Interpretive debt: accumulation and extinction dynamics (complete operational framework)
Interpretive debt is the future cost induced by an ungoverned interpretation today. It does not necessarily manifest as a spectacular error. It manifests as rigidity: the interpretation becomes difficult to shift even after correction, because it has stabilized through repetition, aggregation, neighborhood, and inertia.
This framework describes how debt forms, how it accumulates, how it stabilizes, and how to extinguish it without creating new instability.
Operational definition
Interpretive debt: durable gap between the canon and the dominant interpretation produced by AI systems, whose correction requires endogenous + exogenous intervention, response condition governance, and version discipline.
Why debt accumulates
- Ungoverned inference: the model fills beyond the perimeter.
- Compression: summaries and simplifications that erase nuances.
- Contaminated neighborhood: dominant co-occurrences that redefine the entity.
- Remanence: persistence of a former state after correction.
- Trail: slow and uneven propagation of updates.
Debt lifecycle (DI-1 to DI-5)
DI-1: formation
A plausible inference fills a gap (weak canon, ambiguity, ungoverned conflict).
DI-2: amplification
The same interpretation is repeated, cited, aggregated, compared.
DI-3: stabilization
It becomes the most probable output. Occasional correction no longer has a global effect.
DI-4: rigidification
Correction cost increases. The system resists displacement (inertia).
DI-5: extinction (or regression)
Either the debt is resolved through governed correction, or it reappears (remanence).
Observable symptoms
- stable or growing canon-output gap
- inconsistent responses depending on formulation
- recurring identity confusions
- corrections that “hold” for 2 days then regress
- increasing cost of clarification and re-testing.
Minimum measurement
- Canon-output gap (level and trend)
- Compliance drift (increase over time)
- Remanence index (reappearance post-correction)
- Propagation delay (trail)
- Identity incidents (collisions, contaminations, capture).
Extinction playbook (DIX-1 to DIX-10)
DIX-1: diagnose the root cause
Collision, capture, weak canon, authority conflict, non-timestamped dynamic state.
DIX-2: strengthen the canon
Definitions, exclusions, relations, canon version.
DIX-3: govern inference
Q-Layer, response conditions, legitimate non-response.
DIX-4: require evidence
Interpretation trace and fidelity proof on critical attributes.
DIX-5: correct the exogenous
Dominant sources, aggregators, secondary profiles, comparison pages.
DIX-6: disciplined release
Version, changelog, post-release validation.
DIX-7: adversarial re-tests
Multi-formulation, multi-turn, trap queries.
DIX-8: LTS monitoring
Alert thresholds, cadence, propagation and remanence tracking.
DIX-9: multi-AI stabilization
Compare models to detect “partial” debt.
DIX-10: prevention
Continuous canonization, dynamic state governance, version discipline.
Expected artifacts
- Debt registry (cases, type, surface, severity).
- Prioritized extinction plan (endogenous + exogenous).
- Release and validation journal.
- Metrics dashboard (gap, drift, remanence, trail).
- Versioned test battery.
FAQ
What is the most frequent cause?
An implicit canon: AI fills, then that inference becomes dominant.
Why does debt return after correction?
Remanence + uncorrected neighborhood. Endogenous correction alone is not enough.
What is the best success signal?
A durable decrease in the canon-output gap, and a decrease in post-release remanence.
