Skip to content

Framework

AI perception stability matrix

Matrix for qualifying AI perception stability across identity, category, perimeter, evidence, temporality, recommendability, and cross-system convergence.

CollectionFramework
TypeMatrix
Layertransversal
Version1.0
Stabilization2026-05-15
Published2026-05-15
Updated2026-08-08

AI perception stability matrix

This matrix qualifies representational stability without requiring identical answers. It tracks material brand invariants across models, languages, regions, intents and time.

Prerequisite: a dated baseline

The baseline separates identity; category and specialization; scope, audiences and exclusions; provable differentiators; relations; temporal state; and external claims that are not governed by the brand canon.

Axes

Axis Stable Fragile Drifted
Identity Correct entity Depends on exact name Fusion or misattribution
Category Role and specialization preserved Broad but compatible Wrong market or material generalization
Scope Offer and limits respected Minor omission Invented, old or out-of-audience offer
Differentiation Invariants recognizable Partial Interchangeable brand
Sources Competent authority by claim Incomplete attribution Wrong source class
Time Current state and dates Ambiguous Old state shown as current
Relations Correct affiliations and competitors Fluctuating neighbourhood Wrong entity or relation
Recommendability Aligned use cases Weak reasons Wrong use or unjustified exclusion
Cross-model Compatible variation One system diverges Persistent incompatible versions
Cross-language Invariant parity Local gaps Different identity or category

Levels

0, insufficient presence; 1, fragile presence; 2, partial representation; 3, conditional fidelity; 4, cross-system stability; 5, governable stability with versioned observation and classified gaps.

Sampling

Test definition, comparison, recommendation, problem, alternatives, verification and risk. Separate brand-named from generic prompts. Cover material languages and regions. Preserve observation count and confidence for every cell.

Reading results

High stability with low fidelity means a repeated error. Low stability with medium fidelity means variability. High visibility with category drift means presence success and positioning failure. Positive sentiment with false scope is flattering but unsafe. Visible official sources with structuring third parties indicate exogenous governance.

Limits

The matrix does not predict future outputs or prove correction effects. It does not penalize diverse wording when invariants remain. It does not make official canon authoritative over external reputation.

It determines where to investigate, which drift to classify and which conditions to reproduce during re-observation.