AI perception stability matrix
The AI perception stability matrix qualifies the quality of a generated representation. It does not only measure whether the entity is visible. It measures whether the entity is reconstructed with enough fidelity to remain recognizable, comparable, and governable.
Reading axes
| Axis | Question | Stability signal | Drift signal |
|---|---|---|---|
| Identity | Who is reconstructed? | Entity is named correctly | Fusion, confusion, ambiguous attribution |
| Category | In which frame? | Market or role is exact | Category is too broad or wrong |
| Perimeter | What does the entity do? | Limits are preserved | Older or invented offer |
| Evidence | What supports the answer? | Canonical or admissible sources | Secondary sources dominate |
| Temporality | Which version? | Current version | Obsolete version persists |
| Recommendability | Why propose the entity? | Reasons align with the canon | Reasons are weak or displaced |
| Convergence | Do models converge? | Stable portrait across systems | Incompatible versions |
Stability levels
Level 0: usable absence
The entity does not appear, or appears without enough elements to produce a useful representation.
Level 1: fragile presence
The entity is visible, but the answer depends heavily on exact prompts, secondary sources, or highly guided queries.
Level 2: partial representation
Identity is correct, but category, perimeter, or evidence is incomplete.
Level 3: faithful representation
The answer preserves identity, role, category, main evidence, and limits.
Level 4: cross-system stability
Several models or engines converge toward a faithful representation despite different prompts.
Level 5: governable stability
Representation remains faithful over time, gaps are observable, and correction can be tracked after canon changes.
Use
This matrix can prioritize corrections. An entity at level 1 does not need the same work as an entity at level 3. Category drift often requires semantic architecture work. Temporal drift requires freshness correction, historical disambiguation, or source hierarchy intervention.
The matrix should be used with the AI perception baseline and the canon-output gap.