How a Foresain assessment reaches its verdict — and why it stands up to scrutiny: before the board, before the insurer, and in court.
An AI assessor reads the sources and extracts facts — each with its source URL and, where possible, a verbatim quote. A second, independent verification step checks every single fact against the sources. That is the agentic edge: language understanding exactly where language understanding is needed.
Everything after that is deterministic code without AI involvement: weighting, score, rating, red flags and evidence grades follow versioned, disclosed rules and formulas. Two runs on the same version and the same source snapshot produce the same result at the core.
Open research as of the record date; every document is frozen with URL, publication date and SHA-256 checksum (WORM principle — written once, never altered).
The assessor answers one guiding question per indicator — from the sources only, no prior knowledge, no assumptions. The absence of evidence is disclosed as well.
A second review step confirms, weakens or rejects every fact. Only confirmed facts carry scores and red flags.
Scored strictly against published scale anchors per indicator. What cannot be assessed is disclosed, not guessed — coverage is stated in the report.
Weighted module score → GDS (0–100, lower is better) → rating A–D, plus rule-based red flags and evidence grades. Reproducible and auditable.
The Governance Distress Score condenses a module's weighted maturity levels onto a 0–100 scale: a module with top marks throughout would score 0, one without any evidence 100. The rating bands are fixed:
| A | robust — GDS below 25. Documented, effective governance processes. |
| B | watch — below 45. Solid substance with identifiable gaps. |
| C | elevated — below 65. Structural weaknesses with distress relevance. |
| D | acute — 65 and above. Governance distress requiring immediate action. |
Red flags mark findings that constitute a warning signal in their own right — such as a documented incident without any board involvement. They are not assigned by the AI but derived from versioned rules: from a critical maturity constellation or from a verified event fact. Each red flag adds +5 points to its module's GDS and is individually substantiated in the report — including the source that triggered it.
Every finding carries an evidence grade — rule-based, derived from the source class of the supporting source and the verification result, not from the model's self-assessment:
| A | Primary source (company filing, register, court ruling), statement quoted verbatim. |
| B | Solid secondary source with a concrete, quotable company reference. |
| C | Derived from multiple sources, or generic without direct reference. |
| D | No robust evidence. |
The Evidence Confidence Index (ECI) condenses the evidence grades into a weighted figure (0–100%): it answers the question "how well is this verdict substantiated?" — regardless of what the verdict is. Maturity and evidence grade are deliberately separate: a weak finding can be excellently evidenced, and that is exactly what makes it defensible.
When several module assessments exist for the same company, the composite GDS is the weighted mean of the module GDS values across versioned portfolio weights. Since each module GDS already contains its red flags, a flag acts on the overall picture in proportion to its module's weight — the number of modules does not distort the result. Missing modules are transparently renormalised and disclosed as weight coverage.
Before release, every module is calibrated against a golden case: an independently hand-researched reference report on the same company. The automated run must reproduce it within defined tolerances — per indicator, in score, in GDS, in the exact set of red flags, in fact coverage and in the key findings. Only then does a module go into production.
Everything that influences the result — indicators, scale anchors, weights, rules, source classes — is versioned and disclosed in the report. Each run's source corpus is stored as a WORM snapshot. The deterministic core is protected by a growing regression test suite that runs on every change.
The GDS is not a rating within the meaning of the EU Credit Rating Agencies Regulation and no legal, tax or investment advice. Assessments cover the publicly evidenced outside view as of the record date; inside-view modules (stage 2) are separately marked and mandate-only.