Methodology · v3.3

RCTHistorique réglementaire et légal

Composite negative-signal index for {legal_name} regulatory/legal exposure (5y rolling). HIGHER = CLEANER track record. EXTRACTION RULES: return type="composite_signal" with sub-signals, each 0-100 + source_url. ANTI-PLACEHOLDER STRICT: each source_url MUST point to a SPECIFIC enforcement filing / court docket / news article about ONE incident. Generic homepage URLs (ofac.treasury.gov root, reuters.com root, ft.com root, ec.europa.eu root, sec.gov root) are REJECTED as evidence — citing a homepage means you did not find a specific event. If no events for a sub-signal, OMIT it from signals array (do NOT emit with placeholder value=95). The composite_signal_score works on 2-3 real sub-signals; an all-95-placeholder composite is a protocol violation. Required sub-signals: (a) enforcement_actions — 0-100 inverse map from documented OFAC/SEC/DOJ/EU/national-regulator enforcement settlements in 5y (0 events→95, 1-2→70, 3-5→45, 6+→20). CATEGORIZATION RULE: FCPA / anti-bribery / antitrust / sanctions settlements WITH a monetary penalty are ENFORCEMENT_ACTIONS (NOT controversy_signal) — e.g. a $788M FCPA bribery settlement in the 5y window is a MAJOR enforcement action and MUST set enforcement_actions to ≤45 (a single ≥$100M penalty = at least the "1-2 events major" band, NOT 95). Do NOT score enforcement_actions=95 (clean) while simultaneously describing a large settlement in the reasoning — that is contradictory. A settlement ≥$100M in 5y → enforcement_actions ≤45; ≥$500M → ≤35. REQUIRED URL FORMAT: ofac.treasury.gov/recent-actions/<date>, sec.gov/litigation/litreleases/<year>/<release-id>, justice.gov/usao-XXX/pr/<release-slug>, ec.europa.eu/commission/presscorner/detail/<lang>/<release-id>, ASCII full URL not homepage. (b) litigation_volume — 0-100 inverse from named litigation events 5y (court docket URLs from courtlistener.com, pacer.gov, justia.com — court CASE-specific URL required, not homepage). (c) active_investigations — 0-100 inverse from ongoing probes (EU Commission case URL, DOJ Antitrust press release, CMA case page — case-specific). (d) controversy_signal — 0-100 inverse from documented ESG/governance controversies in tier-1 press 5y (reuters.com/article-specific URL, bloomberg.com/news/articles/<id>, ft.com/content/<id> — never homepages). Cite each event with date + URL + severity (minor/major/critical). For mega-cap public companies with extensive disclosure, low-severity events are EXPECTED (1-2 events ≠ red flag). NEGATIVE-SIGNAL CLEAN-RECORD ENFORCEMENT (mandatory): this is an inverse index — a clean top-band result (sub-signals → 90-95) is published ONLY when your reasoning AFFIRMS a verified search. To emit any clean/near-clean sub-signal (≥90 because "no events"), the reasoning MUST explicitly state (a) the HOSTNAME(s) of the official regulator/court source(s) you actually searched (e.g. ofac.treasury.gov, sec.gov, justice.gov, ec.europa.eu, courtlistener.com — or the company's home-country equivalents), (b) the EXPLICIT 60-month (5-year) rolling window searched, and (c) an EXPLICIT negation: "no enforcement action / litigation / investigation / controversy against this specific company was found in that window". A clean sub-signal WITHOUT all three of (a)+(b)+(c) is an INCOMPLETE search, not a verified clean record — OMIT that sub-signal rather than scoring it 95. Never describe a real settlement/probe/controversy in the reasoning while scoring the matching sub-signal clean (contradiction → that sub-signal must reflect the event). Server applies composite_signal_score. PRESERVES methodology of legacy compliance-history, litigation-footprint, regulatory-heat-map, controversy-history.