{"version": "0.2.2", "doi": "10.5281/zenodo.22674547", "forms": [{"id": "form-012", "version": 1, "legacy_line": 12, "name": "Corpus volume", "name_fr": "Volume du corpus", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "CORPUS_PARAMETER", "class": "REFUSABLE", "layer": "⑨ structure", "injection": "PARAMETER", "legacy_severity": "G4", "legacy_mechanism": "", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles XML JATS d'(internal corpus of 272 JATS articles). Le fonds compte 8 strates : bio 34 · cas 34 · essai 34 · meta 34 · moderne 17 · nonEN 51 · phys 34 · retract 34. Le prélèvement des 100 premiers par ordre de nom donne {bio 34, cas 34, essai 32} et PERD 5 strates sur 8 (meta, moderne, nonEN, phys, retract) ; celui des 200 premiers en perd encore 2 (phys, retract). Les 272 identifiants sont distincts (272/272), donc l'emboîtement à l'octet est vrai — c'est la représentativité qui est fausse.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 JATS XML articles of the internal corpus. The corpus has 8 strata: bio 34 · case 34 · trial 34 · meta 34 · modern 17 · nonEN 51 · phys 34 · retract 34. Taking the first 100 by file name gives {bio 34, case 34, trial 32} and LOSES 5 strata out of 8 (meta, modern, nonEN, phys, retract); the first 200 still lose 2 (phys, retract). The 272 identifiers are distinct (272/272), so byte-level nesting holds — it is representativeness that fails.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": [], "history": [{"date": "2026-08-09", "event": "catalogue line 12 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=CORPUS_PARAMETER injection=PARAMETER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 5 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-013", "version": 1, "legacy_line": 13, "name": "Distribution of relation types", "name_fr": "Distribution des types de relation", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "CORPUS_PARAMETER", "class": "IRREDUCIBLE", "layer": "⑦ schema", "injection": "PARAMETER", "legacy_severity": "G2", "legacy_mechanism": "M12", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ sur le palier P6, 1045 relations, relevé au catalogue le 08/08/2026 : 5 types portent 97,3 % des relations, « est fournisseur de » à lui seul 552 soit 52,8 %, et 21 types sur 26 n'ont qu'1 ou 2 occurrences — donc le F1 par type y vaut 0 ou 1 par tirage.", "lang": "fr", "excerpt_en": "MEASURED on tier P6, 1,045 relations, recorded in the catalogue on 2026-08-08: 5 types carry 97.3 % of the relations, \"is a supplier of\" alone 552, i.e. 52.8 %, and 21 types out of 26 have only 1 or 2 occurrences — so the per-type F1 is 0 or 1 by chance.", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": [], "history": [{"date": "2026-08-09", "event": "catalogue line 13 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=CORPUS_PARAMETER injection=PARAMETER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-014", "version": 1, "legacy_line": 14, "name": "Entity typing", "name_fr": "Typage des entités", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_LABEL", "class": "IRREDUCIBLE", "layer": "⑦ schema", "injection": "DESIGN", "legacy_severity": "G2", "legacy_mechanism": "M12", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 XML de (internal corpus of 272 JATS articles) : 63/272 (23 %) font voisiner une organisation balisée et une étude/cohorte NOMMÉE (« the Fremantle Diabetes Study », « the Brazilian Longitudinal Study ») — deux types confusables dans la même phrase. Seuls 9/272 (3 %) portent l'apposition qui tranche (« , a multicenter … »). Rappel du palier P6 (annexe 14) : 920/925 entités sont Organization ou Person — le typage n'est aujourd'hui AUCUNE difficulté.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 XML files of the internal corpus: 63/272 (23 %) place a tagged organisation next to a NAMED study/cohort (\"the Fremantle Diabetes Study\", \"the Brazilian Longitudinal Study\") — two confusable types in the same sentence. Only 9/272 (3 %) carry the apposition that settles it (\", a multicenter …\"). Reminder from tier P6 (annex 14): 920/925 entities are Organization or Person — typing is currently NO difficulty at all.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "partial", "note": "an extra label can be deleted, a missing one cannot be added"}, "judgeable_by": ["construction", "curated"], "history": [{"date": "2026-08-09", "event": "catalogue line 14 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_LABEL injection=DESIGN (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-015", "version": 1, "legacy_line": 15, "name": "Schema constraints", "name_fr": "Contraintes de schéma", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_LABEL", "class": "IRREDUCIBLE", "layer": "⑦ schema", "injection": "CONSTRAINT", "legacy_severity": "G2", "legacy_mechanism": "M12", "status": "migrated_unreviewed", "seen": [], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "partial", "note": "an extra label can be deleted, a missing one cannot be added"}, "judgeable_by": ["construction", "curated"], "history": [{"date": "2026-08-09", "event": "catalogue line 15 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_LABEL injection=CONSTRAINT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-016", "version": 1, "legacy_line": 16, "name": "Fact redundancy across documents", "name_fr": "Redondance du fait", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "CORPUS_PARAMETER", "class": "REFUSABLE", "layer": "⑩ retrieval", "injection": "DUPLICATE", "legacy_severity": "G4", "legacy_mechanism": "M9", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 (internal corpus of 272 JATS articles) : 1 753 énoncés chiffrés distincts extraits des résumés, dont 1 749 à UN SEUL témoin (99,8 %). Les 4 seuls énoncés à deux témoins ne sont pas des faits : licence Creative Commons et avis de rétractation. En revanche 60 articles sur 272 (22 %) répètent le MÊME énoncé dans le résumé et dans le corps — deux ancrages pour un seul témoin. Contrôle positif de la sonde : le même article sous deux noms → 2/2 énoncés vus à deux témoins.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026 (internal corpus): 1,753 distinct numerical statements extracted from the abstracts, of which 1,749 have a SINGLE witness (99.8 %). The only 4 statements with two witnesses are not facts: a Creative Commons licence and a retraction notice. However 60 articles out of 272 (22 %) repeat the SAME statement in the abstract and in the body — two anchors for a single witness. Positive control of the probe: the same article under two names → 2/2 statements seen with two witnesses.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": [], "history": [{"date": "2026-08-09", "event": "catalogue line 16 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=CORPUS_PARAMETER injection=DUPLICATE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 3 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-017", "version": 1, "legacy_line": 17, "name": "Diversity of document templates", "name_fr": "Diversité des gabarits", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "② reading", "injection": "FORMAT", "legacy_severity": "G3", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles XML JATS d'(internal corpus of 272 JATS articles) (Europe PMC accès libre 2023-2026) : 272/272 déclarent un article-type, et on compte 15 gabarits DISTINCTS. Formes réelles au sommet : research-article 145, case-report 35, retraction 33, review-article 23, editorial 14, letter 9, brief-report 4, book-review 4, systematic-review 3 — et article-type=\"other\", gabarit NON DÉCLARÉ, dans 2 fichiers (bio-PMC12674831.xml, nonEN-PMC13228121.xml). 24 gabarits « translation » supplémentaires n'apparaissent que sur des : ils ne sont PAS le gabarit du fichier.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 JATS XML articles of the internal corpus (Europe PMC open access 2023-2026): 272/272 declare an article-type, and 15 DISTINCT templates are counted. Real forms at the top: research-article 145, case-report 35, retraction 33, review-article 23, editorial 14, letter 9, brief-report 4, book-review 4, systematic-review 3 — and article-type=\"other\", template NOT DECLARED, in 2 files (bio-PMC12674831.xml, nonEN-PMC13228121.xml). 24 additional \"translation\" templates appear only on : they are NOT the template of the file.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 17 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 5 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-018", "version": 1, "legacy_line": 18, "name": "Variable length and chunking", "name_fr": "Longueur variable et découpe", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "② reading", "injection": "FORMAT", "legacy_severity": "G3", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 : présent dans 243 (89 %). Concentration : nonEN 51, bio 34, cas 34, essai 34, meta 34, phys 34, moderne 17, retract 5. Formes réelles : bio-PMC12583535 = 48 845 car. → 25 morceaux | 29 documents tiennent en un seul morceau", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026: present in 243 (89 %). Concentration: nonEN 51, bio 34, case 34, trial 34, meta 34, phys 34, modern 17, retract 5. Real forms: bio-PMC12583535 = 48,845 characters → 25 chunks | 29 documents fit in a single chunk.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 18 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 2 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-019", "version": 1, "legacy_line": 19, "name": "Salience and distractors", "name_fr": "Saillance et distracteurs", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "IRREDUCIBLE", "layer": "④ extraction", "injection": "INSERT", "legacy_severity": "G2", "legacy_mechanism": "M12", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : sur les 75 articles porteurs de mots-clés, 163 des 523 mots-clés (31 %) n'apparaissent au plus qu'UNE fois dans le corps, tandis que le mot-clé le plus cité de chaque article y revient 24 fois (médiane). Un hub, une longue traîne de mentions uniques : c'est exactement la forme que ce module fabrique.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): of the 75 articles carrying keywords, 163 of the 523 keywords (31 %) appear at most ONCE in the body, while the most cited keyword of each article recurs 24 times (median). One hub, a long tail of single mentions: exactly the shape this module fabricates.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 19 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=INSERT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-020", "version": 1, "legacy_line": 20, "name": "Layout furniture (headers, footers)", "name_fr": "Mise en forme et habillage", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "DEFEATED_IF_XML", "layer": "② reading", "injection": "FORMAT", "legacy_severity": "G3", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles XML JATS Europe PMC (2023-2026) : présent dans 181/272 (66 %). Formes réelles : bio-PMC12674831 porte 154 et 75 ; premier = « Position Statement on Cardiometabolic Health… » — un intitulé et sa valeur alignés, sans verbe. bio-PMC12583535 : 51 , premier = « Database Search ».", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 Europe PMC JATS XML articles (2023-2026): present in 181/272 (66 %). Real forms: bio-PMC12674831 carries 154 and 75 ; first = \"Position Statement on Cardiometabolic Health…\" — a heading and its value aligned, without a verb. bio-PMC12583535: 51 , first = \"Database Search\".", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 20 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-021", "version": 1, "legacy_line": 21, "name": "Multiple spellings of one entity", "name_fr": "Graphies multiples d'une entité", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "REFUSABLE", "layer": "⑥ resolution", "injection": "RENAME", "legacy_severity": "G1", "legacy_mechanism": "M1", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles XML JATS Europe PMC accès libre (internal corpus of 272 JATS articles) : présent dans 209 (77 %). Concentration : essai 34, phys 34, bio 33, meta 33, nonEN 32, cas 27, moderne 14, retract 2. Formes réelles : « C-reactive protein (CRP) » (bio-PMC12674831) · « Diabetes mellitus (DM) » (bio-PMC12590050) · « Risk of Bias (RoB) » (bio-PMC12583535).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC JATS XML articles (internal corpus): present in 209 (77 %). Concentration: trial 34, phys 34, bio 33, meta 33, nonEN 32, case 27, modern 14, retract 2. Real forms: \"C-reactive protein (CRP)\" (bio-PMC12674831) · \"Diabetes mellitus (DM)\" (bio-PMC12590050) · \"Risk of Bias (RoB)\" (bio-PMC12583535).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}, {"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 : 209/272 articles (77 %) glosent puis abrègent — « C-reactive protein (CRP) » puis « CRP » seul.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08: 209/272 articles (77 %) gloss then abbreviate — \"C-reactive protein (CRP)\" then \"CRP\" alone.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 21 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPLIT injection=RENAME (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 5 cases and 2 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-022", "version": 1, "legacy_line": 22, "name": "Pronominal references", "name_fr": "Reprises pronominales", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MERGE", "class": "IRREDUCIBLE", "layer": "③ utterance", "injection": "RENAME", "legacy_severity": "G1", "legacy_mechanism": "M2", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles XML JATS de l'atelier : 212/272 (78 %) ouvrent au moins une phrase par une reprise nue (It/They/These/This), 3 260 occurrences au total ; 49/272 (18 %) portent une reprise lexicale explicite — relevé « In the latter study, the risk of developing T2DM was significantly lower » (bio-PMC12915353).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 JATS XML articles of the workshop: 212/272 (78 %) open at least one sentence with a bare anaphor (It/They/These/This), 3,260 occurrences in total; 49/272 (18 %) carry an explicit lexical anaphor — recorded: \"In the latter study, the risk of developing T2DM was significantly lower\" (bio-PMC12915353).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 22 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MERGE injection=RENAME (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-023", "version": 1, "legacy_line": 23, "name": "Typos and OCR confusions", "name_fr": "Bruit de frappe et confusions d'OCR", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "SILENT_FALSE", "layer": "② reading", "injection": "NOISE", "legacy_severity": "G3", "legacy_mechanism": "M10", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 : 80 % des documents anciens observés sont des scans OCRisés, où l/1, O/0 et S/5 se confondent.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08: 80 % of the old documents observed are OCR'd scans, where l/1, O/0 and S/5 are confused.", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 23 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPLIT injection=NOISE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-024", "version": 1, "legacy_line": 24, "name": "Acronyms and abbreviations", "name_fr": "Sigles et abréviations", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MERGE", "class": "SILENT_FALSE", "layer": "⑥ resolution", "injection": "RENAME", "legacy_severity": "G1", "legacy_mechanism": "M1", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : 224/272 articles (82 %) portent au moins une glose « Nom complet (SIGLE) », 2 104 gloses au total, soit 9,4 par article concerné.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): 224/272 articles (82 %) carry at least one gloss \"Full name (ACRONYM)\", 2,104 glosses in total, i.e. 9.4 per affected article.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 24 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MERGE injection=RENAME (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-025", "version": 1, "legacy_line": 25, "name": "Registry identifiers", "name_fr": "Identifiants de registre", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MERGE", "class": "DEFEATED", "layer": "⑥ resolution", "injection": "ALTER", "legacy_severity": "G3", "legacy_mechanism": "M2", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 (internal corpus of 272 JATS articles) : présent dans 31 articles (11 %). Concentration : phys 10/34, meta 9/34, bio 6/34, moderne 4/17, essai 2/34. Registres réellement rencontrés : ROR ×60, GRID ×52, ISNI ×49, Ringgold ×3, plus 22 identifiants de FINANCEUR (doi/FundRef/CrossRef), qui ne désignent PAS l'organisation. Formes réelles : ROR:https://ror.org/02z1vqm45 | GRID:grid.411472.5 | ISNI:0000 0004 1764 1621", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026 (internal corpus): present in 31 articles (11 %). Concentration: phys 10/34, meta 9/34, bio 6/34, modern 4/17, trial 2/34. Registries actually met: ROR ×60, GRID ×52, ISNI ×49, Ringgold ×3, plus 22 FUNDER identifiers (doi/FundRef/CrossRef), which do NOT designate the organisation. Real forms: ROR:https://ror.org/02z1vqm45 | GRID:grid.411472.5 | ISNI:0000 0004 1764 1621", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 25 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MERGE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 5 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-026", "version": 1, "legacy_line": 26, "name": "Strict homonyms", "name_fr": "Homonymes stricts", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MERGE", "class": "IRREDUCIBLE", "layer": "⑥ resolution", "injection": "RENAME", "legacy_severity": "G1", "legacy_mechanism": "M2", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles XML JATS de (internal corpus of 272 JATS articles) : 1 215 patronymes sur 9 160 (13,3 %) sont portés par au moins deux prénoms différents — « Wang » par 95, « Li » par 92, « Chen » par 88 ; et 461 noms complets identiques (nom + prénoms) apparaissent dans au moins deux articles aux affiliations DISJOINTES (Zhang W, Li S, Singh S…). Faute vérifiée le 08/08/2026 par ailleurs : le contrôle d'atteignabilité ne regarde pas les assertions `distinct` — un corpus en portant 27 affichait « 0 piège » et passait au vert.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 JATS XML articles of the internal corpus: 1,215 surnames out of 9,160 (13.3 %) are borne by at least two different given names — \"Wang\" by 95, \"Li\" by 92, \"Chen\" by 88; and 461 identical full names (surname + given names) appear in at least two articles with DISJOINT affiliations (Zhang W, Li S, Singh S…). Fault verified the same day elsewhere: the reachability check does not look at `distinct` assertions — a corpus carrying 27 of them displayed \"0 traps\" and passed green.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 26 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MERGE injection=RENAME (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-027", "version": 1, "legacy_line": 27, "name": "Name-to-address bridge", "name_fr": "Pont nom ↔ adresse", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "IRREDUCIBLE", "layer": "⑥ resolution", "injection": "RENAME", "legacy_severity": "G1", "legacy_mechanism": "M2", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 XML : 54/272 (20 %) portent, DANS LA MÊME balise , le nom de l'établissement ET son adresse postale (code postal ou numéro de rue) — « Department of Urology, Peking University … 100034 Beijing ». 5/272 (2 %) donnent un numéro de rue explicite (« 140 Huguenot Street », « 654 Cherry Street »), et 60/272 (22 %) un couple code postal + ville. La matière du pont existe ; ce qui manque au corpus fabriqué, c'est sa RÉPARTITION sur plusieurs documents.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 XML files: 54/272 (20 %) carry, IN THE SAME tag, the name of the institution AND its postal address (postcode or street number) — \"Department of Urology, Peking University … 100034 Beijing\". 5/272 (2 %) give an explicit street number (\"140 Huguenot Street\", \"654 Cherry Street\"), and 60/272 (22 %) a postcode + city pair. The material for the bridge exists; what the fabricated corpus lacks is its DISTRIBUTION across several documents.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 27 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPLIT injection=RENAME (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-028", "version": 1, "legacy_line": 28, "name": "One person in two roles", "name_fr": "Une personne dans deux rôles", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MERGE", "class": "IRREDUCIBLE", "layer": "⑥ resolution", "injection": "DESIGN", "legacy_severity": "G1", "legacy_mechanism": "M2", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles de (internal corpus of 272 JATS articles) : dans 39/272 (14 %) un AUTEUR de l'article est aussi auteur d'une référence citée — une seule personne, deux rôles (signataire ici, source citée là) ; ex. bio-PMC12915353 (Azizi F., Golzarand M., Mirmiran P.). Et 51/272 (19 %) portent un contributeur déclaré sous au moins deux rôles distincts.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 articles of the internal corpus: in 39/272 (14 %) an AUTHOR of the article is also an author of a cited reference — one person, two roles (signatory here, cited source there); e.g. bio-PMC12915353 (Azizi F., Golzarand M., Mirmiran P.). And 51/272 (19 %) carry a contributor declared under at least two distinct roles.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 28 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MERGE injection=DESIGN (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-029", "version": 1, "legacy_line": 29, "name": "Entities that change identity", "name_fr": "Entités qui changent d'identité", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "IRREDUCIBLE", "layer": "⑥ resolution", "injection": "RENAME", "legacy_severity": "G1", "legacy_mechanism": "M5", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : seulement 6/272 articles (2 %) portent une formule de changement d'identité dans le corps (formerly / renamed / previously known as / merged with / acquired by). ⇒ Le corpus SCIENTIFIQUE ne l'enseigne presque pas ; c'est le fonds d'entreprise qui la porte (annexe 29). Le module l'injecte donc à un taux choisi, pas observé.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): only 6/272 articles (2 %) carry an identity-change phrase in the body (formerly / renamed / previously known as / merged with / acquired by). ⇒ The SCIENTIFIC corpus hardly teaches it; corporate archives carry it (annex 29). The module therefore injects it at a chosen rate, not an observed one.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 29 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPLIT injection=RENAME (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-030", "version": 1, "legacy_line": 30, "name": "Facts spread across documents", "name_fr": "Faits répartis entre documents", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "IRREDUCIBLE", "layer": "⑩ retrieval", "injection": "MOVE", "legacy_severity": "G3", "legacy_mechanism": "M11", "status": "migrated_unreviewed", "seen": [], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 30 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=MOVE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-031", "version": 1, "legacy_line": 31, "name": "Degree of assertion of a link", "name_fr": "Degré d'affirmation du lien", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "IRREDUCIBLE", "layer": "④ extraction", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M5", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : 233/272 articles (86 %) portent au moins un marqueur de non-affirmation (may · might · could · suggests · appears to · likely · has been proposed) ; 4 518 occurrences au total, soit 19,4 par article concerné.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): 233/272 articles (86 %) carry at least one non-assertion marker (may · might · could · suggests · appears to · likely · has been proposed); 4,518 occurrences in total, i.e. 19.4 per affected article.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 31 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-032", "version": 2, "legacy_line": 32, "name": "Link status: never, or expired", "name_fr": "Statut du lien : jamais ou révolu", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "ANACHRONISM", "class": "IRREDUCIBLE", "layer": "⑧ coherence", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M5", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 : 39/272 articles (14 %) portent une version successive ou une rétractation — un lien affirmé puis annulé.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08: 39/272 articles (14 %) carry a successive version or a retraction — a link asserted then cancelled.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "partial", "note": "deleting the anachronistic edge removes a fact that was true within its validity interval; the true repair bounds the validity, it does not delete (reading of Coralie Bagnol-Lebon, 2026-09-09)"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 32 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=ANACHRONISM injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "repair.reachable_by_deletion changed from yes to partial: a deletion loses the fact's past validity; only a spurious edge is fully repaired by deletion", "by": "Coralie Bagnol-Lebon (second reader), applied by Fault Atlas"}]}, {"id": "form-033", "version": 1, "legacy_line": 33, "name": "Confusable relation types", "name_fr": "Confusabilité des types de relation", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_LABEL", "class": "IRREDUCIBLE", "layer": "⑦ schema", "injection": "RENAME", "legacy_severity": "G2", "legacy_mechanism": "M12", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ, relevé au catalogue le 08/08/2026 : F1 82,8 % quand on ignore le type, F1 30,1 % quand on le compte — on relie les bonnes entités et on se trompe de lien trois fois sur quatre. Et vérifié dans le code le même jour : le vocabulaire est annoncé fermé à 26 types, la table des inverses en accepte 31.", "lang": "fr", "excerpt_en": "MEASURED, recorded in the catalogue on 2026-08-08: F1 82.8 % when the type is ignored, F1 30.1 % when it is counted — the right entities are linked and the link is wrong three times out of four. And verified in the code the same day: the vocabulary is announced closed at 26 types, the inverse table accepts 31.", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "partial", "note": "an extra label can be deleted, a missing one cannot be added"}, "judgeable_by": ["construction", "curated"], "history": [{"date": "2026-08-09", "event": "catalogue line 33 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_LABEL injection=RENAME (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-034", "version": 1, "legacy_line": 34, "name": "Hard negatives for typing", "name_fr": "Négatifs durs de typage", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_LABEL", "class": "IRREDUCIBLE", "layer": "④ extraction", "injection": "INSERT", "legacy_severity": "G2", "legacy_mechanism": "M12", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 XML : 19/272 (7 %) portent DEUX formulations de types voisins pour le même fait de financement — « funded by » ET « supported by » dans le même article ; 100/272 (37 %) emploient au moins un terme de relation contractuelle confusable (collaboration, contrat, prestataire, fournisseur). Le corpus réel offre donc le vocabulaire du piège ; aucun de nos corpus fabriqués ne le pose.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 XML files: 19/272 (7 %) carry TWO neighbouring-type formulations for the same funding fact — \"funded by\" AND \"supported by\" in the same article; 100/272 (37 %) use at least one confusable contractual relation term (collaboration, contract, provider, supplier). The real corpus thus offers the vocabulary of the trap; none of our fabricated corpora sets it.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "partial", "note": "an extra label can be deleted, a missing one cannot be added"}, "judgeable_by": ["construction", "curated"], "history": [{"date": "2026-08-09", "event": "catalogue line 34 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_LABEL injection=INSERT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-035", "version": 1, "legacy_line": 35, "name": "Undated contradiction", "name_fr": "Contradiction sans date", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "IRREDUCIBLE", "layer": "⑧ coherence", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M5", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles de (internal corpus of 272 JATS articles) : 15 049 des 16 411 paragraphes de corps (92 %) ne contiennent AUCUNE année. L'affirmation non datée est donc la norme, pas l'exception — et deux affirmations non datées ne peuvent pas être départagées par la date. (Sonde : comptage des

du portant un jeton d'année 19xx/20xx.)", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 articles of the internal corpus: 15,049 of the 16,411 body paragraphs (92 %) contain NO year. The undated statement is therefore the norm, not the exception — and two undated statements cannot be separated by date. (Probe: count of

in carrying a 19xx/20xx year token.)", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 35 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-036", "version": 1, "legacy_line": 36, "name": "Relative dates", "name_fr": "Dates relatives", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "REFUSABLE", "layer": "⑧ coherence", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M5", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles XML JATS d'(internal corpus of 272 JATS articles) : présent dans 64 (23 %), corps de l'article seul (front/back retirés). Concentration : cas 20, essai 16, bio 11, moderne 5. Formes réelles les plus fréquentes : « the past decade » ×5, « the last decade » ×4, « the past year » ×4, « the previous year » ×3, « last week » ×3, « 1 year after » ×3, « two years after » ×2, « 3 years after » ×2, « the next decade » ×2. PASSAGE RÉEL du filtre sur ces 272 documents (08/08/2026) : il résout une expression au moins dans 9 documents (« the previous year » ×3, « ten years ago », « 4 years ago », « 1 year ago », « three years ago », « last year ») et s'ABSTIENT sur les 54 autres — portées, ancres inconnues, granularité inférieure à l'année. C'est ce passage qui a fait sortir « 000 years ago », amputé de « 10,000 years ago » : sans garde-fou le filtre rendait 2024 − 0 = 2024, une date FABRIQUÉE. D'où le refus au-delà de 150 ans.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 JATS XML articles of the internal corpus: present in 64 (23 %), article body only (front/back removed). Concentration: case 20, trial 16, bio 11, modern 5. Most frequent real forms: \"the past decade\" ×5, \"the last decade\" ×4, \"the past year\" ×4, \"the previous year\" ×3, \"last week\" ×3, \"1 year after\" ×3, \"two years after\" ×2, \"3 years after\" ×2, \"the next decade\" ×2. REAL RUN of the filter on these 272 documents (2026-08-08): it resolves at least one expression in 9 documents (\"the previous year\" ×3, \"ten years ago\", \"4 years ago\", \"1 year ago\", \"three years ago\", \"last year\") and ABSTAINS on the other 54 — spans, unknown anchors, granularity below the year. This run is what surfaced \"000 years ago\", truncated from \"10,000 years ago\": without a guard-rail the filter returned 2024 − 0 = 2024, a FABRICATED date. Hence the refusal beyond 150 years.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 36 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 11 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-037", "version": 1, "legacy_line": 37, "name": "Density: documents of no interest", "name_fr": "Densité : documents sans intérêt", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "CORPUS_PARAMETER", "class": "SILENT_FALSE", "layer": "⑩ retrieval", "injection": "INSERT", "legacy_severity": "G4", "legacy_mechanism": "", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : sur 11 380 paragraphes de plus de 200 caractères, 2 483 (22 %) ne contiennent AUCUN chiffre — du texte de cadre, pas de résultat. Le corpus réel dilue donc déjà à un cinquième, à l'intérieur même des documents.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): of 11,380 paragraphs longer than 200 characters, 2,483 (22 %) contain NO digit — framing text, not results. The real corpus therefore already dilutes to one fifth, inside the documents themselves.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": [], "history": [{"date": "2026-08-09", "event": "catalogue line 37 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=CORPUS_PARAMETER injection=INSERT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-038", "version": 1, "legacy_line": 38, "name": "Exact and near duplicates", "name_fr": "Doublons exacts et quasi-doublons", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "DEFEATED", "layer": "⑨ structure", "injection": "DUPLICATE", "legacy_severity": "G1", "legacy_mechanism": "M9", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "deux copies du même article comptées comme deux témoins indépendants — faux double accord (analyse de gravité, 08/08/2026)", "lang": "fr", "excerpt_en": "two copies of the same article counted as two independent witnesses — a false double agreement (severity analysis, 2026-08-08)", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 38 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=DUPLICATE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 2 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-039", "version": 2, "legacy_line": 39, "name": "Successive versions of a document", "name_fr": "Versions successives d'un document", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "ANACHRONISM", "class": "REFUSABLE", "layer": "⑧ coherence", "injection": "DUPLICATE", "legacy_severity": "G2", "legacy_mechanism": "M5", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 : présent dans 39 (14 %). Concentration : retract 33, meta 3, bio 1, essai 1, phys 1. Formes réelles : related-article-type=retracted-article ×45 | retraction-forward ×14 | final-edited-article ×2 | preprint ×1 | 3 ×2 | Version of Record ×1", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026: present in 39 (14 %). Concentration: retract 33, meta 3, bio 1, trial 1, phys 1. Real forms: related-article-type=retracted-article ×45 | retraction-forward ×14 | final-edited-article ×2 | preprint ×1 | 3 ×2 | Version of Record ×1", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "partial", "note": "deleting the anachronistic edge removes a fact that was true within its validity interval; the true repair bounds the validity, it does not delete (reading of Coralie Bagnol-Lebon, 2026-09-09)"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 39 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=ANACHRONISM injection=DUPLICATE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 5 cases and 2 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "repair.reachable_by_deletion changed from yes to partial: a deletion loses the fact's past validity; only a spurious edge is fully repaired by deletion", "by": "Coralie Bagnol-Lebon (second reader), applied by Fault Atlas"}]}, {"id": "form-040", "version": 1, "legacy_line": 40, "name": "Cited documents absent from the corpus", "name_fr": "Documents cités et absents du fonds", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "REFUSABLE", "layer": "⑩ retrieval", "injection": "DELETE", "legacy_severity": "G3", "legacy_mechanism": "M11", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles XML JATS Europe PMC (2023-2026), 10 621 références dont 4 344 portent un DOI, contre 315 DOI présents dans le fonds : 110/272 articles (40 %) citent au moins un document identifié ABSENT du fonds, et 0/272 citent un document présent. Surtout : 217/272 (80 %) portent au moins une référence SANS aucun identifiant — forme réelle, bio-PMC12583535 : « Hanahan D, Weinberg RA. Hallmarks of cancer: The next generation. Cell 2011 » — irrésolvable, donc abstention obligatoire.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 Europe PMC JATS XML articles (2023-2026), 10,621 references of which 4,344 carry a DOI, against 315 DOIs present in the corpus: 110/272 articles (40 %) cite at least one identified document ABSENT from the corpus, and 0/272 cite a document that is present. Above all: 217/272 (80 %) carry at least one reference WITHOUT any identifier — real form, bio-PMC12583535: \"Hanahan D, Weinberg RA. Hallmarks of cancer: The next generation. Cell 2011\" — unresolvable, hence mandatory abstention.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 40 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=DELETE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-041", "version": 1, "legacy_line": 41, "name": "Truncated or unreadable documents", "name_fr": "Documents tronqués ou illisibles", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "REFUSABLE", "layer": "② reading", "injection": "DELETE", "legacy_severity": "G3", "legacy_mechanism": "M11", "status": "migrated_unreviewed", "seen": [], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 41 — ACCEPTED (measured absent from the modern XML corpus): measured ABSENT from the modern XML corpus", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=DELETE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}], "observed_absent_in": ["Europe PMC JATS 2023-2026 (272 articles, measured 2026-08-08)"]}, {"id": "form-042", "version": 1, "legacy_line": 42, "name": "Names damaged by scanning", "name_fr": "Noms abîmés par le scan", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "REFUSABLE", "layer": "⑥ resolution", "injection": "NOISE", "legacy_severity": "G1", "legacy_mechanism": "M1", "status": "migrated_unreviewed", "seen": [], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 42 — ACCEPTED (measured absent from the modern XML corpus): measured ABSENT from the modern XML corpus", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPLIT injection=NOISE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}], "observed_absent_in": ["Europe PMC JATS 2023-2026 (272 articles, measured 2026-08-08)"]}, {"id": "form-043", "version": 1, "legacy_line": 43, "name": "Tables with multi-level headers", "name_fr": "Tableaux à en-têtes multiples", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "DEFEATED_IF_XML", "layer": "② reading", "injection": "FORMAT", "legacy_severity": "G2", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles XML JATS d'(internal corpus of 272 JATS articles) : présent dans 73 (26 %). Concentration : essai 21, nonEN 12, meta 11, bio 10, phys 10. Formes réelles : nonEN-PMC13193189.xml porte 36 à colspan/rowspan ≥2 ; essai-PMC13205731.xml en porte 27 ; bio-PMC12857832.xml a 4 sur 5 qui comptent plus d'une ligne . PASSAGE RÉEL du filtre (08/08/2026) : sur les 670 du corpus il en classe 151 « en-têtes multiples », 469 « en-tête simple », et s'ABSTIENT sur 50 sans . Au niveau du document : 71/272, contre 73 pour la sonde — l'écart de 2 vient de , que le filtre refuse de prendre pour un en-tête.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 JATS XML articles of the internal corpus: present in 73 (26 %). Concentration: trial 21, nonEN 12, meta 11, bio 10, phys 10. Real forms: nonEN-PMC13193189.xml carries 36 out of 5 with more than one row. REAL RUN of the filter (2026-08-08): of the 670
à colspan situés HORS
with colspan/rowspan ≥2; trial-PMC13205731.xml carries 27; bio-PMC12857832.xml has 4
in the corpus it classifies 151 as \"multi-level headers\", 469 as \"single header\", and ABSTAINS on 50 without . At document level: 71/272, against 73 for the probe — the difference of 2 comes from colspan , which the filter refuses to take as a header.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 43 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 5 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-044", "version": 1, "legacy_line": 44, "name": "Several languages in one corpus", "name_fr": "Plusieurs langues dans un même fonds", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "REFUSABLE", "layer": "② reading", "injection": "TRANSLATE", "legacy_severity": "G1", "legacy_mechanism": "M9", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 : présent dans 27 (10 %), TOUS dans la famille nonEN. Formes réelles : ×26 | xml:lang=\"es\" ×9 | xml:lang=\"pt\" ×3 | ×8. Exemple nonEN-PMC13143285 : titre PT « Conquistar corpos, domar a selva… » + trans-title EN « Conquering bodies, taming the jungle… »", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026: present in 27 (10 %), ALL in the nonEN family. Real forms: ×26 | xml:lang=\"es\" ×9 | xml:lang=\"pt\" ×3 | ×8. Example nonEN-PMC13143285: PT title \"Conquistar corpos, domar a selva…\" + EN trans-title \"Conquering bodies, taming the jungle…\"", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 44 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPLIT injection=TRANSLATE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 5 cases and 2 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-045", "version": 2, "legacy_line": 45, "name": "Genuinely scanned pages", "name_fr": "Vraies pages scannées", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "SILENT_FALSE", "layer": "① pixel", "injection": "IMAGE", "legacy_severity": "G3", "legacy_mechanism": "M7", "status": "migrated_unreviewed", "seen": [{"corpus": "Biomedical PDFs 1957-1987 (migraine corpus, 45 PDFs), first observation wave", "document": "Malcolm 1973, The avian egg and the retina (published conference, 2.9 MB, scanned PDF); régime measured on the 45 PDFs", "date": "2026-08-08", "observer": "house observer (LLM agent), first wave of August 2026, biomedical PDFs 1957-1987 (migraine corpus)", "excerpt": "Malcolm 1973, *The avian egg and the retina* (2,9 Mo) — conférence publiée, PDF SCANNÉ. Mesuré le 08/08 sur les 45 PDF du corpus migraine : natif numérique 8/45 (17 %) · scan + couche OCR 36/45 (80 %) · image pure 1/45 (2 %). Vu à l'œil : « IN 1948, WHIII I was working » pour « WHILE I ».", "excerpt_en": "Malcolm 1973, *The avian egg and the retina* (2.9 MB) — published conference, SCANNED PDF. Measured on 2026-08-08 on the 45 PDFs of the migraine corpus: native digital 8/45 (17 %) · scan + OCR layer 36/45 (80 %) · pure image 1/45 (2 %). Seen by eye: \"IN 1948, WHIII I was working\" for \"WHILE I\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 45 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=IMAGE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "observation transcribed from the August annex and the wave-1 journal, after the second reader noted the record had no excerpt", "by": "Fault Atlas"}]}, {"id": "form-046", "version": 1, "legacy_line": 46, "name": "Incremental ingestion", "name_fr": "Ingestion incrémentale", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "CORPUS_PARAMETER", "class": "DEFEATED", "layer": "⑨ structure", "injection": "PARAMETER", "legacy_severity": "G4", "legacy_mechanism": "", "status": "migrated_unreviewed", "seen": [], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": [], "history": [{"date": "2026-08-09", "event": "catalogue line 46 — ACCEPTED (measured absent from the modern XML corpus): measured ABSENT from the modern XML corpus", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=CORPUS_PARAMETER injection=PARAMETER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}], "observed_absent_in": ["Europe PMC JATS 2023-2026 (272 articles, measured 2026-08-08)"]}, {"id": "form-062", "version": 1, "legacy_line": 62, "name": "Document reduced to its title", "name_fr": "Document réduit à son titre", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "IRREDUCIBLE", "layer": "② reading", "injection": "DELETE", "legacy_severity": "G3", "legacy_mechanism": "M11", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "COMPTÉ le 08/08/2026 sur un corpus PubMed réel (annexe 62) : 731 articles sur 3 642 — 20 % du corpus arginine — n'ont aucun corps de texte ; 321 sur 866 — 37 % — sur un second corpus du même dossier. Conséquence mesurée le 31/07 : le juge exige une phrase-preuve, qui sur ces documents ne peut exister.", "lang": "fr", "excerpt_en": "COUNTED on 2026-08-08 on a real PubMed corpus (annex 62): 731 articles out of 3,642 — 20 % of the arginine corpus — have no body text at all; 321 out of 866 — 37 % — on a second corpus of the same folder. Consequence measured on 2026-07-31: the judge requires an evidence sentence, which on these documents cannot exist.", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 62 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=DELETE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-063", "version": 1, "legacy_line": 63, "name": "The title carries the fact", "name_fr": "Le titre porte le fait", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "IRREDUCIBLE", "layer": "② reading", "injection": "MOVE", "legacy_severity": "G3", "legacy_mechanism": "M11", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles XML JATS de (internal corpus of 272 JATS articles) : 263/272 titres (96,7 %) ne portent aucun verbe conjugué — ils sont nominaux ; 197/272 (72 %) portent au moins deux tokens capitalisés ou sigles, donc au moins deux entités candidates ; longueur médiane 117 caractères. Et 59/272 (22 %) n'ont aucun résumé : sur ceux-là le titre est la seule source de la notice.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 JATS XML articles of the internal corpus: 263/272 titles (96.7 %) carry no conjugated verb — they are nominal; 197/272 (72 %) carry at least two capitalised tokens or acronyms, hence at least two candidate entities; median length 117 characters. And 59/272 (22 %) have no abstract: for those the title is the only source of the record.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 63 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=MOVE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-064", "version": 1, "legacy_line": 64, "name": "Period typographic conventions", "name_fr": "Conventions typographiques d'époque", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "IRREDUCIBLE", "layer": "② reading", "injection": "FORMAT", "legacy_severity": "G3", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 XML (PMC récents) : ZÉRO titre en capitales intégrales, mais 70/272 (26 %) portent au moins un intertitre entièrement capitalisé (« ABSTRACT », « KEY WORDS: ») et 30/272 (11 %) un mot de 4 lettres ou plus tout en capitales dans le titre. Le zéro est lui-même la mesure : la convention est DATÉE. L'observation d'origine, 08/08/2026 sur le corpus PubMed ancien (annexe 64), comptait 184 titres en capitales intégrales — « STUDIES ON THE NATURE OF RESISTANCE OF GRAM-NEGATIVE BACILLI TO PENICILLIN ».", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 XML files (recent PMC): ZERO titles in full capitals, but 70/272 (26 %) carry at least one fully capitalised subheading (\"ABSTRACT\", \"KEY WORDS:\") and 30/272 (11 %) a word of 4 letters or more in full capitals in the title. The zero is itself the measurement: the convention is DATED. The original observation, 2026-08-08 on the old PubMed corpus (annex 64), counted 184 titles in full capitals — \"STUDIES ON THE NATURE OF RESISTANCE OF GRAM-NEGATIVE BACILLI TO PENICILLIN\".", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 64 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPLIT injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-065", "version": 1, "legacy_line": 65, "name": "Temporal span of the corpus", "name_fr": "Étendue temporelle du fonds", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "CORPUS_PARAMETER", "class": "IRREDUCIBLE", "layer": "⑨ structure", "injection": "PARAMETER", "legacy_severity": "G4", "legacy_mechanism": "", "status": "migrated_unreviewed", "seen": [], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": [], "history": [{"date": "2026-08-09", "event": "catalogue line 65 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=CORPUS_PARAMETER injection=PARAMETER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-066", "version": 2, "legacy_line": 66, "name": "Multi-column layout", "name_fr": "Mise en page multi-colonnes", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "② reading", "injection": "FORMAT", "legacy_severity": "G2", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "Biomedical PDFs 1957-1987 (migraine corpus), first observation wave", "document": "Hachinski et al., Cerebral Hemodynamics in Migraine, Canadian Journal of Neurological Sciences, 1977 (PDF 3.3 MB)", "date": "2026-08-08", "observer": "house observer (LLM agent), first wave of August 2026, biomedical PDFs 1957-1987 (migraine corpus)", "excerpt": "Observé le 08/08 sur un article COMPLET — *Cerebral Hemodynamics in Migraine*, Hachinski et al., Canadian Journal of Neurological Sciences, 1977 (PDF 3,3 Mo). Le texte n'est pas une colonne : il en compte deux ou trois. La première page en a trois — résumé anglais, résumé français, corps de l'article. Une lecture en travers des colonnes produit un texte grammaticalement correct et sémantiquement faux.", "excerpt_en": "Observed on 2026-08-08 on a COMPLETE article — *Cerebral Hemodynamics in Migraine*, Hachinski et al., Canadian Journal of Neurological Sciences, 1977 (PDF 3.3 MB). The text is not one column: it has two or three. The first page has three — English abstract, French abstract, body. Reading across the columns yields text that is grammatically correct and semantically false.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 66 — ACCEPTED (measured absent from the modern XML corpus): measured ABSENT from the modern XML corpus", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "observation transcribed from the August annex and the wave-1 journal, after the second reader noted the record had no excerpt", "by": "Fault Atlas"}], "observed_absent_in": ["Europe PMC JATS 2023-2026 (272 articles, measured 2026-08-08)"]}, {"id": "form-067", "version": 1, "legacy_line": 67, "name": "Bilingual document", "name_fr": "Document bilingue", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "REFUSABLE", "layer": "② reading", "injection": "TRANSLATE", "legacy_severity": "G1", "legacy_mechanism": "M9", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 (internal corpus of 272 JATS articles) : présent dans 27 articles (10 %), TOUS dans le lot non anglophone — 27 sur 51, soit 53 % de ce lot, et 0 % des sept autres. Formes réelles : nonEN-PMC13185728 porte pt→en (title), pt→es (title), pt→en (abstract) — un article portugais dont le titre est répété en anglais ET en espagnol, et le résumé en anglais.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026 (internal corpus): present in 27 articles (10 %), ALL in the non-English batch — 27 out of 51, i.e. 53 % of that batch, and 0 % of the seven others. Real forms: nonEN-PMC13185728 carries pt→en (title), pt→es (title), pt→en (abstract) — a Portuguese article whose title is repeated in English AND Spanish, and the abstract in English.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 67 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPLIT injection=TRANSLATE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 2 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-068", "version": 1, "legacy_line": 68, "name": "The fact is in the figure", "name_fr": "Le fait est dans la figure", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "IRREDUCIBLE", "layer": "① pixel + ⑤ anchoring", "injection": "IMAGE", "legacy_severity": "G3", "legacy_mechanism": "M7", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : 215/272 articles (79 %) portent au moins une figure — 1 059 figures au total — et 138/272 (51 %) contiennent au moins un renvoi de la forme « (Fig. N) » dans le corps ; 1 791 renvois pour 1 059 figures, soit 1,7 renvoi par figure. Le texte pointe l'image bien plus souvent qu'il ne la redit.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): 215/272 articles (79 %) carry at least one figure — 1,059 figures in total — and 138/272 (51 %) contain at least one reference of the form \"(Fig. N)\" in the body; 1,791 references for 1,059 figures, i.e. 1.7 references per figure. The text points to the image far more often than it restates it.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 68 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=IMAGE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-069", "version": 1, "legacy_line": 69, "name": "Several documents in one file", "name_fr": "Plusieurs documents dans un fichier", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MERGE", "class": "DEFEATED_IF_XML", "layer": "⑨ structure", "injection": "DUPLICATE", "legacy_severity": "G1", "legacy_mechanism": "M4", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles XML JATS d'(internal corpus of 272 JATS articles) : présent dans 30 (11 %). Concentration : nonEN 24, bio 3, meta 2, phys 1. Formes réelles, par article-type de : translation 24, associated-data 9, aggregated-review-documents 6, author-comment 3, editor-report 3. Exemple : bio-PMC13252842.xml porte 5 (aggregated-review-documents, associated-data, author-comment, editor-report) dans un seul fichier.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 JATS XML articles of the internal corpus: present in 30 (11 %). Concentration: nonEN 24, bio 3, meta 2, phys 1. Real forms, by article-type: translation 24, associated-data 9, aggregated-review-documents 6, author-comment 3, editor-report 3. Example: bio-PMC13252842.xml carries 5 (aggregated-review-documents, associated-data, author-comment, editor-report) in a single file.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 69 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MERGE injection=DUPLICATE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 6 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-070", "version": 1, "legacy_line": 70, "name": "Bibliographic references", "name_fr": "Références bibliographiques", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "SILENT_FALSE", "layer": "④ extraction", "injection": "MOVE", "legacy_severity": "G2", "legacy_mechanism": "M4", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles XML JATS : 251/272 (92 %) portent un bloc de références final — 10 912 références au total, médiane 36 par article, maximum 450 — et 247/272 (91 %) portent des appels de citation dans le corps (17 210 appels).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 JATS XML articles: 251/272 (92 %) carry a final reference block — 10,912 references in total, median 36 per article, maximum 450 — and 247/272 (91 %) carry citation calls in the body (17,210 calls).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 70 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=MOVE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-071", "version": 1, "legacy_line": 71, "name": "Multiple dates on one document", "name_fr": "Dates multiples d'un même document", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "REFUSABLE", "layer": "⑧ coherence", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M5", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 : présent dans 111 (40 %). Concentration : nonEN 28, meta 16, cas 14. Formes réelles : accepted | collection | collection", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026: present in 111 (40 %). Concentration: nonEN 28, meta 16, case 14. Real forms: accepted | collection | collection", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 71 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 3 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-072", "version": 1, "legacy_line": 72, "name": "Redundancy between text and table", "name_fr": "Redondance entre le texte et le tableau", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "CORPUS_PARAMETER", "class": "REFUSABLE", "layer": "⑤ anchoring", "injection": "DUPLICATE", "legacy_severity": "G4", "legacy_mechanism": "M9", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 : présent dans 95 (35 %), 144 abstentions (documents sans tableau chiffré). Concentration : essai 24, phys 18, nonEN 15, meta 14, bio 11, cas 7, moderne 6. Formes réelles vérifiées à la main : bio-PMC12915353 texte « a 10.6-year CVD risk » ↔ ligne de tableau « 2863 (≥ 19 y) | 10.6 » ; essai-PMC13154027 texte « median [IQR] age, 58.5 [52.0-64.0] years » ↔ tableau de base ; essai-PMC13169472 « (p = 0.004) » ↔ tableau. Sans le seuil de 3 chiffres : 107/272, dont des « 0.3 » coïncidents", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026: present in 95 (35 %), 144 abstentions (documents without a numerical table). Concentration: trial 24, phys 18, nonEN 15, meta 14, bio 11, case 7, modern 6. Real forms checked by hand: bio-PMC12915353 text \"a 10.6-year CVD risk\" ↔ table row \"2863 (≥ 19 y) | 10.6\"; trial-PMC13154027 text \"median [IQR] age, 58.5 [52.0-64.0] years\" ↔ baseline table; trial-PMC13169472 \"(p = 0.004)\" ↔ table. Without the 3-digit threshold: 107/272, including coincidental \"0.3\"s.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": [], "history": [{"date": "2026-08-09", "event": "catalogue line 72 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=CORPUS_PARAMETER injection=DUPLICATE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 5 cases and 3 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-073", "version": 1, "legacy_line": 73, "name": "Values with quantified uncertainty", "name_fr": "Valeurs avec incertitude quantifiée", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "IRREDUCIBLE", "layer": "④ extraction + ⑦ schema", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M3", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 : présent dans 69 (25 %). Concentration : essai 20, meta 17, bio 15. Formes réelles : 0.08 ± | 0.094 ± | 0.63 (95% CI", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026: present in 69 (25 %). Concentration: trial 20, meta 17, bio 15. Real forms: 0.08 ± | 0.094 ± | 0.63 (95% CI", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 73 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 3 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-074", "version": 1, "legacy_line": 74, "name": "Content quoted from elsewhere", "name_fr": "Contenu repris d'ailleurs", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "IRREDUCIBLE", "layer": "⑤ anchoring", "injection": "INSERT", "legacy_severity": "G2", "legacy_mechanism": "M4", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : 13/272 articles (5 %) portent un signal explicite de reprise — « reproduced with permission », « adapted from », « reprinted from » ; 22/272 (8 %) avec les formes larges. Observé en clair le 08/08 sous une figure du corpus migraine : « (Reproduced by permission of the British Journal of Clinical Pharmacology.) »", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): 13/272 articles (5 %) carry an explicit reuse signal — \"reproduced with permission\", \"adapted from\", \"reprinted from\"; 22/272 (8 %) with the broad forms. Observed in the open on 2026-08-08 under a figure of the migraine corpus: \"(Reproduced by permission of the British Journal of Clinical Pharmacology.)\"", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 74 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=INSERT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-075", "version": 1, "legacy_line": 75, "name": "Multiple citation conventions", "name_fr": "Conventions de citation multiples", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "REFUSABLE", "layer": "④ extraction", "injection": "FORMAT", "legacy_severity": "G3", "legacy_mechanism": "M4", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles XML JATS Europe PMC (2023-2026) : les deux conventions coexistent bien dans le même fonds — 207/272 en appel numérique (76 %), 13/272 en auteur-année (5 %), 25 sans appel, 0 mélangé dans un même article. La forme DANGEREUSE est mesurée à part : dans 37/272 articles (14 %) un appel numérique colle directement à une lettre — bio-PMC12583535 donne, une fois les balises retirées, « s1 », « e2 », « y4 », « n10 ». Forme auteur-année réelle : nonEN-PMC13228121 — « Bauman, 1999 », « Carvalho, 2022 ».", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 Europe PMC JATS XML articles (2023-2026): the two conventions do coexist in the same corpus — 207/272 with numeric calls (76 %), 13/272 author-year (5 %), 25 without calls, 0 mixed within one article. The DANGEROUS form is measured separately: in 37/272 articles (14 %) a numeric call sticks directly to a letter — bio-PMC12583535 gives, once the tags are removed, \"s1\", \"e2\", \"y4\", \"n10\". Real author-year form: nonEN-PMC13228121 — \"Bauman, 1999\", \"Carvalho, 2022\".", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 75 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-076", "version": 1, "legacy_line": 76, "name": "Anonymised entities", "name_fr": "Entités anonymisées", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MERGE", "class": "DEFEATED", "layer": "⑥ resolution", "injection": "RENAME", "legacy_severity": "G1", "legacy_mechanism": "M2", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "« Patient 1 » dans un cas clinique portugais de 1986 ET dans un article de 2025 — même surface, entités sans aucun rapport (vagues 1 et 3, 08/08/2026)", "lang": "fr", "excerpt_en": "\"Patient 1\" in a Portuguese clinical case of 1986 AND in an article of 2025 — same surface, entirely unrelated entities (waves 1 and 3, 2026-08-08)", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 76 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MERGE injection=RENAME (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 3 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-077", "version": 1, "legacy_line": 77, "name": "Local and ambiguous date formats", "name_fr": "Formats de date locaux et ambigus", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "REFUSABLE", "layer": "⑧ coherence", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M5", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles XML JATS Europe PMC accès libre (internal corpus of 272 JATS articles) : date numérique AMBIGUË présente dans 4 articles (1,5 %) — moderne 2, cas 1, essai 1 ; date numérique NON ambiguë (jour > 12) dans 3 autres. Formes réelles : « report dated 05/03/2026 » (cas-PMC13291881) · « date of registration: 11.7.2011 » (essai-PMC13254492) · « (accessed 10/3/2023) » (moderne-PMC13264966) · « 06/06/2022 », « 09/07/2019 », « 07/06/2018 » en colonne de tableau (moderne-PMC13274555).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC JATS XML articles (internal corpus): AMBIGUOUS numeric date present in 4 articles (1.5 %) — modern 2, case 1, trial 1; NON-ambiguous numeric date (day > 12) in 3 others. Real forms: \"report dated 05/03/2026\" (case-PMC13291881) · \"date of registration: 11.7.2011\" (trial-PMC13254492) · \"(accessed 10/3/2023)\" (modern-PMC13264966) · \"06/06/2022\", \"09/07/2019\", \"07/06/2018\" in a table column (modern-PMC13274555).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 77 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 6 cases and 3 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-078", "version": 1, "legacy_line": 78, "name": "Structured abstract in numbered points", "name_fr": "Résumé structuré en points numérotés", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "IRREDUCIBLE", "layer": "③ utterance", "injection": "FORMAT", "legacy_severity": "G3", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles XML JATS de (internal corpus of 272 JATS articles) : 0/272 résumés commencent par un numéro de point, et 0/272 portent trois points « 1. … 2. … 3. » ou plus. La forme MODERNE équivalente est le résumé à sections déclarées : 160/272 (59 %) ont un dans leur . ⇒ la mesure DATE la ligne : observée le 08/08/2026 sur Apperley, Humphrey & Levy, Br. J. Pharmac. 1976 (résumé en NEUF points numérotés), absente du fonds moderne.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 JATS XML articles of the internal corpus: 0/272 abstracts start with a point number, and 0/272 carry three points \"1. … 2. … 3.\" or more. The MODERN equivalent form is the abstract with declared sections: 160/272 (59 %) have a in their . ⇒ the measurement DATES the line: observed on 2026-08-08 on Apperley, Humphrey & Levy, Br. J. Pharmac. 1976 (abstract in NINE numbered points), absent from the modern corpus.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 78 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-079", "version": 1, "legacy_line": 79, "name": "Scientific notation inside names", "name_fr": "Notation scientifique dans les noms", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MERGE", "class": "SILENT_FALSE", "layer": "⑥ resolution", "injection": "NOISE", "legacy_severity": "G1", "legacy_mechanism": "M1", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 XML : 64/272 (24 %) portent un nom à lettre grecque (« β-cells »), 38/272 (14 %) une notation exponentielle (« 6 × 10 9 »), 17/272 (6 %) une constante indicée (« Ki67 », « LD 0 ») et 17/272 (6 %) un nom chiffre-tiret-sigle du type « 5-HT ». Seuls 8/272 (3 %) gardent les caractères d'indice/exposant Unicode (₂, ⁻) : ailleurs la notation est déjà APLATIE par le transport, ce qui est la fusion même que la ligne décrit. Observation d'origine 08/08/2026 : Apperley, Humphrey & Levy, Br. J. Pharmac. 1976 (annexe 79).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 XML files: 64/272 (24 %) carry a name with a Greek letter (\"β-cells\"), 38/272 (14 %) an exponential notation (\"6 × 10 9\"), 17/272 (6 %) a subscripted constant (\"Ki67\", \"LD 0\") and 17/272 (6 %) a digit-hyphen-acronym name of the type \"5-HT\". Only 8/272 (3 %) keep the Unicode subscript/superscript characters (₂, ⁻): elsewhere the notation is already FLATTENED by transport, which is the very merge the line describes. Original observation 2026-08-08: Apperley, Humphrey & Levy, Br. J. Pharmac. 1976 (annex 79).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 79 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MERGE injection=NOISE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-080", "version": 1, "legacy_line": 80, "name": "Relative quantified relations", "name_fr": "Relations quantifiées relatives", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "IRREDUCIBLE", "layer": "⑦ schema", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M3", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles de (internal corpus of 272 JATS articles) : 10/272 articles portent au moins une comparaison de la forme « N fois / N-fold plus (ou moins) X » (ex. bio-PMC12927065 : 3.34, 2.74, 2.62), et 36/272 (13 %) portent un odds/hazard ratio — un fait dont la valeur EST le facteur. L'observation qui fonde la ligne est datée du 08/08/2026 sur Apperley, Humphrey & Levy, Br. J. Pharmac. 1976 : « about 1000 times less potent », « 5-HT about 700 times less potent », « methysergide about 4500 times less potent ».", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 articles of the internal corpus: 10/272 articles carry at least one comparison of the form \"N times / N-fold more (or less) X\" (e.g. bio-PMC12927065: 3.34, 2.74, 2.62), and 36/272 (13 %) carry an odds/hazard ratio — a fact whose value IS the factor. The observation founding the line is dated 2026-08-08 on Apperley, Humphrey & Levy, Br. J. Pharmac. 1976: \"about 1000 times less potent\", \"5-HT about 700 times less potent\", \"methysergide about 4500 times less potent\".", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 80 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-081", "version": 1, "legacy_line": 81, "name": "Equipment and supplier entities", "name_fr": "Entités de matériel et de fournisseurs", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_LABEL", "class": "IRREDUCIBLE", "layer": "④ extraction", "injection": "INSERT", "legacy_severity": "G4", "legacy_mechanism": "", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : 57/272 articles (21 %) portent dans le corps au moins une mention de fournisseur entre parenthèses de la forme « (… Ltd/Inc/GmbH/Corp/B.V./AG …) » ; 50/272 (18 %) portent en outre une référence catalogue (cat. no., Cat#, catalogue number).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): 57/272 articles (21 %) carry in the body at least one parenthesised supplier mention of the form \"(… Ltd/Inc/GmbH/Corp/B.V./AG …)\"; 50/272 (18 %) additionally carry a catalogue reference (cat. no., Cat#, catalogue number).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "partial", "note": "an extra label can be deleted, a missing one cannot be added"}, "judgeable_by": ["construction", "curated"], "history": [{"date": "2026-08-09", "event": "catalogue line 81 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_LABEL injection=INSERT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-082", "version": 2, "legacy_line": 82, "name": "Landscape-oriented table", "name_fr": "Tableau en orientation paysage", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "① pixel + ② reading", "injection": "IMAGE", "legacy_severity": "G3", "legacy_mechanism": "M7", "status": "migrated_unreviewed", "seen": [{"corpus": "Biomedical PDFs 1957-1987 (migraine corpus), first observation wave", "document": "Slater, Benign recurrent vertigo, J. Neurol. Neurosurg. Psychiatry 1979 (PDF 565 KB), sixth PDF opened", "date": "2026-08-08", "observer": "house observer (LLM agent), first wave of August 2026, biomedical PDFs 1957-1987 (migraine corpus)", "excerpt": "Observé le 08/08 — Slater, *Benign recurrent vertigo*, J. Neurol. Neurosurg. Psychiatry 1979. Le tableau occupe une page entière, tourné à 90 degrés. Dix-sept colonnes, une ligne par patient. L'ordre du texte extrait n'a plus aucun rapport avec l'ordre de lecture ; le tableau pivoté contient la totalité des sept cas de l'étude.", "excerpt_en": "Observed on 2026-08-08 — Slater, *Benign recurrent vertigo*, J. Neurol. Neurosurg. Psychiatry 1979. The table fills a whole page, rotated 90 degrees. Seventeen columns, one row per patient. The order of the extracted text no longer bears any relation to the reading order; the rotated table holds all seven cases of the study.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 82 — ACCEPTED (measured absent from the modern XML corpus): measured ABSENT from the modern XML corpus", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=IMAGE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "observation transcribed from the August annex and the wave-1 journal, after the second reader noted the record had no excerpt", "by": "Fault Atlas"}], "observed_absent_in": ["Europe PMC JATS 2023-2026 (272 articles, measured 2026-08-08)"]}, {"id": "form-083", "version": 1, "legacy_line": 83, "name": "The table carries its own definitions", "name_fr": "Le tableau porte ses propres définitions", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "IRREDUCIBLE", "layer": "⑥ resolution", "injection": "MOVE", "legacy_severity": "G3", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles XML JATS : 180/272 (66 %) portent au moins un tableau, 128/272 (47 %) un pied de tableau, et 47/272 (17 %) y définissent leurs sigles — relevé « ST: title screening; SR: abstract screening; LC: full-text review; SF: studies finally included » (bio-PMC12909487) et « LC, drug loading content; EE, encapsulation efficiency » (bio-PMC12583535). 351 pieds de tableau au total.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 JATS XML articles: 180/272 (66 %) carry at least one table, 128/272 (47 %) a table footer, and 47/272 (17 %) define their acronyms there — recorded \"ST: title screening; SR: abstract screening; LC: full-text review; SF: studies finally included\" (bio-PMC12909487) and \"LC, drug loading content; EE, encapsulation efficiency\" (bio-PMC12583535). 351 table footers in total.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 83 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=MOVE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-084", "version": 1, "legacy_line": 84, "name": "Non-atomic cells", "name_fr": "Cellules non atomiques", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "SILENT_FALSE", "layer": "② reading + ⑤ anchoring", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M3", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : 180/272 articles (66 %) portent au moins un tableau ; 47 164 cellules lues, dont 4 179 non atomiques (9 %) — 494 tirets seuls, 507 intervalles, 3 178 cellules de six mots ou plus. 144/272 articles (53 %) en portent au moins une.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): 180/272 articles (66 %) carry at least one table; 47,164 cells read, of which 4,179 non-atomic (9 %) — 494 lone dashes, 507 ranges, 3,178 cells of six words or more. 144/272 articles (53 %) carry at least one.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 84 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-085", "version": 1, "legacy_line": 85, "name": "Non-standard decimal separator", "name_fr": "Séparateur décimal non standard", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "IRREDUCIBLE", "layer": "② reading", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M3", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 : les DEUX conventions coexistent dans 16 articles sur 34 (47 %) d'un corpus Europe PMC en accès libre 2024-2026. Formes réelles relevées : « 2,8 » et « 1,3 » côtoyant des décimaux à point dans le même texte. Confirmé par deux observations indépendantes du même jour : un tableau 2026 mêlant les deux, et un article 2023 écrivant « 10,6 % » en résumé français et « 10.6 % » en résumé anglais — MÊME DOCUMENT.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08: BOTH conventions coexist in 16 articles out of 34 (47 %) of an open-access Europe PMC corpus 2024-2026. Real forms recorded: \"2,8\" and \"1,3\" next to point-decimals in the same text. Confirmed by two independent observations the same day: a 2026 table mixing both, and a 2023 article writing \"10,6 %\" in the French abstract and \"10.6 %\" in the English abstract — SAME DOCUMENT.", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 85 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 5 cases and 3 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-086", "version": 1, "legacy_line": 86, "name": "One compound, several names", "name_fr": "Un composé, plusieurs noms", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "IRREDUCIBLE", "layer": "⑥ resolution", "injection": "RENAME", "legacy_severity": "G1", "legacy_mechanism": "M1", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles XML JATS de (internal corpus of 272 JATS articles) : 224/272 articles (82 %) introduisent au moins une abréviation par glose parenthétique « nom développé (SIGLE) », 2 463 gloses au total, soit 9 par article en moyenne — la parenthèse est le SEUL pont entre le nom long et le sigle. Observé le 08/08/2026 sur Cumings & Hilton, Br. J. Pharmac. 1971 : quatre natures de nom pour une substance, et 5-HT (1976) contre 5HT (1971).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 JATS XML articles of the internal corpus: 224/272 articles (82 %) introduce at least one abbreviation by parenthetical gloss \"expanded name (ACRONYM)\", 2,463 glosses in total, i.e. 9 per article on average — the parenthesis is the ONLY bridge between the long name and the acronym. Observed on 2026-08-08 on Cumings & Hilton, Br. J. Pharmac. 1971: four kinds of name for one substance, and 5-HT (1976) versus 5HT (1971).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 86 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPLIT injection=RENAME (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-087", "version": 1, "legacy_line": 87, "name": "Mixed units and conventions", "name_fr": "Unités et conventions mixtes", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "REFUSABLE", "layer": "④ extraction", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M3", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles XML JATS d'(internal corpus of 272 JATS articles) : au moins TROIS familles de mesure distinctes dans le seul corps de l'article pour 62 documents (22 %). Concentration : essai 17, cas 16, moderne 9, bio 8. Formes réelles relevées : bio-PMC12583535.xml → masse ×22, pourcent ×12, molaire ×3, durée ×2 ; bio-PMC12674831.xml → pourcent ×141, masse ×92, durée ×12, volume ×2. Formes d'écriture citées au catalogue (Cumings & Hilton, Br. J. Pharmac. 1971) : « 37° C », « −20° C », « 1,500 r.p.m. », « 0·9 % w/v », « 4 ml », « 5 × 10⁻⁸ mol », « 2,5 mg », « 30-60 min », « 2 hours ».", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 JATS XML articles of the internal corpus: at least THREE distinct measurement families in the article body alone for 62 documents (22 %). Concentration: trial 17, case 16, modern 9, bio 8. Real forms recorded: bio-PMC12583535.xml → mass ×22, percent ×12, molar ×3, duration ×2; bio-PMC12674831.xml → percent ×141, mass ×92, duration ×12, volume ×2. Written forms quoted in the catalogue (Cumings & Hilton, Br. J. Pharmac. 1971): \"37° C\", \"−20° C\", \"1,500 r.p.m.\", \"0·9 % w/v\", \"4 ml\", \"5 × 10⁻⁸ mol\", \"2,5 mg\", \"30-60 min\", \"2 hours\".", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 87 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 7 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-088", "version": 1, "legacy_line": 88, "name": "Collective entity", "name_fr": "Entité collective", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MERGE", "class": "IRREDUCIBLE", "layer": "⑥ resolution", "injection": "RENAME", "legacy_severity": "G1", "legacy_mechanism": "M2", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 XML : 101/272 (37 %) ouvrent sur un effectif suivi d'un nom de groupe (« 127 animals », « 910 participants »), 96/272 (35 %) portent un « n = … », et 87/272 (32 %) associent l'effectif à une dispersion (moyenne, SD, ±). Le sujet collectif est donc la forme MAJORITAIRE ; aucun de nos corpus fabriqués n'en contient un. Observation d'origine 08/08/2026 : Duncker et al., Br. J. Pharmac. 1986 (annexe 88).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 XML files: 101/272 (37 %) open with a headcount followed by a group noun (\"127 animals\", \"910 participants\"), 96/272 (35 %) carry an \"n = …\", and 87/272 (32 %) associate the headcount with a dispersion (mean, SD, ±). The collective subject is therefore the MAJORITY form; none of our fabricated corpora contains one. Original observation 2026-08-08: Duncker et al., Br. J. Pharmac. 1986 (annex 88).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 88 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MERGE injection=RENAME (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-089", "version": 1, "legacy_line": 89, "name": "Complex attributions and affiliations", "name_fr": "Attributions et affiliations complexes", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "IRREDUCIBLE", "layer": "④ extraction", "injection": "FORMAT", "legacy_severity": "G2", "legacy_mechanism": "M4", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles de (internal corpus of 272 JATS articles) : 106/272 (39 %) déclarent au moins deux ET au moins un auteur rattaché à deux affiliations par renvoi de note — bio-PMC12674831 en porte 57 affiliations pour 11 auteurs multi-affiliés, bio-PMC12882215 en a 4 pour 4 auteurs. Le lien auteur→institution passe par le symbole, jamais par une phrase. L'observation qui fonde la ligne est datée du 08/08/2026 sur Duncker et al., Br. J. Pharmac. 1986.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 articles of the internal corpus: 106/272 (39 %) declare at least two AND at least one author attached to two affiliations by footnote reference — bio-PMC12674831 carries 57 affiliations for 11 multi-affiliated authors, bio-PMC12882215 has 4 for 4 authors. The author→institution link goes through the symbol, never through a sentence. The observation founding the line is dated 2026-08-08 on Duncker et al., Br. J. Pharmac. 1986.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 89 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-090", "version": 1, "legacy_line": 90, "name": "Text-to-figure reference by label", "name_fr": "Renvoi texte ↔ figure par étiquette", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "IRREDUCIBLE", "layer": "⑤ anchoring", "injection": "MOVE", "legacy_severity": "G3", "legacy_mechanism": "M7", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : 63/272 articles (23 %) portent au moins une étiquette de panneau « (A) » dans une légende de figure, et 44/272 (16 %) en portent au moins DEUX dans la même légende — c'est-à-dire un couplage texte↔figure par lettre, pas une simple numérotation.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): 63/272 articles (23 %) carry at least one panel label \"(A)\" in a figure caption, and 44/272 (16 %) carry at least TWO in the same caption — that is, a text↔figure coupling by letter, not a simple numbering.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 90 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=MOVE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-091", "version": 2, "legacy_line": 91, "name": "Dense relative chronology", "name_fr": "Chronologie relative dense", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "ANACHRONISM", "class": "REFUSABLE", "layer": "⑧ coherence", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M5", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 : présent dans 3 (1,1 %), TOUS dans la famille cas (récits cliniques) — exactement là où l'annexe l'attendait. Formes réelles : cas-PMC13301386 = 16 repères relatifs / 0 date absolue (« Nine days after », « One day later », « The following day », « Three days after ») ; cas-PMC13294281 = 12 / 3 (« 3 weeks later », « On day 2 ») ; cas-PMC13293901 = 11 / 6 (« Since the onset »). Distribution du corpus : 140 documents à 0 repère, 45 à 1, décrochage net après 7.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026: present in 3 (1.1 %), ALL in the case family (clinical narratives) — exactly where the annex expected it. Real forms: case-PMC13301386 = 16 relative markers / 0 absolute dates (\"Nine days after\", \"One day later\", \"The following day\", \"Three days after\"); case-PMC13294281 = 12 / 3 (\"3 weeks later\", \"On day 2\"); case-PMC13293901 = 11 / 6 (\"Since the onset\"). Corpus distribution: 140 documents with 0 markers, 45 with 1, a clear drop after 7.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "partial", "note": "deleting the anachronistic edge removes a fact that was true within its validity interval; the true repair bounds the validity, it does not delete (reading of Coralie Bagnol-Lebon, 2026-09-09)"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 91 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=ANACHRONISM injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 2 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "repair.reachable_by_deletion changed from yes to partial: a deletion loses the fact's past validity; only a spurious edge is fully repaired by deletion", "by": "Coralie Bagnol-Lebon (second reader), applied by Fault Atlas"}]}, {"id": "form-092", "version": 2, "legacy_line": 92, "name": "Drop cap", "name_fr": "Lettrine", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "② reading", "injection": "NOISE", "legacy_severity": "G3", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "Biomedical PDFs 1957-1987 (migraine corpus), first observation wave", "document": "Imaizumi et al., Japanese Circulation Journal 1987 (PDF 591 KB), eleventh PDF opened", "date": "2026-08-08", "observer": "house observer (LLM agent), first wave of August 2026, biomedical PDFs 1957-1987 (migraine corpus)", "excerpt": "Observé le 08/08 — Imaizumi et al., *Japanese Circulation Journal* 1987. Le premier mot de l'article commence par une lettrine : « **V** ARIANT angina is caused by coronary artery vasospasm. » À l'extraction, la lettrine est un objet graphique séparé : le texte devient « V ARIANT ». Le mot cassé est le premier de l'article, et le nom de la maladie étudiée.", "excerpt_en": "Observed on 2026-08-08 — Imaizumi et al., *Japanese Circulation Journal* 1987. The first word of the article starts with a drop cap: \"**V** ARIANT angina is caused by coronary artery vasospasm.\" At extraction the drop cap is a separate graphic object: the text becomes \"V ARIANT\". The broken word is the first of the article, and the name of the disease studied.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 92 — ACCEPTED (measured absent from the modern XML corpus): measured ABSENT from the modern XML corpus", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=NOISE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "observation transcribed from the August annex and the wave-1 journal, after the second reader noted the record had no excerpt", "by": "Fault Atlas"}], "observed_absent_in": ["Europe PMC JATS 2023-2026 (272 articles, measured 2026-08-08)"]}, {"id": "form-093", "version": 1, "legacy_line": 93, "name": "Content added by digitisation", "name_fr": "Contenu ajouté par la numérisation", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "SILENT_FALSE", "layer": "② reading", "injection": "INSERT", "legacy_severity": "G2", "legacy_mechanism": "M10", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles XML JATS : 0/272 portent un tampon de chaîne de numérisation (le seul appariement, « scanned by the bare plate », phys-PMC13257740, est un faux positif). ⇒ la mesure CONFIRME le domaine « ancien seulement » du catalogue. L'observation qui fonde la ligne est celle du 08/08/2026 sur Imaizumi et al., Japanese Circulation Journal 1987 : « NII-Electronic Library Service » en pied de CHAQUE page (annexe 93).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 JATS XML articles: 0/272 carry a digitisation-chain stamp (the only match, \"scanned by the bare plate\", phys-PMC13257740, is a false positive). ⇒ the measurement CONFIRMS the \"old only\" domain of the catalogue. The observation founding the line is the one of 2026-08-08 on Imaizumi et al., Japanese Circulation Journal 1987: \"NII-Electronic Library Service\" at the foot of EVERY page (annex 93).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 93 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=INSERT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-094", "version": 1, "legacy_line": 94, "name": "Editorial metadata", "name_fr": "Métadonnées éditoriales", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "SILENT_FALSE", "layer": "④ extraction", "injection": "INSERT", "legacy_severity": "G4", "legacy_mechanism": "", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : 89/272 articles (33 %) portent un bloc de mots-clés balisé ; 98/272 (36 %) une phrase de financement (« was supported by », « funded by », « grant from »), dont 59 (22 %) dans une balise ou dédiée.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): 89/272 articles (33 %) carry a tagged keyword block; 98/272 (36 %) a funding sentence (\"was supported by\", \"funded by\", \"grant from\"), of which 59 (22 %) in a dedicated or tag.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 94 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=INSERT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-095", "version": 1, "legacy_line": 95, "name": "The study type qualifies all its facts", "name_fr": "Le type d'étude qualifie tous ses faits", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_LABEL", "class": "IRREDUCIBLE", "layer": "⑦ schema", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M12", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles XML JATS de (internal corpus of 272 JATS articles) : 90/272 résumés (33 %) nomment explicitement leur dispositif (randomized, double-blind, placebo-controlled, retrospective, cohort, case-control, cross-sectional, meta-analysis, in vitro/in vivo) — une seule ligne qui qualifie tout l'article. Observé aussi le 08/08/2026 sur Gomersall & Stuart, J. Neurol. Neurosurg. Psychiatry 1973 : « a double-blind controlled clinical trial of crossover design was conducted in 26 volunteers ».", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 JATS XML articles of the internal corpus: 90/272 abstracts (33 %) explicitly name their design (randomized, double-blind, placebo-controlled, retrospective, cohort, case-control, cross-sectional, meta-analysis, in vitro/in vivo) — a single line that qualifies the whole article. Also observed on 2026-08-08 on Gomersall & Stuart, J. Neurol. Neurosurg. Psychiatry 1973: \"a double-blind controlled clinical trial of crossover design was conducted in 26 volunteers\".", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "partial", "note": "an extra label can be deleted, a missing one cannot be added"}, "judgeable_by": ["construction", "curated"], "history": [{"date": "2026-08-09", "event": "catalogue line 95 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_LABEL injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-096", "version": 1, "legacy_line": 96, "name": "Methods as entities with a genealogy", "name_fr": "Les méthodes sont des entités avec généalogie", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "SILENT_FALSE", "layer": "⑥ resolution", "injection": "RENAME", "legacy_severity": "G3", "legacy_mechanism": "M1", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 XML (PMC récents) : 6/272 (2 %) nomment une méthode par son auteur (« method of … », « criteria of Guillain-Barré »), 3/272 (1 %) sous la forme canonique « using/according to the method of », et 4/272 (1 %) portent une filiation explicite (« adapted from Page », « adapted from MacDougall »). En regard, 55/272 (20 %) portent un « Nom (année) » de citation — soit DIX FOIS PLUS de leurres que de cibles, avec la même forme de surface. Observation d'origine 08/08/2026 : Carpi & Virno, Brit. J. Pharmacol. 1957 (annexe 96).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 XML files (recent PMC): 6/272 (2 %) name a method by its author (\"method of …\", \"criteria of Guillain-Barré\"), 3/272 (1 %) in the canonical form \"using/according to the method of \", and 4/272 (1 %) carry an explicit lineage (\"adapted from Page\", \"adapted from MacDougall\"). By contrast, 55/272 (20 %) carry a citation \"Name (year)\" — TEN TIMES MORE decoys than targets, with the same surface form. Original observation 2026-08-08: Carpi & Virno, Brit. J. Pharmacol. 1957 (annex 96).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 96 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPLIT injection=RENAME (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-097", "version": 2, "legacy_line": 97, "name": "The three PDF regimes (native, OCR layer, pure image)", "name_fr": "Les trois régimes de PDF (natif · OCR posé · image pure)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "REFUSABLE", "layer": "① pixel + ② reading", "injection": "IMAGE", "legacy_severity": "G3", "legacy_mechanism": "M7", "status": "migrated_unreviewed", "seen": [{"corpus": "Biomedical PDFs 1957-1987 (migraine corpus, 45 PDFs), first observation wave", "document": "the 45 PDFs of the migraine corpus", "date": "2026-08-08", "observer": "house observer (LLM agent), first wave of August 2026, biomedical PDFs 1957-1987 (migraine corpus)", "excerpt": "Mesuré le 08/08 sur les 45 PDF du corpus migraine : natif numérique (plusieurs polices déclarées) 8/45, 17 % · scan + couche OCR (une seule police, beaucoup de texte) 36/45, 80 % · image pure (aucun opérateur de texte) 1/45, 2 %. Ligne née d'un doute déclaré par un observateur : « aucun des six n'a été testé pour savoir s'il porte une couche de texte ». Le doute se lève par une mesure de trente secondes : compter les polices déclarées et les opérateurs de texte.", "excerpt_en": "Measured on 2026-08-08 on the 45 PDFs of the migraine corpus: native digital (several declared fonts) 8/45, 17 % · scan + OCR layer (a single font, much text) 36/45, 80 % · pure image (no text operator) 1/45, 2 %. Line born from a doubt declared by an observer: \"none of the six was tested for a text layer\". The doubt is lifted by a thirty-second measurement: count declared fonts and text operators.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 97 — ACCEPTED (measured absent from the modern XML corpus): measured ABSENT from the modern XML corpus", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=IMAGE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "observation transcribed from the August annex and the wave-1 journal, after the second reader noted the record had no excerpt", "by": "Fault Atlas"}], "observed_absent_in": ["Europe PMC JATS 2023-2026 (272 articles, measured 2026-08-08)"]}, {"id": "form-100", "version": 1, "legacy_line": 100, "name": "Sample size written in words", "name_fr": "L'effectif est écrit en toutes lettres", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "REFUSABLE", "layer": "④ extraction", "injection": "ALTER", "legacy_severity": "G3", "legacy_mechanism": "M3", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 : présent dans 80 (29 %). Concentration : essai 24, bio 13, phys 13. Formes réelles : two | five | three", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026: present in 80 (29 %). Concentration: trial 24, bio 13, phys 13. Real forms: two | five | three", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 100 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-101", "version": 1, "legacy_line": 101, "name": "Sample size is computed, not stated", "name_fr": "L'effectif ne se lit nulle part, il se calcule", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "REFUSABLE", "layer": "⑩ retrieval", "injection": "MOVE", "legacy_severity": "G3", "legacy_mechanism": "M11", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 (internal corpus of 272 JATS articles) : présent dans 25 articles (9 %). Concentration : essai 10/34, meta 7/34, bio 4/34, nonEN 4/51. Formes réelles : bio-PMC12590050 → 910 = 387 + 523 ; bio-PMC12953170 → 27 = 8 + 19 ; et bio-PMC12674831 porte TROIS décompositions concurrentes (105=45+60, 109=33+76, 109=4+105) — c'est ce cas mesuré qui fonde la clause de refus.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026 (internal corpus): present in 25 articles (9 %). Concentration: trial 10/34, meta 7/34, bio 4/34, nonEN 4/51. Real forms: bio-PMC12590050 → 910 = 387 + 523; bio-PMC12953170 → 27 = 8 + 19; and bio-PMC12674831 carries THREE competing decompositions (105=45+60, 109=33+76, 109=4+105) — this measured case is what founds the refusal clause.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 101 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=MOVE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-102", "version": 1, "legacy_line": 102, "name": "Furniture language differs from body language", "name_fr": "La langue du mobilier n'est pas celle du corps", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "② reading", "injection": "FORMAT", "legacy_severity": "G3", "legacy_mechanism": "M4", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 : présent dans 53 (19 %). Concentration : nonEN 51, bio 1, meta 1. Formes réelles : en | fa | de", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026: present in 53 (19 %). Concentration: nonEN 51, bio 1, meta 1. Real forms: en | fa | de", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 102 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 3 cases and 2 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-103", "version": 1, "legacy_line": 103, "name": "Significance marker with two meanings", "name_fr": "Marqueur de significativité à double sens", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_LABEL", "class": "IRREDUCIBLE", "layer": "⑦ schema", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M12", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles de (internal corpus of 272 JATS articles) : 1/272 seulement porte une note de tableau définissant DEUX symboles de seuil (nonEN-PMC13224957). Le chiffre CONFIRME le domaine « ancien seulement » porté par le catalogue — la difficulté est morte dans la science courante et vivante pour qui dépouille des archives. L'observation qui fonde la ligne est datée du 08/08/2026, sur les PDF du corpus migraine (1976-1986) : « + p<0·01, ++ p<0·001, +++ p<0·02 as compared with migraineurs and + p<0·02, ++ p<0·01 as compared with controls ».", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 articles of the internal corpus: only 1/272 carries a table note defining TWO threshold symbols (nonEN-PMC13224957). The figure CONFIRMS the \"old only\" domain carried by the catalogue — the difficulty is dead in current science and alive for whoever mines archives. The observation founding the line is dated 2026-08-08, on the PDFs of the migraine corpus (1976-1986): \"+ p<0·01, ++ p<0·001, +++ p<0·02 as compared with migraineurs and + p<0·02, ++ p<0·01 as compared with controls\".", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "partial", "note": "an extra label can be deleted, a missing one cannot be added"}, "judgeable_by": ["construction", "curated"], "history": [{"date": "2026-08-09", "event": "catalogue line 103 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_LABEL injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-104", "version": 1, "legacy_line": 104, "name": "No apparatus at all", "name_fr": "Absence totale d'appareil", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "② reading", "injection": "FORMAT", "legacy_severity": "G3", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles XML JATS d'(internal corpus of 272 JATS articles) : présent dans 25 (9 %) — aucun , aucun , aucun titre de section dans . Concentration nette : retract 22, nonEN 3 (ex. nonEN-PMC13170112.xml, nonEN-PMC13224953.xml, nonEN-PMC13285344.xml). Un avis de rétractation est un corps de prose nu. Forme voisine citée au catalogue : de Salles, « CEREBRAL VASOSPASM » — titre, auteurs, puis prose continue, pour tout appareil « Department of Physiology, Medical College of Virginia ».", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 JATS XML articles of the internal corpus: present in 25 (9 %) — no , no , no section title in . Sharp concentration: retract 22, nonEN 3 (e.g. nonEN-PMC13170112.xml, nonEN-PMC13224953.xml, nonEN-PMC13285344.xml). A retraction notice is a bare body of prose. Neighbouring form quoted in the catalogue: de Salles, \"CEREBRAL VASOSPASM\" — title, authors, then continuous prose, with \"Department of Physiology, Medical College of Virginia\" as the whole apparatus.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 104 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 5 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-105", "version": 1, "legacy_line": 105, "name": "Abbreviation dot inside a unit", "name_fr": "Point abréviatif dans l'unité", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "REFUSABLE", "layer": "② reading", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M3", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "« c.c. » et « m.g. » relevés dans Br. J. Pharmac. 1971-1976 (vague 1, 08/08/2026)", "lang": "fr", "excerpt_en": "\"c.c.\" and \"m.g.\" recorded in Br. J. Pharmac. 1971-1976 (wave 1, 2026-08-08)", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 105 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-106", "version": 2, "legacy_line": 106, "name": "Signature mark of a gathering", "name_fr": "Signature de cahier", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "② reading", "injection": "NOISE", "legacy_severity": "G3", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "Biomedical PDFs 1957-1987 (migraine corpus), first observation wave", "document": "a PDF of the migraine corpus (folio 149), observed by a delegated observer, proof quoted", "date": "2026-08-08", "observer": "house observer (LLM agent), first wave of August 2026, biomedical PDFs 1957-1987 (migraine corpus)", "excerpt": "Observé le 08/08 par un observateur délégué, sur les PDF du corpus migraine : « un 4 seul, en bas à gauche sous la dernière ligne d'un tableau, alors que le folio de la page est 149 en haut à droite ». Un repère de pliage des cahiers, vestige du façonnage du livre : un chiffre nu, sans contexte, rattaché par proximité à la dernière valeur du tableau au-dessus.", "excerpt_en": "Observed on 2026-08-08 by a delegated observer, on the PDFs of the migraine corpus: \"a lone 4, bottom left under the last row of a table, while the page folio is 149 top right\". A gathering signature from the bookbinder, a vestige of the printing: a bare digit, without context, attached by proximity to the last value of the table above.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 106 — ACCEPTED (measured absent from the modern XML corpus): measured ABSENT from the modern XML corpus", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=NOISE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "observation transcribed from the August annex and the wave-1 journal, after the second reader noted the record had no excerpt", "by": "Fault Atlas"}], "observed_absent_in": ["Europe PMC JATS 2023-2026 (272 articles, measured 2026-08-08)"]}, {"id": "form-107", "version": 1, "legacy_line": 107, "name": "Several DOIs per document", "name_fr": "Plusieurs DOI par document", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MERGE", "class": "DEFEATED", "layer": "④ extraction", "injection": "INSERT", "legacy_severity": "G1", "legacy_mechanism": "M1", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 : présent dans 27 (9 %). Concentration : nonEN 24, bio 1, meta 1. Formes réelles : 10.1371/journal.pcbi.1014325 | 10.1371/journal.pcbi.1014325.r001 | 10.1371/journal.pmed.1004653", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026: present in 27 (9 %). Concentration: nonEN 24, bio 1, meta 1. Real forms: 10.1371/journal.pcbi.1014325 | 10.1371/journal.pcbi.1014325.r001 | 10.1371/journal.pmed.1004653", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 107 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MERGE injection=INSERT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 3 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-108", "version": 1, "legacy_line": 108, "name": "Partial ORCID", "name_fr": "ORCID partiel", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "DEFEATED", "layer": "⑥ resolution", "injection": "ALTER", "legacy_severity": "G1", "legacy_mechanism": "M1", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "ORCID tronqués ou mal recopiés dans les blocs auteurs d'articles modernes (vague hors-bio, 08/08/2026)", "lang": "fr", "excerpt_en": "Truncated or miscopied ORCIDs in the author blocks of modern articles (non-bio wave, 2026-08-08)", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 108 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPLIT injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-109", "version": 1, "legacy_line": 109, "name": "Structured abstract", "name_fr": "Résumé structuré", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "③ utterance", "injection": "FORMAT", "legacy_severity": "G3", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles XML JATS Europe PMC (2023-2026) : présent dans 106/272 (39 %). Forme réelle, bio-PMC12882215 : le résumé est fait de Background… puis Methods, Results, Conclusions. Absent des articles 1957-1987 : la difficulté est née avec l'édition numérique.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 Europe PMC JATS XML articles (2023-2026): present in 106/272 (39 %). Real form, bio-PMC12882215: the abstract is made of Background… then Methods, Results, Conclusions. Absent from the 1957-1987 articles: the difficulty was born with digital publishing.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 109 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-110", "version": 1, "legacy_line": 110, "name": "Pre-registration identifier", "name_fr": "Préenregistrement (identifiant de registre)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_LABEL", "class": "DEFEATED", "layer": "⑩ retrieval", "injection": "INSERT", "legacy_severity": "G3", "legacy_mechanism": "M6", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "identifiants de registre en fin de résumé, articles biomédicaux 2024-2026 (vague moderne, 08/08/2026)", "lang": "fr", "excerpt_en": "registry identifiers at the end of the abstract, biomedical articles 2024-2026 (modern wave, 2026-08-08)", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true}, "repair": {"reachable_by_deletion": "partial", "note": "an extra label can be deleted, a missing one cannot be added"}, "judgeable_by": ["construction", "curated"], "history": [{"date": "2026-08-09", "event": "catalogue line 110 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_LABEL injection=INSERT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 3 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-111", "version": 1, "legacy_line": 111, "name": "Forest plot as image", "name_fr": "Forest plot en image", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "IRREDUCIBLE", "layer": "① pixel + ⑤ anchoring", "injection": "IMAGE", "legacy_severity": "G2", "legacy_mechanism": "M7", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : 36/272 articles (13 %) mentionnent un « forest plot », et 28/272 (10 %) le mentionnent DANS UNE LÉGENDE DE FIGURE — c'est-à-dire que le résultat groupé y est un objet graphique, pas une phrase.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): 36/272 articles (13 %) mention a \"forest plot\", and 28/272 (10 %) mention it INSIDE A FIGURE CAPTION — that is, the pooled result is a graphical object there, not a sentence.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 111 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=IMAGE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-112", "version": 1, "legacy_line": 112, "name": "Ordinal scale by colour", "name_fr": "Échelle ordinale par couleur", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "SILENT_FALSE", "layer": "① pixel + ④ extraction", "injection": "IMAGE", "legacy_severity": "G3", "legacy_mechanism": "M7", "status": "migrated_unreviewed", "seen": [], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 112 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=IMAGE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-113", "version": 1, "legacy_line": 113, "name": "Supplementary material cited and absent", "name_fr": "Matériel supplémentaire cité et absent", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "REFUSABLE", "layer": "⑩ retrieval", "injection": "DELETE", "legacy_severity": "G3", "legacy_mechanism": "M11", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles XML JATS Europe PMC accès libre (internal corpus of 272 JATS articles) : objet supplémentaire NOMMÉ cité SANS aucun fichier attaché dans 1 article (0,4 %) — meta-PMC13271776, qui appelle « Supplementary Fig. S1 », « Supplementary Fig. S2 », « Supplementary Fig. S3 » sans un seul ni ; cité ET attaché dans 81 (30 %) : bio 18, cas 4, essai 14, meta 31, moderne 9, phys 5. ⚠️ Prévalence FAIBLE, dite telle quelle : 1/272. Les deux versions antérieures de la sonde donnaient 115 puis 21 porteurs, tous du bruit — c'est la mesure qui a été corrigée, pas le chiffre choisi.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC JATS XML articles (internal corpus): NAMED supplementary object cited WITHOUT any attached file in 1 article (0.4 %) — meta-PMC13271776, which calls \"Supplementary Fig. S1\", \"Supplementary Fig. S2\", \"Supplementary Fig. S3\" without a single or ; cited AND attached in 81 (30 %): bio 18, case 4, trial 14, meta 31, modern 9, phys 5. ⚠️ LOW prevalence, stated as such: 1/272. The two earlier versions of the probe gave 115 then 21 carriers, all noise — it is the measurement that was corrected, not the figure that was chosen.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 113 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=DELETE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 2 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-114", "version": 1, "legacy_line": 114, "name": "Author declarations", "name_fr": "Déclarations sur les auteurs", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "DEFEATED", "layer": "④ extraction", "injection": "INSERT", "legacy_severity": "G4", "legacy_mechanism": "", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 (internal corpus of 272 JATS articles) : présent dans 190 articles (69 %) — le défaut le plus répandu de la série. Concentration : cas 34/34, meta 34/34, phys 31/34, bio 30/34, essai 28/34, moderne 15/17, nonEN 17/51, retract 1/34. Formes réelles : titres « Author contributions », « Competing interests », « Data availability », et fn-type COI-statement ×30, financial-disclosure ×12, con ×4 — contre fn-type=\"other\" ×127, qui ne dit rien et fonde la clause de refus.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026 (internal corpus): present in 190 articles (69 %) — the most widespread item of the series. Concentration: case 34/34, meta 34/34, phys 31/34, bio 30/34, trial 28/34, modern 15/17, nonEN 17/51, retract 1/34. Real forms: headings \"Author contributions\", \"Competing interests\", \"Data availability\", and fn-type COI-statement ×30, financial-disclosure ×12, con ×4 — against fn-type=\"other\" ×127, which says nothing and founds the refusal clause.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 114 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=INSERT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 5 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-115", "version": 1, "legacy_line": 115, "name": "Licence on page 1", "name_fr": "Licence en page 1", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "DEFEATED", "layer": "② reading", "injection": "INSERT", "legacy_severity": "G4", "legacy_mechanism": "", "status": "migrated_unreviewed", "seen": [], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 115 — ACCEPTED (measured absent from the modern XML corpus): measured ABSENT from the modern XML corpus", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=INSERT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}], "observed_absent_in": ["Europe PMC JATS 2023-2026 (272 articles, measured 2026-08-08)"]}, {"id": "form-116", "version": 1, "legacy_line": 116, "name": "Open peer-review furniture", "name_fr": "Mobilier de relecture ouverte", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "SILENT_FALSE", "layer": "④ extraction", "injection": "INSERT", "legacy_severity": "G4", "legacy_mechanism": "", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : 5/272 articles (2 %) portent un mobilier de relecture ouverte nommé (peer review file, reviewer report, editor's evaluation, eLife assessment) et 6/272 (2 %) un ; le mobilier voisin de déclarations sur les auteurs est lui à 203/272 (75 %). Observée le 08/08 sur 15 articles 2023-2026 (VERDICT ancien contre moderne, N10) — absente du corpus 1957-1987.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): 5/272 articles (2 %) carry named open peer-review furniture (peer review file, reviewer report, editor's evaluation, eLife assessment) and 6/272 (2 %) a ; the neighbouring furniture of author declarations is at 203/272 (75 %). Observed on 2026-08-08 on 15 articles 2023-2026 (VERDICT old versus modern, N10) — absent from the 1957-1987 corpus.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 116 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=INSERT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-117", "version": 2, "legacy_line": 117, "name": "Crossmark badge (retraction status)", "name_fr": "Badge Crossmark", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "ANACHRONISM", "class": "DEFEATED", "layer": "② reading", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M6", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "badge Crossmark relevé sur des articles 2025-2026 ; et un document de la vague « genres rares » dont un article cité AVAIT ÉTÉ ANNULÉ (observé le 08/08/2026)", "lang": "fr", "excerpt_en": "Crossmark badge recorded on 2025-2026 articles; and a document of the \"rare genres\" wave in which a cited article HAD BEEN RETRACTED (observed 2026-08-08)", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true}, "repair": {"reachable_by_deletion": "partial", "note": "deleting the anachronistic edge removes a fact that was true within its validity interval; the true repair bounds the validity, it does not delete (reading of Coralie Bagnol-Lebon, 2026-09-09)"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 117 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=ANACHRONISM injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "repair.reachable_by_deletion changed from yes to partial: a deletion loses the fact's past validity; only a spurious edge is fully repaired by deletion", "by": "Coralie Bagnol-Lebon (second reader), applied by Fault Atlas"}]}, {"id": "form-118", "version": 1, "legacy_line": 118, "name": "Unicode traps in native PDF", "name_fr": "Pièges Unicode du PDF natif", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "DEFEATED", "layer": "② reading", "injection": "NOISE", "legacy_severity": "G1", "legacy_mechanism": "M10", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "1 569 caractères U+200B mesurés dans UN seul PDF d'éditeur, injectés ENTRE les caractères des DOI de la bibliographie (observé le 08/08/2026). Et 8 écritures non latines observées dans la vague « langues » du même jour, où U+200C/U+200D sont PORTEURS DE SENS.", "lang": "fr", "excerpt_en": "1,569 U+200B characters measured in ONE single publisher PDF, injected BETWEEN the characters of the DOIs in the bibliography (observed 2026-08-08). And 8 non-Latin scripts observed in the \"languages\" wave the same day, where U+200C/U+200D CARRY MEANING.", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 118 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPLIT injection=NOISE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 7 cases and 5 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-119", "version": 1, "legacy_line": 119, "name": "Coloured hyperlinks carrying a type", "name_fr": "Hyperliens colorés porteurs de type", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_LABEL", "class": "DEFEATED_IF_XML", "layer": "⑤ anchoring", "injection": "IMAGE", "legacy_severity": "G3", "legacy_mechanism": "M4", "status": "migrated_unreviewed", "seen": [], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "partial", "note": "an extra label can be deleted, a missing one cannot be added"}, "judgeable_by": ["construction", "curated"], "history": [{"date": "2026-08-09", "event": "catalogue line 119 — ACCEPTED (measured absent from the modern XML corpus): measured ABSENT from the modern XML corpus", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_LABEL injection=IMAGE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}], "observed_absent_in": ["Europe PMC JATS 2023-2026 (272 articles, measured 2026-08-08)"]}, {"id": "form-120", "version": 1, "legacy_line": 120, "name": "Running head with dotted initials", "name_fr": "Running head à initiales pointées", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "REFUSABLE", "layer": "⑥ resolution", "injection": "RENAME", "legacy_severity": "G1", "legacy_mechanism": "M1", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles XML JATS Europe PMC (2023-2026) : présent dans 23/272 (8,5 %). Forme réelle, bio-PMC13162546 — la page 1 donne XianyunShao, et le running head porte « X. Shao et al. », graphie qui n'apparaît nulle part en page 1. Autres relevés : « Lisbeth Mogensen » / « L. Mogensen » (essai-PMC13169472), « Chloe Puett » / « C. Puett » (essai-PMC13205709).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 Europe PMC JATS XML articles (2023-2026): present in 23/272 (8.5 %). Real form, bio-PMC13162546 — page 1 gives XianyunShao, and the running head carries \"X. Shao et al.\", a spelling that appears nowhere on page 1. Other records: \"Lisbeth Mogensen\" / \"L. Mogensen\" (trial-PMC13169472), \"Chloe Puett\" / \"C. Puett\" (trial-PMC13205709).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 120 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPLIT injection=RENAME (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-121", "version": 1, "legacy_line": 121, "name": "Access format of the document (PDF, HTML, XML)", "name_fr": "Le format d'accès au document (PDF · HTML · XML)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "CORPUS_PARAMETER", "class": "DEFEATED", "layer": "② reading", "injection": "PARAMETER", "legacy_severity": "G4", "legacy_mechanism": "", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles d'(internal corpus of 272 JATS articles) : 272/272 sont du XML JATS INTÉGRAL — body ET sections balisées — dans les 8 strates (nonEN 51, bio/cas/essai/meta/phys/retract 34 chacune, moderne 17). 0/272 en front seul, 0/272 en PDF. Une CONDITION mesurée à 100 % est le résultat attendu d'une collecte qui a choisi ce format ; ce qu'elle dit de neuf, c'est que la branche PDF de ce filtre n'est éprouvée par AUCUN document réel.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 articles of the internal corpus: 272/272 are FULL JATS XML — body AND tagged sections — across the 8 strata (nonEN 51, bio/case/trial/meta/phys/retract 34 each, modern 17). 0/272 front-only, 0/272 PDF. A CONDITION measured at 100 % is the expected result of a collection that chose this format; what it says that is new is that the PDF branch of this filter is tested by NO real document.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "handle, or abstain and declare — never guess"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": [], "history": [{"date": "2026-08-09", "event": "catalogue line 121 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=CORPUS_PARAMETER injection=PARAMETER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 5 cases and 2 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-122", "version": 1, "legacy_line": 122, "name": "Numbered display equation", "name_fr": "Équation numérotée en display", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "⑤ anchoring", "injection": "FORMAT", "legacy_severity": "G3", "legacy_mechanism": "M7", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 (internal corpus of 272 JATS articles) : présent dans 33 articles (12 %), dont 29 sur 34 du lot physique/maths (85 %) contre 1 sur 34 du lot essais cliniques et 0 sur 34 du lot biomédical — la difficulté n'existe quasiment pas en biomédecine, ce qui explique qu'elle ait manqué au catalogue. Formes réelles : étiquettes « (1) », « (2) » (moderne-PMC13264967) et « 1 », « 2 », « 3 » sans parenthèses (moderne-PMC13262411) — deux conventions dans le même corpus.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026 (internal corpus): present in 33 articles (12 %), of which 29 of the 34 in the physics/maths batch (85 %) against 1 of 34 in the clinical-trial batch and 0 of 34 in the biomedical batch — the difficulty hardly exists in biomedicine, which explains why it was missing from the catalogue. Real forms: labels \"(1)\", \"(2)\" (modern-PMC13264967) and \"1\", \"2\", \"3\" without parentheses (modern-PMC13262411) — two conventions in the same corpus.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 122 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-123", "version": 1, "legacy_line": 123, "name": "Bare-number reference with ambiguous target", "name_fr": "Renvoi par numéro nu, cible ambiguë", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "SILENT_FALSE", "layer": "⑤ anchoring", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M4", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles XML JATS de (internal corpus of 272 JATS articles) : 2 280 renvois de la forme « (N) » dans le corps, répartis sur 140/272 articles (51 %) ; et 2 161 d'entre eux (94,8 %) n'ont AUCUNE amorce devant — ni « Eq. », ni « Fig. », ni « Ref. ». Seuls 117 sont annoncés par « Eq./equation » et 2 par « Ref. ». ⇒ le numéro nu est la règle, l'amorce est l'exception. Ligne repérée le 08/08/2026 sur 14 articles 2021-2026 hors biomédecine (arXiv, PLOS, Copernicus, HAL, Scientific Reports).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 JATS XML articles of the internal corpus: 2,280 references of the form \"(N)\" in the body, spread over 140/272 articles (51 %); and 2,161 of them (94.8 %) have NO lead-in — neither \"Eq.\", nor \"Fig.\", nor \"Ref.\". Only 117 are announced by \"Eq./equation\" and 2 by \"Ref.\". ⇒ the bare number is the rule, the lead-in the exception. Line spotted on 2026-08-08 on 14 articles 2021-2026 outside biomedicine (arXiv, PLOS, Copernicus, HAL, Scientific Reports).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 123 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-124", "version": 1, "legacy_line": 124, "name": "Quantifier in a column right of the equation", "name_fr": "Quantificateur en colonne à droite de l'équation", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "IRREDUCIBLE", "layer": "⑦ schema", "injection": "FORMAT", "legacy_severity": "G2", "legacy_mechanism": "M12", "status": "migrated_unreviewed", "seen": [], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 124 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-125", "version": 1, "legacy_line": 125, "name": "Numbered theorem whose body is the claim", "name_fr": "Théorème numéroté dont le corps est l'affirmation", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "IRREDUCIBLE", "layer": "③ utterance", "injection": "MOVE", "legacy_severity": "G3", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles de (internal corpus of 272 JATS articles) : 2/272 en tout, et les DEUX dans la strate physique (2/34, soit 6 %) — phys-PMC13099575 (« Theorem 1. For a SINDy-class model with a library of functions… », puis Theorem 2) et phys-PMC13113438 (Theorem 1 à 3). ZÉRO dans les six strates biomédicales du corpus. Le comptage confirme l'annexe : la ligne est hors-biomédicale, elle n'aurait jamais été vue sans changer de discipline.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 articles of the internal corpus: 2/272 in all, and BOTH in the physics stratum (2/34, i.e. 6 %) — phys-PMC13099575 (\"Theorem 1. For a SINDy-class model with a library of functions…\", then Theorem 2) and phys-PMC13113438 (Theorems 1 to 3). ZERO in the six biomedical strata of the corpus. The count confirms the annex: the line is non-biomedical, it would never have been seen without changing discipline.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 125 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=MOVE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-126", "version": 2, "legacy_line": 126, "name": "Pseudo-code with double numbering", "name_fr": "Pseudo-code à double numérotation", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "② reading", "injection": "FORMAT", "legacy_severity": "G3", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "14 articles 2021-2026 outside biomedicine (arXiv, PLOS, Copernicus, HAL, Scientific Reports), 'rare genres' wave", "document": "document identifiers not kept by the observer (lesson of 2026-08-08: 81 documents observed, no specimen archived)", "date": "2026-08-08", "observer": "house observer (LLM agent), 'rare genres' wave of August 2026", "excerpt": "Observé le 08/08 sur 14 articles de 2021-2026 hors biomédecine — physique, maths, informatique, économie, linguistique — sur arXiv, PLOS, Copernicus, HAL, Scientific Reports : chaque ligne du pseudo-code porte DEUX compteurs décalés d'un cran, celui du composeur et celui de l'algorithme. Aucune ligne du catalogue ne la couvrait ; elle n'aurait jamais été trouvée sans changer de discipline. Mesurée ensuite ABSENTE des 272 articles JATS : 11 blocs « Algorithm N », aucune double numérotation.", "excerpt_en": "Observed on 2026-08-08 on 14 articles 2021-2026 outside biomedicine — physics, maths, computer science, economics, linguistics — on arXiv, PLOS, Copernicus, HAL, Scientific Reports: each pseudo-code line carries TWO counters offset by one, the typesetter's and the algorithm's. No catalogue line covered it; it would never have been found without changing discipline. Then measured ABSENT from the 272 JATS articles: 11 'Algorithm N' blocks, no double numbering.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 126 — ACCEPTED (measured absent from the modern XML corpus): measured ABSENT from the modern XML corpus", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "observation transcribed from the August annex and the wave-1 journal, after the second reader noted the record had no excerpt", "by": "Fault Atlas"}], "observed_absent_in": ["Europe PMC JATS 2023-2026 (272 articles, measured 2026-08-08)"]}, {"id": "form-127", "version": 1, "legacy_line": 127, "name": "Structured object presented as a figure", "name_fr": "Objet structuré présenté comme figure", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "IRREDUCIBLE", "layer": "② reading", "injection": "MOVE", "legacy_severity": "G3", "legacy_mechanism": "M7", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : ZÉRO bloc ou sur les 272 articles, et les 5 légendes contenant des accolades sont du LaTeX mathématique, pas des objets structurés. ⇒ L'ABSENCE est la mesure : ce corpus est PMC (biomédical + physique) et la difficulté n'y existe pas. Elle confirme l'annexe 127, observée le 08/08/2026 sur 14 articles 2021-2026 hors biomédecine (arXiv, HAL, Copernicus) : elle n'aurait jamais été trouvée sans changer de discipline.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): ZERO or blocks across the 272 articles, and the 5 captions containing braces are mathematical LaTeX, not structured objects. ⇒ The ABSENCE is the measurement: this corpus is PMC (biomedical + physics) and the difficulty does not exist there. It confirms annex 127, observed on 2026-08-08 on 14 articles 2021-2026 outside biomedicine (arXiv, HAL, Copernicus): it would never have been found without changing discipline.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 127 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=MOVE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-128", "version": 1, "legacy_line": 128, "name": "Acronym defined by typography alone", "name_fr": "Sigle défini par la typographie seule", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "SILENT_FALSE", "layer": "⑥ resolution", "injection": "FORMAT", "legacy_severity": "G1", "legacy_mechanism": "M1", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles XML JATS : 214/272 (79 %) emploient au moins un sigle tout-capitales 3 fois ou plus SANS jamais l'introduire par « (SIGLE) » — 1 389 sigles concernés hors liste d'arrêt. La forme en apposition nue est relevée 14 fois : « NADPH, an important antioxidant » (bio-PMC12583535), « LVAD, the flow patterns » (cas-PMC13294096), « AUC, the linear trapezoidal method » (essai-PMC13254492).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 JATS XML articles: 214/272 (79 %) use at least one all-capitals acronym 3 times or more WITHOUT ever introducing it by \"(ACRONYM)\" — 1,389 acronyms concerned outside the stop list. The bare apposition form is recorded 14 times: \"NADPH, an important antioxidant\" (bio-PMC12583535), \"LVAD, the flow patterns\" (case-PMC13294096), \"AUC, the linear trapezoidal method\" (trial-PMC13254492).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 128 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPLIT injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-129", "version": 1, "legacy_line": 129, "name": "Repository cover page prepended to the article", "name_fr": "Couverture de dépôt ajoutée devant l'article", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "DEFEATED_IF_XML", "layer": "② reading", "injection": "INSERT", "legacy_severity": "G2", "legacy_mechanism": "M4", "status": "migrated_unreviewed", "seen": [], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 129 — ACCEPTED (measured absent from the modern XML corpus): measured ABSENT from the modern XML corpus", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=INSERT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}], "observed_absent_in": ["Europe PMC JATS 2023-2026 (272 articles, measured 2026-08-08)"]}, {"id": "form-130", "version": 1, "legacy_line": 130, "name": "Two conflicting paginations", "name_fr": "Deux paginations en conflit", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "⑨ structure", "injection": "FORMAT", "legacy_severity": "G2", "legacy_mechanism": "M4", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 : présent dans 102 (37 %). Concentration : nonEN 24, cas 20, bio 14. Formes réelles : fpage=646 + elocation=38430 | fpage=59 + elocation=e-rbgo200 | fpage=25 + elocation=491", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026: present in 102 (37 %). Concentration: nonEN 24, case 20, bio 14. Real forms: fpage=646 + elocation=38430 | fpage=59 + elocation=e-rbgo200 | fpage=25 + elocation=491", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 130 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 3 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-131", "version": 2, "legacy_line": 131, "name": "Display mathematics inside the abstract", "name_fr": "Le résumé contient des mathématiques en display", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "③ utterance", "injection": "FORMAT", "legacy_severity": "G3", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "14 articles 2021-2026 outside biomedicine (arXiv, PLOS, Copernicus, HAL, Scientific Reports), 'rare genres' wave", "document": "document identifiers not kept by the observer (lesson of 2026-08-08: 81 documents observed, no specimen archived)", "date": "2026-08-08", "observer": "house observer (LLM agent), 'rare genres' wave of August 2026", "excerpt": "Observé le 08/08 sur 14 articles de 2021-2026 hors biomédecine : le résumé s'interrompt sur une inégalité pleine largeur ; le résumé cesse d'être du texte. Jamais observé en biomédical. Mesuré ensuite sur les 272 JATS : 6 articles ont une formule EN LIGNE dans le résumé, tous en physique — ce n'est pas la même ligne.", "excerpt_en": "Observed on 2026-08-08 on 14 articles 2021-2026 outside biomedicine: the abstract breaks off on a full-width inequality; the abstract stops being text. Never observed in biomedicine. Then measured on the 272 JATS: 6 articles have an INLINE formula in the abstract, all in physics — not the same line.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 131 — ACCEPTED (measured absent from the modern XML corpus): measured ABSENT from the modern XML corpus", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "observation transcribed from the August annex and the wave-1 journal, after the second reader noted the record had no excerpt", "by": "Fault Atlas"}], "observed_absent_in": ["Europe PMC JATS 2023-2026 (272 articles, measured 2026-08-08)"]}, {"id": "form-132", "version": 1, "legacy_line": 132, "name": "Clickable zone differs from semantic unit", "name_fr": "Zone cliquable ≠ unité sémantique", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "⑤ anchoring", "injection": "FORMAT", "legacy_severity": "G3", "legacy_mechanism": "M4", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles XML JATS Europe PMC accès libre (internal corpus of 272 JATS articles) : citation COUPÉE par la zone cliquable dans 21 articles (7,7 %) — essai 15, meta 4, moderne 2 ; contre-forme (le lien porte la citation ENTIÈRE) dans 16 autres. Formes réelles : « Steinmetz et al. 2023 » (essai-PMC13169472) · « Victora et al. 2008 » (essai-PMC13205709) · « UNICEF. WHO. World_Bank. 2025 » (essai-PMC13205731). ⚠️ Le sens décrit dans l'annexe — lien sur le nom, année dehors — est à 0/272 : cherché sous 3 motifs, jamais trouvé.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC JATS XML articles (internal corpus): citation CUT by the clickable zone in 21 articles (7.7 %) — trial 15, meta 4, modern 2; counter-form (the link carries the WHOLE citation) in 16 others. Real forms: \"Steinmetz et al. 2023\" (trial-PMC13169472) · \"Victora et al. 2008\" (trial-PMC13205709) · \"UNICEF. WHO. World_Bank. 2025\" (trial-PMC13205731). ⚠️ The direction described in the annex — link on the name, year outside — is at 0/272: searched under 3 patterns, never found.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 132 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 5 cases and 3 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-133", "version": 1, "legacy_line": 133, "name": "HTML fragments the text", "name_fr": "Le format html fragmente le texte", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "② reading", "injection": "FORMAT", "legacy_severity": "G2", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 (internal corpus of 272 JATS articles) : présent dans 38 articles (13 %), dont 23 sur 34 du lot physique/maths. Formes réelles, lecture naïve contre lecture assemblée : bio-PMC13252842 → 274 octets, « M H F N » au lieu de « MHFN » ; essai-PMC13263161 → « V ˙ » au lieu de « V˙ » ; essai-PMC13238360 → 1 032 octets, « PV t = PV bl × 1 + Hct bl Hct t − » au lieu de « PVt=PVbl×1+HctblHctt−11−Hctbl » — et la version naïve est en plus TRONQUÉE.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026 (internal corpus): present in 38 articles (13 %), of which 23 of the 34 in the physics/maths batch. Real forms, naive reading versus assembled reading: bio-PMC13252842 → 274 bytes, \"M H F N\" instead of \"MHFN\"; trial-PMC13263161 → \"V ˙\" instead of \"V˙\"; trial-PMC13238360 → 1,032 bytes, \"PV t = PV bl × 1 + Hct bl Hct t −\" instead of \"PVt=PVbl×1+HctblHctt−11−Hctbl\" — and the naive version is moreover TRUNCATED.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 133 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 4 cases and 2 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-134", "version": 1, "legacy_line": 134, "name": "Page 1 is not representative", "name_fr": "Page 1 non représentative", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "DEFEATED_IF_XML", "layer": "⑨ structure", "injection": "INSERT", "legacy_severity": "G2", "legacy_mechanism": "M4", "status": "migrated_unreviewed", "seen": [], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 134 — ACCEPTED (measured absent from the modern XML corpus): measured ABSENT from the modern XML corpus", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=INSERT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}], "observed_absent_in": ["Europe PMC JATS 2023-2026 (272 articles, measured 2026-08-08)"]}, {"id": "form-135", "version": 1, "legacy_line": 135, "name": "Translated verbatim", "name_fr": "Verbatim traduit", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "SILENT_FALSE", "layer": "⑤ anchoring", "injection": "TRANSLATE", "legacy_severity": "G2", "legacy_mechanism": "M9", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : 1/272 article (0,4 %) porte un marqueur de traduction, pour 11/272 (4 %) qui portent une citation en bloc (102 blocs) — la difficulté est QUASI ABSENTE du biomédical, et c'est le résultat : elle a été observée le 08/08 sur 14 articles 2021-2026 HORS biomédecine (arXiv, PLOS, Copernicus, HAL, Scientific Reports). Elle n'aurait jamais été trouvée sans changer de discipline.", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): 1/272 article (0.4 %) carries a translation marker, against 11/272 (4 %) carrying a block quotation (102 blocks) — the difficulty is NEARLY ABSENT from biomedicine, and that is the result: it was observed on 2026-08-08 on 14 articles 2021-2026 OUTSIDE biomedicine (arXiv, PLOS, Copernicus, HAL, Scientific Reports). It would never have been found without changing discipline.", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 135 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=TRANSLATE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-136", "version": 1, "legacy_line": 136, "name": "Columns of different natures", "name_fr": "Colonnes de natures différentes", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "SILENT_FALSE", "layer": "② reading", "injection": "FORMAT", "legacy_severity": "G2", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles XML JATS de (internal corpus of 272 JATS articles) : 124/272 (45,6 %) portent un bloc « Funding / Competing interests / Acknowledgements » À L'INTÉRIEUR de — donc dans le flux de lecture du corps, pas dans l'appareil de fin ; 202/272 (74 %) en portent un quelque part. Ligne repérée le 08/08/2026 sur 14 articles 2021-2026 hors biomédecine (arXiv, PLOS, Copernicus, HAL, Scientific Reports).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 JATS XML articles of the internal corpus: 124/272 (45.6 %) carry a \"Funding / Competing interests / Acknowledgements\" block INSIDE — hence in the reading flow of the body, not in the end apparatus; 202/272 (74 %) carry one somewhere. Line spotted on 2026-08-08 on 14 articles 2021-2026 outside biomedicine (arXiv, PLOS, Copernicus, HAL, Scientific Reports).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 136 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=SPURIOUS_EDGE injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-137", "version": 1, "legacy_line": 137, "name": "Three kinds of number in one string", "name_fr": "Trois natures de nombre dans une chaîne", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "SILENT_FALSE", "layer": "④ extraction", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M3", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 XML : 48/272 (18 %) portent un indice chimique balisé , 2/272 (1 %) un appel de citation multiple en (« 73, », « 2– »), et 74/272 (27 %) contiennent déjà, EN TEXTE PLAT, un mot terminé par un chiffre collé à un nombre décimal (« i6.2 », « iid3.1 »). Un seul article (1/272) montre la collision complète en texte plat (« S3.38-58 ») : en JATS le balisage sépare encore les trois natures — LA COLLISION NAÎT AU RENDU, et c'est le rendu qu'on donne au lecteur. Observation d'origine 08/08/2026, 14 articles 2021-2026 hors biomédecine (annexe 137).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 XML files: 48/272 (18 %) carry a tagged chemical index, 2/272 (1 %) a multiple citation call in (\"73,\", \"2–\"), and 74/272 (27 %) already contain, IN FLAT TEXT, a word ending in a digit glued to a decimal number (\"i6.2\", \"iid3.1\"). A single article (1/272) shows the full collision in flat text (\"S3.38-58\"): in JATS the markup still separates the three kinds — THE COLLISION IS BORN AT RENDERING, and rendering is what the reader is given. Original observation 2026-08-08, 14 articles 2021-2026 outside biomedicine (annex 137).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 137 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}]}, {"id": "form-138", "version": 1, "legacy_line": 138, "name": "Mnemonic superscript that is not a power", "name_fr": "Exposant mnémonique qui n'est pas une puissance", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "SILENT_FALSE", "layer": "⑦ schema", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M12", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur les 272 articles de (internal corpus of 272 JATS articles) : 9/272 portent un exposant en lettres non arithmétique — dont meta-PMC13260247 avec « CD16^dim » et « CD16^hi » dans la même phrase, moderne-PMC13262411 avec « ^ABS » et « ^PLA ». 3 des 9 sont dans la strate physique, aucune ne relève d'un ordinal (th/st/nd écartés).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the 272 articles of the internal corpus: 9/272 carry a non-arithmetic lettered superscript — including meta-PMC13260247 with \"CD16^dim\" and \"CD16^hi\" in the same sentence, modern-PMC13262411 with \"^ABS\" and \"^PLA\". 3 of the 9 are in the physics stratum, none is an ordinal (th/st/nd excluded).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 138 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=WRONG_VALUE injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}]}, {"id": "form-139", "version": 1, "legacy_line": 139, "name": "Citation with internal locator", "name_fr": "Citation avec localisateur interne", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "⑤ anchoring", "injection": "FORMAT", "legacy_severity": "G3", "legacy_mechanism": "M4", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 : présent dans 71 (26 %). Concentration : essai 17, meta 16, phys 13. Formes réelles : (Fig. 1 ) | (Fig. 3 ) | (Fig. 1 )", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026: present in 71 (26 %). Concentration: trial 17, meta 16, phys 13. Real forms: (Fig. 1 ) | (Fig. 3 ) | (Fig. 1 )", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 139 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=FORMAT (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 3 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-140", "version": 2, "legacy_line": 140, "name": "Editorial status printed in the document", "name_fr": "Statut éditorial imprimé dans le document", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "ANACHRONISM", "class": "SILENT_FALSE", "layer": "④ extraction", "injection": "ALTER", "legacy_severity": "G2", "legacy_mechanism": "M6", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (272 Europe PMC JATS articles)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur (internal corpus of 272 JATS articles) (272 XML JATS) : 52/272 articles (19 %) portent en tête une déclaration de statut éditorial — 33 notices de rétractation (article-type=\"retraction\", 12 %), 39 documents avec un (version antérieure ou correction, 14 %), 2 déclarant explicitement une version préprint antérieure sur medRxiv. ⚠️ Le « A PREPRINT » d'arXiv, lui, est absent de ce corpus par construction : PMC ne publie que du publié (annexe 140, observée le 08/08/2026 hors PMC).", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on the internal corpus (272 JATS XML): 52/272 articles (19 %) carry an editorial-status declaration at the top — 33 retraction notices (article-type=\"retraction\", 12 %), 39 documents with a (earlier version or correction, 14 %), 2 explicitly declaring an earlier preprint version on medRxiv. ⚠️ arXiv's \"A PREPRINT\", for its part, is absent from this corpus by construction: PMC publishes only what is published (annex 140, observed on 2026-08-08 outside PMC).", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false}, "repair": {"reachable_by_deletion": "partial", "note": "deleting the anachronistic edge removes a fact that was true within its validity interval; the true repair bounds the validity, it does not delete (reading of Coralie Bagnol-Lebon, 2026-09-09)"}, "judgeable_by": ["construction", "registry", "dated_future", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 140 — GOLDEN (kept for the benchmark): irreducible or silent false", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=ANACHRONISM injection=ALTER (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-09-09", "event": "internal tool paths removed from observation notes before publication; wording otherwise unchanged", "by": "Fault Atlas"}, {"date": "2026-09-09", "event": "repair.reachable_by_deletion changed from yes to partial: a deletion loses the fact's past validity; only a spurious edge is fully repaired by deletion", "by": "Coralie Bagnol-Lebon (second reader), applied by Fault Atlas"}]}, {"id": "form-141", "version": 1, "legacy_line": 141, "name": "Multiple abstracts from different registries", "name_fr": "Résumés multiples de registres différents", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "August 2026 — observation campaign (Korela For Science)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "DEFEATED_IF_XML", "layer": "③ utterance", "injection": "DUPLICATE", "legacy_severity": "G3", "legacy_mechanism": "M8", "status": "migrated_unreviewed", "seen": [{"corpus": "observation 2026-08 (Europe PMC JATS 2023-2026 and PubMed abstracts)", "date": "2026-08-08", "observer": "house observer (LLM agent), see report", "excerpt": "MESURÉ le 08/08/2026 sur 272 articles Europe PMC accès libre 2023-2026 : présent dans 57 (20 %). Concentration : essai 17, nonEN 12, meta 9. Formes réelles : | ", "lang": "fr", "excerpt_en": "MEASURED on 2026-08-08 on 272 open-access Europe PMC articles 2023-2026: present in 57 (20 %). Concentration: trial 17, nonEN 12, meta 9. Real forms: | ", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "note": "cancelled only when the source is structured (XML/HTML); not on composed PDF"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment on 2026-09-09, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["construction", "curated", "closed_world"], "history": [{"date": "2026-08-09", "event": "catalogue line 141 — DEFEATED (cancelled by code): filter written and green on the bench", "by": "Korela For Science"}, {"date": "2026-09-09", "event": "migrated; classified damage=MISSING injection=DUPLICATE (judgment, unreviewed by a second reader)", "by": "Fault Atlas migration"}, {"date": "2026-08-08", "event": "house filter exists with 3 cases and 1 counter-examples (bench green=True); specimens to be transcribed into this record", "by": "Korela For Science"}]}, {"id": "form-142", "version": 1, "name": "Species-split gene identities", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "UNCLASSIFIED", "layer": "⑥ resolution", "injection": "RENAME", "status": "proposed", "seen": [{"corpus": "PubTator3 frozen 2026-08-18 18:08 (bioconcepts2pubtator3.gz, sha256 856353fb…)", "document": "Gene 9966 (TNFSF15 human) vs 326623 (Tnfsf15 mouse); PD-L1: 29126, 60533, 574058; PD-1: 5133, 18566; TMPRSS2: 7113, 50528; ACE2: 59272, 70008, 103231639", "date": "2026-09-07", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "07-09-2026, table gène × mois (4 929 431 gènes vivants) : « TNFSF15 humain (TL1A) rang 5 715 · Tnfsf15 souris rang 18 396 » à la même photo ; liste de juin 2021 : « 2. ACE2 gène 59272 · 15. ACE2 gène 70008 · 24. ACE2 gène 103231639 » ; « 1. PD-L1 29126 · 3. PD-L1 60533 · 4. PD-L1 574058 » (juin 2018). Réveil des vieux : « 62 places de champions = 32 noms distincts » (08-09-2026), le compte en places étant gonflé par les doublons d'espèce.", "excerpt_en": "2026-09-07, gene × month table (4,929,431 living genes): \"TNFSF15 human (TL1A) rank 5,715 · Tnfsf15 mouse rank 18,396\" in the same snapshot; June 2021 list: \"2. ACE2 gene 59272 · 15. ACE2 gene 70008 · 24. ACE2 gene 103231639\"; \"1. PD-L1 29126 · 3. PD-L1 60533 · 4. PD-L1 574058\" (June 2018). Old-gene revival: \"62 champion places = 32 distinct names\" (2026-09-08), the count in places being inflated by species duplicates.", "lang": "fr", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["curated", "registry"], "related": ["form-026", "form-086"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 1 of 23; effect on a graph: One biological object (the protein, the target) becomes two to four nodes depending on the species studied; counts and degrees split between them; a blind reunion by surface name would merge orthologues that are not the same entity. Prevalence: seen, not counted globally; 3 species duplicates on 20 places of a gene list (June 2018).", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-143", "version": 1, "name": "Gene annotation cliff after 2023", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "UNCLASSIFIED", "layer": "⑨ structure", "injection": "DELETE", "status": "proposed", "seen": [{"corpus": "PubTator3 frozen 2026-08-18 18:08 (bioconcepts2pubtator3.gz, sha256 856353fb…) joined to baseline dates", "document": "all genes; example TNFSF15 (9966)", "date": "2026-09-08", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "08-09-2026 : « liens gène-notice par an : 2022 : 7 420 k · 2023 : 6 369 k · 2024 : 1 273 k · 2025 : 1 272 k » ; « gènes cités par an : 2023 : 741 k · 2024 : 79 k » ; alors que le monde annoté toutes entités reste à « 2024 : 1 438 k · 2025 : 1 649 k » notices. TNFSF15 : « 2022 342 · 2023 250 · 2024 51 ».", "excerpt_en": "2026-09-08: \"gene-record links per year: 2022: 7,420 k · 2023: 6,369 k · 2024: 1,273 k · 2025: 1,272 k\"; \"genes cited per year: 2023: 741 k · 2024: 79 k\"; while the all-entity annotated world stays at \"2024: 1,438 k · 2025: 1,649 k\" records. TNFSF15: \"2022 342 · 2023 250 · 2024 51\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["dated_future"], "related": [], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 2 of 23; effect on a graph: Every gene-record edge after 2023 is missing; a per-gene time series collapses by 80 % while the literature did not move; a three-year judge set in 2021 reads a death. Prevalence: measured: −83 % gene links in 2024 vs 2022, on 40 M records.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-144", "version": 1, "name": "Un-normalised chemical bucket", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "MERGE", "class": "UNCLASSIFIED", "layer": "⑥ resolution", "injection": "RENAME", "status": "proposed", "seen": [{"corpus": "PubTator3 BioCXML section layer (2,576,783,200 annotations), Chemical layer, frozen 2026-08-18", "document": "entity id 4306059, key '-'", "date": "2026-08-21", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "21-08-2026 : « 224 554 246 lignes = 8,71 % du corpus section, 29,62 % des annotations Chemical (758 078 274) — confirmé à l'unité par pointage architecte indépendant (98 s) » ; « le seau franchit tous les filtres légitimes (surface=0, type et id valides) ».", "excerpt_en": "2026-08-21: \"224,554,246 lines = 8.71 % of the section corpus, 29.62 % of Chemical annotations (758,078,274) — confirmed to the unit by an independent architect check (98 s)\"; \"the bucket passes every legitimate filter (surface=0, valid type and id)\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["curated"], "related": [], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 3 of 23; effect on a graph: Nearly a third of molecule mentions point to a single node, which becomes the largest chemical node of the graph, linked to everything. Prevalence: measured, above.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-145", "version": 1, "name": "Two overlapping identifier spaces in one column", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_LABEL", "class": "UNCLASSIFIED", "layer": "⑦ schema", "injection": "ALTER", "status": "proposed", "seen": [{"corpus": "PubTator3 BioCXML section layer, column ent.u32, flag surface.u8", "document": "lines with surface=1 (126,255,447), of which 21.8 M outside the dictionary", "date": "2026-08-21", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "21-08-2026 : « ent.u32 = DEUX espaces d'ids chevauchants, surface.u8 seul discriminant » ; « 104 443 196 lignes surface=1 avec ent dans la plage des entités → 95,25 % de types DISCORDANTS (pièces : lignes Species → entités DNAMutation) ».", "excerpt_en": "2026-08-21: \"ent.u32 = TWO overlapping id spaces, surface.u8 the only discriminant\"; \"104,443,196 lines with surface=1 and ent in the entity range → 95.25 % DISCORDANT types (evidence: Species lines → DNAMutation entities)\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "read surface.u8 first; a line with surface=1 is never an entity id"}, "repair": {"reachable_by_deletion": "partial", "note": "an extra label can be deleted, a missing one cannot be added"}, "judgeable_by": ["curated"], "related": [], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 4 of 23; effect on a graph: A surface-mention identifier read as a normalised-entity identifier yields a node of the wrong type in 95 % of cases. Prevalence: measured, above.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-146", "version": 1, "name": "Entity with no normalised identity (drug hidden in the bucket)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "UNCLASSIFIED", "layer": "⑥ resolution", "injection": "RENAME", "status": "proposed", "seen": [{"corpus": "PubTator3 frozen 2026-08-18 18:08 (bioconcepts2pubtator3.gz, sha256 856353fb…)", "document": "liraglutide (no PubTator entity above F1 0.076); semaglutide", "date": "2026-08-21", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "21-08-2026 : « liraglutide sans identité PubTator — 52 % de ses notices sous la clé `-` id 4306059 » ; « sémaglutide exclu : descripteur 2026 ».", "excerpt_en": "2026-08-21: \"liraglutide without PubTator identity — 52 % of its records under key '-' id 4306059\"; \"semaglutide excluded: descriptor 2026\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["curated"], "related": ["form-003"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 5 of 23; effect on a graph: A major molecule has no node; half of its mentions are in the bucket, the other half scattered. Prevalence: seen on two molecules, not counted.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-147", "version": 2, "name": "Retro-indexing of a MeSH descriptor onto older articles", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "ANACHRONISM", "class": "UNCLASSIFIED", "layer": "⑧ coherence", "injection": "ALTER", "status": "proposed", "seen": [{"corpus": "MeSH history 2026, on the baseline frozen 2026-08-18, liens/notice_descripteur (year, attribution bound)", "document": "descriptor D000077203 (born 2019); PMID 16198559", "date": "2026-08-21", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "21-08-2026 : « D000077203 né 2019, articles dès 2006 » ; « 931 notices achevées < 2018 portent le descripteur » ; « campagne NLM 2019 = 21,22 % des bornes » ; « plancher DateRevised 2000 = re-datage total à la bascule XML ».", "excerpt_en": "2026-08-21: \"D000077203 born 2019, articles from 2006\"; \"931 records completed < 2018 carry the descriptor\"; \"NLM 2019 campaign = 21.22 % of the bounds\"; \"DateRevised floor 2000 = total re-dating at the XML switch\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "partial", "note": "deleting the anachronistic edge removes a fact that was true within its validity interval; the true repair bounds the validity, it does not delete (reading of Coralie Bagnol-Lebon, 2026-09-09)"}, "judgeable_by": ["dated_future", "registry"], "related": [], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 6 of 23; effect on a graph: A concept appears attached to articles earlier than its own birth; any dating of emergence by the descriptor is off by years. Prevalence: measured on one descriptor (931 records); 21.22 % of attribution bounds carry the 2019 campaign date.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}, {"date": "2026-09-09", "event": "repair.reachable_by_deletion changed from yes to partial: a deletion loses the fact's past validity; only a spurious edge is fully repaired by deletion", "by": "Coralie Bagnol-Lebon (second reader), applied by Fault Atlas"}]}, {"id": "form-148", "version": 1, "name": "No absolute clock in the record (dates are revision bounds)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "UNCLASSIFIED", "layer": "⑧ coherence", "injection": "ALTER", "status": "proposed", "seen": [{"corpus": "PubMed/MEDLINE baseline frozen 2026-08-18 (39,994,988 records), noeuds/notices/an.i16, an_revise.i16, an_complete.i16", "document": "all records; OLDMEDLINE break 1966; pre-1946 articles", "date": "2026-08-21", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "21-08-2026 : « le substrat ne porte aucune horloge absolue ; toute mesure temporelle = différence de deux bornes de la même matrice » ; « rupture OLDMEDLINE à 1966, articles pré-1946 re-datés médiane 2010 » ; « borne_attribution == an_revise recopiée — 100,0000 % sur 342 733 118 » ; « DateCompleted min 1965 / 11 222 286 < 2000 ».", "excerpt_en": "2026-08-21: \"the substrate carries no absolute clock; every temporal measure = difference of two bounds of the same matrix\"; \"OLDMEDLINE break at 1966, pre-1946 articles re-dated median 2010\"; \"attribution bound == revised year copied — 100.0000 % over 342,733,118\"; \"DateCompleted min 1965 / 11,222,286 < 2000\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry", "dated_future"], "related": [], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 7 of 23; effect on a graph: A revision date taken for a fact date rejuvenates old articles by half a century; the graph believes something was born in 2010 when it dates from 1947. Prevalence: measured: 342,733,118 attributions with bound = revision date; 11,222,286 records completed before 2000.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-149", "version": 1, "name": "Two publication dates per record (print vs electronic)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "UNCLASSIFIED", "layer": "⑧ coherence", "injection": "ALTER", "status": "proposed", "seen": [{"corpus": "PubMed/MEDLINE baseline frozen 2026-08-18 (39,994,988 records) (pub.an/mois and elec), working table (mois, mois_e)", "document": "all records; example PMID 1: pub 1975-06", "date": "2026-09-05", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "Règle de lecture consignée le 05-09-2026 : « an·12 + (CASE WHEN mois>0 THEN mois WHEN mois_e>0 THEN mois_e ELSE 1 END) » ; année de publication max observée dans la table de travail : 2027 (08-09-2026 : « (2027, 188900) » = 188 900 notices datées 2026, et des notices datées 2027 par avance de parution).", "excerpt_en": "Reading rule recorded on 2026-09-05: \"year·12 + (CASE WHEN month>0 THEN month WHEN month_e>0 THEN month_e ELSE 1 END)\"; maximum publication year observed in the working table: 2027 (2026-09-08: \"(2027, 188900)\" = 188,900 records dated 2026, and records dated 2027 by advance publication).", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": ["form-071"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 8 of 23; effect on a graph: Two possible months for the same fact; records 'from the future' (volume year after the freeze); a missing month replaced by January falsifies the month of birth. Prevalence: seen; share of records with missing month not counted.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-150", "version": 1, "name": "Author keywords missing on most records", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "UNCLASSIFIED", "layer": "⑨ structure", "injection": "DELETE", "status": "proposed", "seen": [{"corpus": "Author keywords of the baseline frozen 2026-08-18 (English, articles and reviews, 2011-2025)", "document": "titles layer 17,034,319 records vs author-keyword layer 10,154,266 on the same population", "date": "2026-09-08", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "08-09-2026 : couche titres 17 034 319 notices, couche mots-clés d'auteur 10 154 266 notices sur la même population ; 18 777 notices anglaises d'un fichier de 2020 « avec résumé et mots-clés » : mots-clés par notice médiane 5.", "excerpt_en": "2026-09-08: titles layer 17,034,319 records, author-keyword layer 10,154,266 records on the same population; 18,777 English records of a 2020 file \"with abstract and keywords\": median 5 keywords per record.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": ["form-062"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 9 of 23; effect on a graph: 40 % of records declare no object at all; a graph built on keywords is blind on them. Prevalence: measured: 6.9 M of 17.0 M without author keyword (40 %), 2011-2025.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-151", "version": 1, "name": "Same object under several author spellings (acronym, plural, appended acronym)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "UNCLASSIFIED", "layer": "⑥ resolution", "injection": "RENAME", "status": "proposed", "seen": [{"corpus": "Author keywords of the baseline frozen 2026-08-18 (English, articles and reviews, 2011-2025)", "document": "published 120 places of the proof dossier: 'artificial intelligence ai'; 'machine learning ml'; 'tumor microenvironment tme'; 'circular rna' / 'circular rnas'; 'explainable ai' / 'explainability'; 'checkpoint inhibitor' / 'immune checkpoint inhibitor'; 'three dimensional printing' / '3d bioprinting'; 'crispr cas9' / 'crispr/cas12a'", "date": "2026-09-08", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "08-09-2026 : deux implémentations du même banc, mêmes totaux (85 champions, 3 morts), « 17 noms différents » entre les deux listes de noms (63 contre 64 sujets distincts) — la variabilité vient des graphies ; « tumor microenvironment tme » : le mot-clé porte son propre sigle.", "excerpt_en": "2026-09-08: two implementations of the same bench, same totals (85 champions, 3 dead), \"17 different names\" between the two name lists (63 vs 64 distinct subjects) — the variability comes from spellings; \"tumor microenvironment tme\": the keyword carries its own acronym.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["curated", "closed_world"], "related": ["form-021", "form-024", "form-086"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 10 of 23; effect on a graph: One concept becomes three to five nodes; counts divide; a lexical reunion rule re-glues some and breaks others. Prevalence: seen everywhere; not counted globally.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-152", "version": 1, "name": "Publication type or institutional tag used as a topic keyword", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_LABEL", "class": "UNCLASSIFIED", "layer": "⑦ schema", "injection": "INSERT", "status": "proposed", "seen": [{"corpus": "Author keywords of the baseline frozen 2026-08-18 (English, articles and reviews, 2011-2025)", "document": "'case report'; 'sdg 3 good health and well being', 'sdg 5 gender equality', 'sdg 10 reduced inequalities'; 'surveys and questionnaires' (120 places, 2017)", "date": "2026-09-01", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "Dossier de preuve §7 : « la règle ×3 mesure une croissance, pas une naissance : elle déclare champion case report (un type de publication) » ; mesure du 01-09-2026 sur la photo 2025 : « abstenu : sdg 5 gender equality … née 2024 ».", "excerpt_en": "Proof dossier §7: \"the ×3 rule measures a growth, not a birth: it declares 'case report' (a publication type) a champion\"; measurement of 2026-09-01 on the 2025 snapshot: \"abstained: sdg 5 gender equality … born 2024\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "partial", "note": "an extra label can be deleted, a missing one cannot be added"}, "judgeable_by": ["curated"], "related": ["form-094", "form-114"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 11 of 23; effect on a graph: A 'topic' node that is actually a document genre or an editorial-policy tag; it links to everything and grows with no science behind it. Prevalence: seen; share of keywords of this kind not counted.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-153", "version": 1, "name": "One real-world event, many nodes (vocabulary satellites)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "UNCLASSIFIED", "layer": "⑥ resolution", "injection": "RENAME", "status": "proposed", "seen": [{"corpus": "Author keywords of the baseline frozen 2026-08-18 (English, articles and reviews, 2011-2025) and MeSH history 2026, on the baseline frozen 2026-08-18", "document": "'covid', 'covid 19', 'coronavirus disease 19', 'sars cov 2', 'covid 19 vaccine', 'covid 19 pneumonia', 'suspected covid-19'; MeSH D000086382 (born 2019) vs PubTator 2697049 (115 pre-pivot articles)", "date": "2026-08-21", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "21-08-2026 : « COVID = 9 satellites du même événement (fond à compter en ÉVÉNEMENTS) » ; « témoin = (identifiant, couche) — D000086382 MeSH naît 2019 mais 2697049 PubTator a 115 articles pré-pivot ».", "excerpt_en": "2026-08-21: \"COVID = 9 satellites of the same event (background to be counted in EVENTS)\"; \"witness = (identifier, layer) — MeSH D000086382 is born 2019 but PubTator 2697049 has 115 pre-pivot articles\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["curated", "dated_future"], "related": ["form-021", "form-086"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 12 of 23; effect on a graph: A single event becomes nine neighbouring nodes; novelty measures count nine births; two layers date the same thing differently. Prevalence: measured on one event (9 spellings); not counted elsewhere.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-154", "version": 2, "name": "New MeSH label for an old science", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "ANACHRONISM", "class": "UNCLASSIFIED", "layer": "⑧ coherence", "injection": "INSERT", "status": "proposed", "seen": [{"corpus": "MeSH history 2026, on the baseline frozen 2026-08-18", "document": "'Heart Disease Risk Factors' (recently created descriptor, 1,411 records at birth)", "date": "2026-08-21", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "21-08-2026 : « fond mêle chocs et événements de vocabulaire (Heart Disease Risk Factors 1 411 : un mot neuf sur une science ancienne) ».", "excerpt_en": "2026-08-21: \"background mixes shocks and vocabulary events (Heart Disease Risk Factors 1,411: a new word on an old science)\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "partial", "note": "deleting the anachronistic edge removes a fact that was true within its validity interval; the true repair bounds the validity, it does not delete (reading of Coralie Bagnol-Lebon, 2026-09-09)"}, "judgeable_by": ["dated_future"], "related": [], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 13 of 23; effect on a graph: A node that is 'born' with more than a thousand articles on its creation day: a vocabulary birth read as an emergence. Prevalence: seen; 'vocabulary events' identified by the dozen in the background, not counted.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}, {"date": "2026-09-09", "event": "repair.reachable_by_deletion changed from yes to partial: a deletion loses the fact's past validity; only a spurious edge is fully repaired by deletion", "by": "Coralie Bagnol-Lebon (second reader), applied by Fault Atlas"}]}, {"id": "form-155", "version": 2, "name": "Retraction status not linked to the retracted record", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "ANACHRONISM", "class": "UNCLASSIFIED", "layer": "⑧ coherence", "injection": "DELETE", "status": "proposed", "seen": [{"corpus": "PubMed/MEDLINE baseline frozen 2026-08-18 (39,994,988 records), liens/notice_correction, descriptors D016440 (Retracted Publication) and D016441 (Retraction of Publication)", "document": "1,429 records; PMID 367632 (1979, 'RetractionIn' towards Circulation 1984 without PMID link); PMID 9241335; file DISCORDANCE-retractations-1429-notices.json", "date": "2026-08-18", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "Fichier DISCORDANCE-retractations-1429-notices.json (Substrat, 18-08-2026) : 1 429 notices ; « D016440_sans_lien 696 · D016441_sans_lien 685 · erratum_autoboucle 48 » ; reftypes « RetractionIn 1 038 · aucun 302 · ErratumFor 49 ».", "excerpt_en": "File DISCORDANCE-retractations-1429-notices.json (Substrat, 2026-08-18): 1,429 records; \"D016440_without_link 696 · D016441_without_link 685 · erratum_self_loop 48\"; reftypes \"RetractionIn 1,038 · none 302 · ErratumFor 49\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "partial", "note": "deleting the anachronistic edge removes a fact that was true within its validity interval; the true repair bounds the validity, it does not delete (reading of Coralie Bagnol-Lebon, 2026-09-09)"}, "judgeable_by": ["registry"], "related": ["form-117", "form-140", "form-032"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 14 of 23; effect on a graph: A retracted article stays a living node with no edge to its retraction; 48 errata correct themselves (loop). Prevalence: measured: 1,429 discordant records over 3,122,924 correction links.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}, {"date": "2026-09-09", "event": "repair.reachable_by_deletion changed from yes to partial: a deletion loses the fact's past validity; only a spurious edge is fully repaired by deletion", "by": "Coralie Bagnol-Lebon (second reader), applied by Fault Atlas"}]}, {"id": "form-156", "version": 1, "name": "Citation targets outside the snapshot, and identifiers beyond 32 bits", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "UNCLASSIFIED", "layer": "⑨ structure", "injection": "DELETE", "status": "proposed", "seen": [{"corpus": "PubMed/MEDLINE baseline frozen 2026-08-18 (39,994,988 records), liens/citation (397,591,740 citations), citation_sans_cible (48,496,098), citation_cible_hors_u32", "document": "PMID 35912960 rank 39 target 2635393000000000; PMID 37341427 rank 2 target 3233922100000000 (ANOMALIE-pmid-hors-u32.json)", "date": "2026-08-18", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "CITATIONS.json : « citations 397 591 740 · cibles hors u32 2 · cibles absentes de la baseline 1 635 · corrections sans pmid 145 487 ».", "excerpt_en": "CITATIONS.json: \"citations 397,591,740 · targets beyond u32 2 · targets absent from the baseline 1,635 · corrections without pmid 145,487\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "a PMID beyond 32 bits is rejected, never guessed"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": ["form-040"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 15 of 23; effect on a graph: 48.5 M citation edges point outside the graph; two point to identifiers that cannot exist (16-digit PMIDs); 145,487 corrections without a target. Prevalence: measured, above.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-157", "version": 1, "name": "Annotations for records newer than the frozen snapshot (false recent decline)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "UNCLASSIFIED", "layer": "⑨ structure", "injection": "DELETE", "status": "proposed", "seen": [{"corpus": "PubTator3 frozen 2026-08-18 18:08 (bioconcepts2pubtator3.gz, sha256 856353fb…) against PubMed/MEDLINE baseline frozen 2026-08-18 (39,994,988 records) (max PMID 41,610,285)", "document": "REJETS-hors-instantane-par-annee.json (2026-08-19)", "date": "2026-08-19", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "REJETS-hors-instantane-par-annee.json (19-08-2026) : relations « lignes rejetées 1 238 995 · PMID distincts 278 093 · au-delà de la borne 270 047 » ; entités « lignes rejetées 6 741 580 · PMID distincts 855 858 · au-delà de la borne 843 713 » ; « 97,1 % des PMID rejetés des relations et 98,6 % de ceux des entités sont au-delà de la borne haute de l'instantané ; le reste — 8 046 pour les relations, 12 145 pour les entités — constitue le vrai manque ».", "excerpt_en": "REJETS-hors-instantane-par-annee.json (2026-08-19): relations \"rejected lines 1,238,995 · distinct PMIDs 278,093 · beyond the bound 270,047\"; entities \"rejected lines 6,741,580 · distinct PMIDs 855,858 · beyond the bound 843,713\"; \"97.1 % of rejected PMIDs for relations and 98.6 % for entities are beyond the upper bound of the snapshot; the rest — 8,046 for relations, 12,145 for entities — is the true gap\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": [], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 16 of 23; effect on a graph: Two sources frozen at different moments leave a recent edge that seems to decline; 8,046 and 12,145 PMIDs annotated but absent from the baseline are true holes. Prevalence: measured, above.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-158", "version": 1, "name": "Anonymous typed mentions (surface only, no dictionary entry)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "UNCLASSIFIED", "layer": "⑥ resolution", "injection": "DELETE", "status": "proposed", "seen": [{"corpus": "PubTator3 BioCXML section layer, frozen 2026-08-18", "document": "0.85 % of section lines; Species 1.94 %", "date": "2026-08-21", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "21-08-2026 : « 0,85 % des lignes section = mentions de surface TYPÉES hors dico (« anonymes ») » ; « Species 1,94 % anonymes (6× gènes) » ; « 21,8 M anonymes à surface = 1, inclus dans les 126,26 M ».", "excerpt_en": "2026-08-21: \"0.85 % of section lines = TYPED surface mentions outside the dictionary ('anonymous')\"; \"Species 1.94 % anonymous (6× genes)\"; \"21.8 M anonymous with surface = 1, included in the 126.26 M\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["curated"], "related": ["form-025"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 17 of 23; effect on a graph: Typed mentions without an identifier can become neither node nor edge; species suffer six times more than genes. Prevalence: measured: 21.8 M lines; 0.85 % of the section corpus; 1.94 % of species.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-159", "version": 1, "name": "Indexing source switch (human to automated) inside one field", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_LABEL", "class": "UNCLASSIFIED", "layer": "⑤ anchoring", "injection": "ALTER", "status": "proposed", "seen": [{"corpus": "PubMed/MEDLINE baseline frozen 2026-08-18 (39,994,988 records), indexing field (indexation.u16), MeSH", "document": "MTIX switch mid-2022 (bound 2021)", "date": "2026-08-21", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "Mémoire du 21-08-2026 : « bascule MTIX mi-2022 (borne 2021) » ; dossier de preuve v8 : « le juge le plus lent qui soit — la création du descripteur officiel ».", "excerpt_en": "Memory note of 2026-08-21: \"MTIX switch mid-2022 (bound 2021)\"; proof dossier v8: \"the slowest judge there is — the creation of the official descriptor\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "partial", "note": "an extra label can be deleted, a missing one cannot be added"}, "judgeable_by": ["registry"], "related": [], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 18 of 23; effect on a graph: Record-descriptor edges before and after mid-2022 are not of the same nature (human indexer, then algorithm); a series that mixes them changes source without saying so. Prevalence: seen, not counted; rests on a memory note, not re-measured in the indexing field.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-160", "version": 1, "name": "Lag between a term's use and its descriptor (concept exists before its label)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "UNCLASSIFIED", "layer": "⑧ coherence", "injection": "DELETE", "status": "proposed", "seen": [{"corpus": "Author keywords of the baseline frozen 2026-08-18 (English, articles and reviews, 2011-2025) against MeSH history 2026, on the baseline frozen 2026-08-18", "document": "federated learning (dated 2020-09, descriptor 2025-01); graph neural networks (2020-12 → 2025-01); spatial transcriptomics (2018-12 → 2026-01); tirzepatide (2020-12 → 2025-01); long covid (2022-01 → 2023-01) — public proof dossier v8 §8", "date": "2026-09-02", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "Dossier v8 : « un à sept ans » entre la datation et le descripteur ; « 3,7 ans avant MeSH en médiane » ; « signalement environ un an avant le descripteur officiel ».", "excerpt_en": "Dossier v8: \"one to seven years\" between the dating and the descriptor; \"3.7 years before MeSH in median\"; \"signalling about a year before the official descriptor\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["dated_future"], "related": [], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 19 of 23; effect on a graph: For years, thousands of articles carry an object the authority vocabulary does not know; at its creation the descriptor is retro-applied (see retro-indexing form). Prevalence: measured on five revealed positions; median 3.7 years over all positions.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-161", "version": 1, "name": "Author-declared affiliation country, detected by pattern", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "UNCLASSIFIED", "layer": "④ extraction", "injection": "ALTER", "status": "proposed", "seen": [{"corpus": "Authors table derived from the baseline (countries detected by patterns in affiliations)", "document": "lower-case countries ('usa', 'israel', 'france'); a record without affiliation has no country", "date": "2026-09-08", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "Dossier v8 : « les pays sont détectés par motifs dans les affiliations des auteurs — une notice compte pour chaque pays présent » ; 08-09-2026 : pays en minuscules (« usa », « israel », « france »), une notice sans affiliation n'a pas de pays.", "excerpt_en": "Dossier v8: \"countries are detected by patterns in the authors' affiliations — a record counts for each country present\"; 2026-09-08: lower-case countries ('usa', 'israel', 'france'), a record without affiliation has no country.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": ["form-089"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 20 of 23; effect on a graph: Record-country edges depend on a pattern; affiliations without a written country vanish; a multi-country record counts several times. Prevalence: seen, not counted (share of records without detected country not measured).", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-162", "version": 1, "name": "Funding acknowledgements: identifiers free-form, agencies multi-spelled", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "UNCLASSIFIED", "layer": "⑥ resolution", "injection": "RENAME", "status": "proposed", "seen": [{"corpus": "Funding table derived from the baseline (liens/notice_financement, 14.6 M links)", "document": "['MRC', 'Medical Research Council'] on one record; 'NIA NIH HHS'", "date": "2026-09-08", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "08-09-2026, table financements : « ['MRC', 'Medical Research Council'] » pour une même notice ; « NIA NIH HHS » ; identifiants normalisés « majuscules, sans espaces ni ponctuation » (pré-enregistrement des capitaux, 05-09-2026) ; part des notices financées 24-27 % sur les sujets suivis.", "excerpt_en": "2026-09-08, funding table: \"['MRC', 'Medical Research Council']\" on one record; \"NIA NIH HHS\"; identifiers normalised \"upper case, no spaces nor punctuation\" (capital pre-registration, 2026-09-05); share of funded records 24-27 % on the followed subjects.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": ["form-034"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 21 of 23; effect on a graph: The same agency becomes two nodes depending on spelling; grants without identifier link nothing. Prevalence: seen; share of links without identifier not counted.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-163", "version": 1, "name": "External vocabulary frozen in time (closed keyword list)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "UNCLASSIFIED", "layer": "⑦ schema", "injection": "PARAMETER", "status": "proposed", "seen": [{"corpus": "OpenAlex keywords, snapshot 2026 (outside PubMed, for comparison)", "document": "65,004 keywords, 99 % created in 2011", "date": "2026-09-06", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files", "excerpt": "06-09-2026 (NUIT B) : « vocabulaire fermé (65 004 mots-clés, 99 % créés en 2011) → 106 sujets, 6 champions ».", "excerpt_en": "2026-09-06 (night B): \"closed vocabulary (65,004 keywords, 99 % created in 2011) → 106 subjects, 6 champions\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["dated_future"], "related": [], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' (part 1; final file sha256 51a338add43cd071…), form 22 of 23; effect on a graph: A subject graph that cannot grow; any emergence after 2011 is invisible. Prevalence: measured: 99 % of keywords dated 2011.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-164", "version": 1, "name": "Homonymy in a reference registry (agency identifiers)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "MERGE", "class": "UNCLASSIFIED", "layer": "⑥ resolution", "injection": "ALTER", "status": "proposed", "seen": [], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": ["form-026"], "history": [{"date": "2026-09-09", "event": "reported second-hand by the Voûte Claude (form 23 of 23): '52 proven homonymies and about 800 estimated' in the reference data of a US agency — the measurement file is not located; recorded as a hypothesis until the evidence is found (a silent grep is not an absence)", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the Substrat measurement files"}]}, {"id": "form-165", "version": 1, "name": "Closed machine-assigned keyword vocabulary (OpenAlex)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "UNCLASSIFIED", "layer": "⑦ schema", "injection": "PARAMETER", "status": "proposed", "seen": [{"corpus": "OpenAlex snapshot 2026 (works parquet), read 2026-09-06, works.keywords and entities/keywords", "document": "65,004 keyword entries", "date": "2026-09-06", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "06-09-2026 : « référentiel de 65 004 entrées, attribué automatiquement, en moyenne 7,9 par notice » ; « 99,14 % du vocabulaire existe déjà en 2011 » ; résultat : « 106 sujets, 6 champions » là où PubMed en donne 85 sur 120.", "excerpt_en": "2026-09-06: \"reference list of 65,004 entries, assigned automatically, 7.9 per record on average\"; \"99.14 % of the vocabulary already exists in 2011\"; result: \"106 subjects, 6 champions\" where PubMed gives 85 out of 120.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["dated_future"], "related": ["form-163"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 24 of 40; effect on a graph: A frozen subject graph: any emergence after 2011 is invisible, and the record-subject link is an algorithm's decision, not an author's declaration (also WRONG_LABEL: machine keyword taken for author keyword). Prevalence: measured, above.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-166", "version": 1, "name": "Missing publication month defaulted to January (OpenAlex)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "UNCLASSIFIED", "layer": "⑧ coherence", "injection": "ALTER", "status": "proposed", "seen": [{"corpus": "OpenAlex snapshot 2026 (works parquet), read 2026-09-06, publication_date absent, publication_year only", "document": "all works 2011-2025", "date": "2026-09-06", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "06-09-2026 : « la part de \"janvier\" va de 47,4 % en 2011 à 21,4 % en 2025 — un quart à près de la moitié des notices n'ont pas de mois réel ».", "excerpt_en": "2026-09-06: \"the share of 'January' goes from 47.4 % in 2011 to 21.4 % in 2025 — a quarter to nearly half of the records have no real month\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": ["form-149"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 25 of 40; effect on a graph: A giant artificial peak every January, decreasing over time; any monthly series is false; PubMed has the same defect, much smaller. Prevalence: measured: 47.4 % (2011) to 21.4 % (2025).", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-167", "version": 1, "name": "Nested field read at the wrong level (institutions under authorships)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "UNCLASSIFIED", "layer": "② reading", "injection": "FORMAT", "status": "proposed", "seen": [{"corpus": "OpenAlex snapshot 2026 (works parquet), read 2026-09-06 parquet; house extraction script (a reader defect revealing a model trap)", "document": "1,305,653 records with an empty institutions column", "date": "2026-09-06", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "06-09-2026 : « 0 notice sur 1 305 653 a une institution » ; « l'extraction visait un champ `institutions` qui n'existe pas au premier niveau — les institutions sont sous `authorships` ; la colonne a été écrite vide, silencieusement ».", "excerpt_en": "2026-09-06: \"0 records out of 1,305,653 have an institution\"; \"the extraction targeted an 'institutions' field that does not exist at the top level — institutions are under 'authorships'; the column was written empty, silently\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "a column written entirely empty is refused, never accepted as data"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": [], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 26 of 40; effect on a graph: Two families of edges absent with no error raised; no 'country' or 'teams' signal measurable. Prevalence: measured: 100 % of records (empty column).", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-168", "version": 1, "name": "Multilingual corpus mixed without a language filter (OpenAlex)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "UNCLASSIFIED", "layer": "⑥ resolution", "injection": "TRANSLATE", "status": "proposed", "seen": [{"corpus": "OpenAlex snapshot 2026 (works parquet), read 2026-09-06, works 2011-2025", "document": "132.5 M works", "date": "2026-09-06", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "06-09-2026 : « 93,4 M de notices en anglais sur 132,5 M ; espagnol 5,4 M, portugais 3,7 M, japonais 3,7 M, allemand 3,3 M, français 3,3 M, russe 3,1 M, 2,8 M sans langue ».", "excerpt_en": "2026-09-06: \"93.4 M records in English out of 132.5 M; Spanish 5.4 M, Portuguese 3.7 M, Japanese 3.7 M, German 3.3 M, French 3.3 M, Russian 3.1 M, 2.8 M without language\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": ["form-044", "form-067"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 27 of 40; effect on a graph: Textual objects duplicate by language (also MISSING: 2.8 M without language); PubMed is de facto English, OpenAlex is not. Prevalence: measured: 30 % of records not in English, 2.1 % without language.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-169", "version": 1, "name": "Domain assigned from the record's primary topic, not from the keyword", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_LABEL", "class": "UNCLASSIFIED", "layer": "⑦ schema", "injection": "ALTER", "status": "proposed", "seen": [{"corpus": "OpenAlex snapshot 2026 (works parquet), read 2026-09-06, primary_topic.domain", "document": "1.55 M records without domain", "date": "2026-09-06", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "06-09-2026 : « Physical Sciences 51,8 M de notices, Life Sciences 16,5 M — un facteur trois » ; « pour un mot-clé transversal (\"Medicine\", \"Computer science\") ce choix est arbitraire » ; « 1,55 M de notices n'ont pas de domaine du tout ».", "excerpt_en": "2026-09-06: \"Physical Sciences 51.8 M records, Life Sciences 16.5 M — a factor of three\"; \"for a cross-cutting keyword ('Medicine', 'Computer science') this choice is arbitrary\"; \"1.55 M records have no domain at all\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "partial", "note": "an extra label can be deleted, a missing one cannot be added"}, "judgeable_by": ["curated"], "related": [], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 28 of 40; effect on a graph: The domain of a subject is that of the majority of its records, not a property of the subject; a cross-cutting keyword is filed at random. Prevalence: 1.55 M records without domain; the rest seen, not counted.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-170", "version": 1, "name": "Snapshot contains merged, retracted and paratext records (OpenAlex)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "MERGE", "class": "UNCLASSIFIED", "layer": "⑨ structure", "injection": "DUPLICATE", "status": "proposed", "seen": [{"corpus": "OpenAlex snapshot 2026 (works parquet), read 2026-09-06 parquet, before any date filter", "document": "510 M records read, ~132 M kept on 2011-2025", "date": "2026-09-06", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "06-09-2026 : « 510 M de notices lues pour ~132 M retenues sur 2011-2025 » ; « le filtre type/paratexte/rétractation seul » retient 51,6 %.", "excerpt_en": "2026-09-06: \"510 M records read for ~132 M kept on 2011-2025\"; \"the type/paratext/retraction filter alone\" keeps 51.6 %.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "records flagged merged, retracted or paratext are excluded by their own flags"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": ["form-038", "form-155"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 29 of 40; effect on a graph: Without a filter, half of the nodes are merged duplicates, retracted records, covers and tables of contents (also ANACHRONISM and WRONG_LABEL). Prevalence: measured: 48.4 % removed by this filter alone.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-171", "version": 1, "name": "Funded projects without text (OpenAIRE, NIH)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "UNCLASSIFIED", "layer": "⑨ structure", "injection": "DELETE", "status": "proposed", "seen": [{"corpus": "OpenAIRE Graph 11.1.1 (Zenodo 20428976), read 2026-09-08, entity project", "document": "NIH 2,343,783 projects", "date": "2026-09-08", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "08-09-2026 : « NIH : 2 343 783 projets · résumé 0 · mots-clés 0 · montant 1 583 418 » ; résumés présents pour UKRI 175 092, Commission européenne 102 580 sur 128 574, ANR 27 260 sur 35 344.", "excerpt_en": "2026-09-08: \"NIH: 2,343,783 projects · abstract 0 · keywords 0 · amount 1,583,418\"; abstracts present for UKRI 175,092, European Commission 102,580 of 128,574, ANR 27,260 of 35,344.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": ["form-062"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 30 of 40; effect on a graph: 60 % of projects have only a title; no project-subject edge is possible for NIH. Prevalence: measured: 534,364 projects with an abstract out of 3,909,902.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-172", "version": 1, "name": "Impossible or placeholder dates in project records (OpenAIRE)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "UNCLASSIFIED", "layer": "⑧ coherence", "injection": "ALTER", "status": "proposed", "seen": [{"corpus": "OpenAIRE Graph 11.1.1 (Zenodo 20428976), read 2026-09-08, project.startDate", "document": "FCT min startDate 0007-03-02; NSF 1900-01-01; Research Council of Finland 1950-01-01; 224,207 projects without start date", "date": "2026-09-08", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "08-09-2026 : « Fundação para a Ciência e a Tecnologia : min startDate 0007-03-02 » ; « National Science Foundation : min startDate 1900-01-01 » ; « Research Council of Finland : 1950-01-01 » ; 224 207 projets sans date de début.", "excerpt_en": "2026-09-08: \"Fundação para a Ciência e a Tecnologia: min startDate 0007-03-02\"; \"National Science Foundation: min startDate 1900-01-01\"; \"Research Council of Finland: 1950-01-01\"; 224,207 projects without a start date.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "a start date before the funder's existence is rejected as placeholder, never used"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": ["form-077"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 31 of 40; effect on a graph: Projects dated year 7 or 1900; a placeholder date taken for a true date. Prevalence: seen; 5.7 % without date; impossible dates not counted.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-173", "version": 1, "name": "Empty funder jurisdiction", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "MISSING", "class": "UNCLASSIFIED", "layer": "④ extraction", "injection": "DELETE", "status": "proposed", "seen": [{"corpus": "OpenAIRE Graph 11.1.1 (Zenodo 20428976), read 2026-09-08, project.fundings[].jurisdiction", "document": "21,292 projects", "date": "2026-09-08", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "08-09-2026 : « ('', 21 292 projets, 15 478 avec montant) ».", "excerpt_en": "2026-09-08: \"('', 21,292 projects, 15,478 with an amount)\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": ["form-161"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 32 of 40; effect on a graph: 21,292 projects without a funder country. Prevalence: measured: 21,292 out of 3,909,902.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-174", "version": 1, "name": "Inferred relations mixed with declared ones (OpenAIRE)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "UNCLASSIFIED", "layer": "⑤ anchoring", "injection": "INSERT", "status": "proposed", "seen": [{"corpus": "OpenAIRE Graph 11.1.1 (Zenodo 20428976), read 2026-09-08, project_produces (10,038,714 links), field provenance", "document": "Inferred by OpenAIRE 6,721,902 · Harvested 3,308,996 · Linked by user 7,816", "date": "2026-09-08", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "08-09-2026 : « Inferred by OpenAIRE 6 721 902 · Harvested 3 308 996 · Linked by user 7 816 ».", "excerpt_en": "2026-09-08: \"Inferred by OpenAIRE 6,721,902 · Harvested 3,308,996 · Linked by user 7,816\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["registry"], "related": [], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 33 of 40; effect on a graph: Two thirds of project-publication edges are deduced by an algorithm, one third declared; without reading the provenance one does not know which one holds (also WRONG_LABEL: provenance). Prevalence: measured: 67 % inferred edges.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-175", "version": 1, "name": "Article titles stored as entity labels (Wikidata)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_LABEL", "class": "UNCLASSIFIED", "layer": "⑥ resolution", "injection": "INSERT", "status": "proposed", "seen": [{"corpus": "Wikidata dump 2026-08-31 (121.5 M items), Almageste Wikidata edition, read 2026-09-06", "document": "tens of millions of scholarly-article items whose label_en is a full title", "date": "2026-09-06", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "06-09-2026, NUIT A : « les 121,5 M d'éléments contiennent des dizaines de millions d'articles scientifiques dont le label_en est un titre entier. Aucun filtre par type n'a été appliqué ».", "excerpt_en": "2026-09-06, night A: \"the 121.5 M items contain tens of millions of scientific articles whose label_en is a full title. No filter by type was applied\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "partial", "note": "an extra label can be deleted, a missing one cannot be added"}, "judgeable_by": ["curated"], "related": ["form-063"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 34 of 40; effect on a graph: An exact match on labels links writings to article titles; a concept is confused with a publication (MERGE by matching). Prevalence: seen; 'tens of millions', not counted.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-176", "version": 1, "name": "Same writing on several entities, no sense chosen (Wikidata)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "MERGE", "class": "UNCLASSIFIED", "layer": "⑥ resolution", "injection": "RENAME", "status": "proposed", "seen": [{"corpus": "Wikidata dump 2026-08-31 (121.5 M items), Almageste Wikidata edition, read 2026-09-06, labels and aliases", "document": "ambiguous writings; the Wiktionary 2017c edition counted 113,000 of 997,000", "date": "2026-09-06", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "06-09-2026 : « quand plusieurs Q portent la même écriture, le script ne choisit pas — il marque ambigu » ; « un objet russe peut être rattaché au mauvais Q par homographie ».", "excerpt_en": "2026-09-06: \"when several Q items carry the same writing, the script does not choose — it flags ambiguous\"; \"a Russian object can be attached to the wrong Q by homography\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["curated"], "related": ["form-026"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 35 of 40; effect on a graph: The writing does not identify the entity; any reunion by label merges homographs (or MISSING if one abstains). Prevalence: seen; number of ambiguous writings not counted in these files.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-177", "version": 1, "name": "Item number mistaken for a creation date (Wikidata)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "UNCLASSIFIED", "layer": "⑧ coherence", "injection": "ALTER", "status": "proposed", "seen": [{"corpus": "Wikidata dump 2026-08-31 (121.5 M items), Almageste Wikidata edition, read 2026-09-06, qnum", "document": "all items", "date": "2026-09-06", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "06-09-2026 : « qnum n'est qu'un ordre de création : un Q bas veut dire créé tôt dans Wikidata, pas concept ancien ; les imports de masse déplacent des millions de numéros en quelques semaines. Aucune date ne doit être calculée depuis qnum ».", "excerpt_en": "2026-09-06: \"qnum is only a creation order: a low Q means created early in Wikidata, not an old concept; mass imports move millions of numbers in a few weeks. No date must ever be computed from qnum\".", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "qnum is never read as a date"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": [], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 36 of 40; effect on a graph: Dating a concept by its item number gives false births by the million (ANACHRONISM). Prevalence: structural.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-178", "version": 1, "name": "Drug names verbatim, active ingredient field absent in early years (FAERS)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "UNCLASSIFIED", "layer": "⑥ resolution", "injection": "RENAME", "status": "proposed", "seen": [{"corpus": "FDA FAERS quarterly files 2013Q1-2026Q2, probe of 2026-09-09", "document": "files 2013-2014 without prod_ai; 49,272 distinct drugs after minimal normalisation, 17,525 with ≥ 5 reports; e.g. 'avobenzone\\octisalate\\octocrylene'", "date": "2026-09-09", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "09-09-2026 : « les fichiers 2013-2014 n'ont pas le champ prod_ai, seulement drugname (nom verbatim) » ; 49 272 médicaments distincts après normalisation minimale, 17 525 à ≥ 5 déclarations ; exemple d'objet : « avobenzone\\octisalate\\octocrylene ».", "excerpt_en": "2026-09-09: \"the 2013-2014 files have no prod_ai field, only drugname (verbatim name)\"; 49,272 distinct drugs after minimal normalisation, 17,525 with ≥ 5 reports; example object: 'avobenzone\\octisalate\\octocrylene'.", "lang": "fr", "organisation": "Loxyn SAS, Lyon", "corpus_id": "corpus-europe-pmc-jats-2023-2026"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry", "curated"], "related": ["form-021", "form-086"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 37 of 40; effect on a graph: The same molecule is one node per spelling and per combination; the seam between two field eras makes objects 'born' in 2014 (ANACHRONISM of seam). Prevalence: seen; share of multiple spellings not counted.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-179", "version": 1, "name": "Mixed character encodings across quarterly files (FAERS)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "WRONG_VALUE", "class": "UNCLASSIFIED", "layer": "② reading", "injection": "NOISE", "status": "proposed", "seen": [{"corpus": "FDA FAERS quarterly files 2013Q1-2026Q2, probe of 2026-09-09, ASCII quarterly files", "document": "one quarter not latin-1 after others read as latin-1", "date": "2026-09-09", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "09-09-2026, 01 h 14 : « Invalid Input Error: File is not latin-1 encoded » sur un trimestre après lecture réussie d'autres en latin-1 ; les fichiers mêlent les encodages selon le trimestre.", "excerpt_en": "2026-09-09, 01:14: \"Invalid Input Error: File is not latin-1 encoded\" on one quarter after others were read as latin-1; the files mix encodings by quarter.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "detect the encoding per file and convert to UTF-8 before reading; a file that fails detection is rejected, not guessed"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": ["form-118", "form-023"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 38 of 40; effect on a graph: Drug and effect names altered or lost depending on the reader (MISSING: rejected lines). Prevalence: seen, not counted.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-180", "version": 1, "name": "Case versions: one case, many records (FAERS)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "SPLIT", "class": "UNCLASSIFIED", "layer": "⑨ structure", "injection": "DUPLICATE", "status": "proposed", "seen": [{"corpus": "FDA FAERS quarterly files 2013Q1-2026Q2, probe of 2026-09-09, DEMO (caseid, caseversion, primaryid)", "document": "17,465,106 deduplicated cases (last caseversion); 14,567,407 cumulative DEMO lines at end 2022 for ~11 M distinct cases", "date": "2026-09-09", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "09-09-2026 : 17 465 106 cas dédupliqués (dernière caseversion) sur un total de lignes DEMO bien supérieur (14 567 407 lignes cumulées à fin 2022 pour ~11 M de cas distincts à cette date).", "excerpt_en": "2026-09-09: 17,465,106 deduplicated cases (last caseversion) out of a much larger total of DEMO lines (14,567,407 cumulative lines at end 2022 for ~11 M distinct cases at that date).", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": true, "refusal_clause": "keep the last caseversion per caseid; never count versions as cases"}, "repair": {"reachable_by_deletion": "no", "note": "deletion cannot restore this; classified by judgment, not measured — the EUR-Lex loop will measure it"}, "judgeable_by": ["registry"], "related": ["form-038", "form-039"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 39 of 40; effect on a graph: Without deduplication, every update of a case is a new 'case' node; pair-month counts are inflated. Prevalence: seen; lines/cases ratio not precisely recorded.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}, {"id": "form-181", "version": 1, "name": "Litigation- and media-driven report waves (FAERS)", "origin": {"organisation": "Loxyn SAS, Lyon", "campaign": "September 2026 — PubMed substrate and other corpora (Voûte)", "contact": "contact@loxyn.ai"}, "damage": "SPURIOUS_EDGE", "class": "UNCLASSIFIED", "layer": "⑨ structure", "injection": "INSERT", "status": "proposed", "seen": [{"corpus": "FDA FAERS quarterly files 2013Q1-2026Q2, probe of 2026-09-09, 2019 and 2021", "document": "'avobenzone\\octisalate\\octocrylene × skin burning sensation' 9,269 reports in twelve months (2019); 'ranitidine hydrochloride × prostate cancer' 15,565 (2021)", "date": "2026-09-09", "observer": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09", "excerpt": "09-09-2026 : « avobenzone\\octisalate\\octocrylene × skin burning sensation 9 269 déclarations en douze mois » (2019) ; « ranitidine hydrochloride × prostate cancer 15 565 » (2021) ; couples de la liste 2021 : 18 sur 20 morts au juge à trois ans. Ce n'est pas une faute de données mais une population.", "excerpt_en": "2026-09-09: \"'avobenzone\\octisalate\\octocrylene × skin burning sensation' 9,269 reports in twelve months\" (2019); \"'ranitidine hydrochloride × prostate cancer' 15,565\" (2021); pairs of the 2021 list: 18 of 20 dead at the three-year judge. Not a data fault but a population.", "lang": "fr", "organisation": "Loxyn SAS, Lyon"}], "specimens": {"cases": [], "counter_examples": []}, "prevention": {"cancellable_by_code": false, "note": "not assessed by the observer; to be classified by the second reader"}, "repair": {"reachable_by_deletion": "yes", "note": "the faulty edge can be deleted"}, "judgeable_by": ["dated_future"], "related": ["form-016"], "history": [{"date": "2026-09-09", "event": "proposed by the Voûte Claude in 'FORMES-PUBMED-pour-Fault-Atlas (09-09-2026)' part 2 (sha256 51a338add43cd071…), form 40 of 40; effect on a graph: A drug-effect edge becomes massive within months through a campaign, then vanishes; without reading the reporter one cannot tell the medical signal from the campaign. Prevalence: measured on two years: 15 of the 20 places of 2019 and 18 of the 20 of 2021 are waves.", "by": "Claude Voûte (LLM agent), session 8d9c5098, from the night reports of 2026-09-06/09"}, {"date": "2026-09-09", "event": "converted to a record; class, prevention and the novel/variant question left to the second reader", "by": "Fault Atlas"}]}]}
located OUTSIDE