Why Integrity Matters
Any rating system creates incentives for rated entities to optimize for the metric rather than the underlying governance quality. Alpha builds anti-gaming protections directly into the methodology.Gaming Resistance Mechanisms
Disclosure Timing Flags
When a company publishes governance improvements within 30 days of an announced Alpha review or rating update, the disclosure receives a timing flag. Flagged disclosures receive reduced weight (0.5x) until independent evidence corroborates the claims.Governance Under Stress (Indicator 2I)
The single highest-value signal in the framework. This measures documented instances where governance constrained commercial decisions. A company that delayed a product launch due to safety review findings scores higher than one that published a safety policy but never applied it. Evidence that governance held under competitive pressure earns the highest marks. This signal is structurally difficult to game because it requires sacrificing short-term commercial outcomes.Signal Tier Weighting
Evidence from T1-T4 sources (regulatory actions, verified documents, expert assessments, proprietary analysis) receives 3x the weight of T5 evidence (public disclosures, corporate websites). A company cannot improve its rating solely by publishing better marketing materials.Cross-Signal Inference
Contradictory signals trigger investigation flags. If a company claims robust safety testing (D2) but has multiple AI incidents on record (AIID database), the cross-signal check flags the inconsistency for analyst review.Independence Safeguards
Two-Track Architecture
Public Profile ratings are produced independently without company engagement. No rated company can influence its Public Profile by purchasing services. Assessed Ratings involve direct engagement but require:- founder review by Steven Wolfe Pereira
- Conflict disclosure on record
- Firewall between advisory and rating functions
Conflict Check Protocol
Before any enterprise advisory engagement exceeding $5K with a rated company:- All active ratings for the company are identified
- Steven Wolfe Pereira is notified for direct review
- Firewall is documented
- Conflict disclosure is added to the rating record
Frontier Model Rating Rule
Alpha rates frontier model companies (Anthropic, OpenAI, Google DeepMind, Meta AI, xAI) on public data without seeking permission or offering courtesy reviews. Independence creates credibility.Concern Ceiling Logic
A critical failure in a high-weight dimension caps the overall grade, preventing companies from “burying” a governance failure with strong scores elsewhere:Rating Error Protocol
When a rated company experiences a material AI governance failure within 24 months of an Alpha rating:- Incident logged within 48 hours
- Post-mortem published within 30 days covering: what the rating said, what signals were present but weighted incorrectly, what signals were absent, and what methodology changes result
- All companies in the same sector with similar profiles reviewed by Steven Wolfe Pereira