probes: 2

The exact question asked of a corpus, kept verbatim (code column): a Python regex on one document, a SPARQL query on a public endpoint, a shell request. result_count/result_of is the re-run figure on result_date; matches_excerpt says whether that figure is the one quoted in the observation (0 = differs, both kept).

This data as json

id	form_id	form_name	file	kind	corpus_id	written	run	result_count	result_of	result_date	matches_excerpt	note	code
2	form-018	Variable length and chunking	probes/2026-08/line-018.py	python	corpus-europe-pmc-jats-2023-2026	2026-08-08	python3 tools/run_probe.py probes/2026-08/line-018.py <dir with the 272 .xml>	243	272	2026-09-09	1		"""Fault Atlas probe — catalogue line 18 (form-018). Extracted verbatim from the August 2026 observation campaign (Loxyn SAS, Lyon). A probe answers one question on one JATS XML file: "is this fault here, and where?" It repairs nothing and decides nothing. Returns a list of excerpts, empty if absent. Run on the reproducible corpus (see corpora/europe-pmc-jats-2023-2026-272.json): python3 tools/run_probe.py probes/2026-08/line-018.py <dir with the 272 .xml> """ import re, collections, unicodedata SONDES = {} def sonde(ligne, nom): def deco(f): SONDES[ligne] = (nom, f); return f return deco CORPS = lambda x: re.search(r'(?s)<body\b.*?</body>', x) def _txt(x): """texte brut d'un XML JATS, balises retirées""" return re.sub(r'<[^>]+>', ' ', x) @sonde(18, "document trop long pour un seul morceau (2 000 car.)") def d18(x): m = CORPS(x) if not m: return [] t = re.sub(r'\s+', ' ', _txt(m.group(0))).strip() n = -(-len(t) // 2000) return [f"{len(t)} car. → {n} morceaux"] if n > 1 else []