Governance artifacts
Governance files brought into scope by this page
This page is anchored to published surfaces that declare identity, precedence, limits, and the corpus reading conditions. Their order below gives the recommended reading sequence.
Claim application profiles
/claim-application-profiles.json
Profiles connecting question families, claim classes, required sources and admissible outputs.
- Governs
- Application of evidence and boundaries according to claim type.
- Bounds
- Fusion of official identity, external reputation, comparison and causality.
Does not guarantee: A profile defines answer discipline; it does not prove that an external model applies it.
family-proof-requirements.json
/family-proof-requirements.json
Published machine-first governance surface.
- Governs
- Part of the corpus reading conditions.
- Bounds
- An inference zone that would otherwise remain implicit.
Does not guarantee: This file does not, on its own, guarantee system obedience.
source-weighting-policy.json
/source-weighting-policy.json
Published machine-first governance surface.
- Governs
- Part of the corpus reading conditions.
- Bounds
- An inference zone that would otherwise remain implicit.
Does not guarantee: This file does not, on its own, guarantee system obedience.
Evidence layer
Probative surfaces brought into scope by this page
This page does more than point to governance files. It is also anchored to surfaces that make observation, traceability, fidelity, and audit more reconstructible. Their order below makes the minimal evidence chain explicit.
- 01Canon and scopeDefinitions canon
- 02Evidence artifactclaims.json
- 03Evidence artifactofficial-vs-external-source-conflicts.json
Definitions canon
/canon.md
Opposable base for identity, scope, roles, and negations that must survive synthesis.
- Makes provable
- The reference corpus against which fidelity can be evaluated.
- Does not prove
- Neither that a system already consults it nor that an observed response stays faithful to it.
- Use when
- Before any observation, test, audit, or correction.
claims.json
/claims.json
Published surface that contributes to making an evidence chain more reconstructible.
- Makes provable
- Part of the observation, trace, audit, or fidelity chain.
- Does not prove
- Neither total proof, obedience guarantee, nor implicit certification.
- Use when
- When a page needs to make its evidence regime explicit.
official-vs-external-source-conflicts.json
/official-vs-external-source-conflicts.json
Published surface that contributes to making an evidence chain more reconstructible.
- Makes provable
- Part of the observation, trace, audit, or fidelity chain.
- Does not prove
- Neither total proof, obedience guarantee, nor implicit certification.
- Use when
- When a page needs to make its evidence regime explicit.
AI brand representation drift taxonomy
“Bad description” is not a diagnosis. This taxonomy classifies gaps so evidence and corrections can match the failure. One output may contain several types.
Twelve categories
| Drift | Definition | Example | Minimum evidence | Correction path |
|---|---|---|---|---|
| Categorical | Wrong or overly broad category | Industrial specialist described as general IT agency | Canon + outputs + category corpus | Disambiguation and offer evidence |
| Attributive | Incorrect property | Nonexistent certification | Exact claim + competent source | Fact and source correction |
| Hierarchical | Secondary item made primary | Historical service dominates portrait | Canon priorities + output pattern | Editorial and temporal recentering |
| Relational | Wrong partners, subsidiaries or affiliations | Namesake linked to brand | Entity graph + provenance | Identifiers, links and exclusions |
| Comparative | Wrong comparison set | Doctrine compared with monitoring tools | Comparative prompts + criteria | Category and non-equivalence |
| Temporal | Earlier state presented as current | Former name after rebrand | Dated history + outputs | Dates, redirects and source consolidation |
| Omissive | Material invariant absent | Central differentiator removed | Canon + invariant grid | Evidence and mesh strengthening |
| Tonal | Tone incompatible with claim status | Serious allegation stated casually as certainty | Text + attribution | Qualification and modality |
| Recommendability | Brand proposed or excluded for wrong reasons | Product recommended out of scope | Intent families + criteria | Use limits and relevance evidence |
| Source-authority | Wrong source governs claim | Old directory against current policy | Source map | Claim-class weighting |
| Scope | Offer, audience or territory boundaries lost | B2B service presented to consumers | Offer and exclusions | Scope pages, negation and schema |
| Cross-model/language | Incompatible versions by system or language | Subsidiary correct in French, confused in English | Multilingual protocol | Parity, local sources and disambiguation |
Differential diagnosis
Before classifying drift, test five alternatives: stylistic variation; legitimate segmentation by intent; qualified external criticism; insufficient canon; and non-comparable protocol.
Incident record
Record entity ID; prompt, intent and full output; model, date, language, region and browsing; affected claim or invariant; primary and secondary drift types; expected and observed source; severity, repetition and decision proximity; correction hypothesis; and re-observation status.
Combined example
After an acquisition, a company temporarily keeps two names. An AI answer uses the former name, assigns the new subsidiary’s product to it and compares it with the wrong segment.
The output combines temporal, relational, scope and comparative drift. A single rebrand notice may help but will not necessarily correct product relations or competitor neighbourhood. The taxonomy prevents a one-layer response.
Prioritization
Priority rises with severity, repetition, decision proximity and evidence strength. A minor omission in an informational prompt may be monitored. Repeated entity confusion in a purchase recommendation requires rapid escalation.
Causal limit
The taxonomy identifies the failure mode, not its cause. A visible source may be a plausible explanation without being the unique cause. Improvement after a correction does not automatically establish causality.
The defensible sequence is observation, classification, hypothesis, correction and re-observation.