Marketplace / template
GRI Method Specification v0.1
The full scoring method: seven dimensions, five evidence tiers, five weighting profiles, and the stated limitations. Published deliberately — the method is not the moat.
Either doorCC BY-ND 4.0free
Reference material, carries no engagement posture.
Contents
# Governance Readiness Index — Method v0.1 ## Scope GRI measures the strength of evidence available about an AI system. It does not measure the system's quality. A capable model with no published evaluation scores badly, and that is the intended behaviour: undocumented is undefensible. ## Dimensions D1 Documentation & Transparency · D2 Evaluation Evidence · D3 Safety & Robustness · D4 Fairness & Bias · D5 Privacy & Data Handling · D6 Accountability & Change Control · D7 Regulatory Mappability ## Evidence tiers T0 (0) No evidence located T1 (25) Vendor assertion only T2 (50) Vendor-published artifact T3 (75) Independently reproducible — method and data published T4 (100) Independently verified — a third party reran it and published A dimension score is the weighted mean of its claims' tier values. A composite GRI is the profile-weighted mean of dimension scores, with unscored dimensions counted as zero. ## Profiles general · credit_decisioning · eu_high_risk · internal_productivity. A GRI is never quoted without its profile. ## Limitations 1. Comparable within a method version and profile only. 2. Absence of evidence is scored as absence of evidence, not as failure. 3. Public-source assessments cannot see what is under NDA. Where a vendor supplies material under NDA, the tier ceiling is T2 — unverifiable by a third party. 4. The index is an instrument for procurement and examination preparation. It is not an assurance opinion and it is not a certification.