research(interoception): bind exclusion decisions to realized evidence artifacts

Start by reviewing ExclusionDecisionReceipt, ExclusionCriterionDecision, the current qualified-target lineage, and the exact v0.1 head 1007949d5c60fd2d7dd650e8bb4521e2b2803c48. Design the typed artifact registry and its binding to realized evidence packages before deciding whether a schema-lineage bump is required. Use the seven suggested adversarial tests as acceptance criteria, including failed validation for missing, duplicate, mismatched, or removed artifacts and preservation of excluded or indeterminate evidence.

Problem

ExclusionDecisionReceipt currently requires each preregistered exclusion criterion to receive an evidence-bearing decision, including NotTriggered, and validates evidence_sha256 as a lowercase SHA-256 digest.

That is useful but incomplete: a syntactically valid digest is not yet mechanically bound to a known artifact in the realized study evidence package. A caller could provide an arbitrary well-formed 64-hex value and still satisfy the receipt's structural digest check.

This is an evidence-lineage integrity issue, not a native interoceptive model-semantics issue.

Proposed direction

Design a typed exclusion-evidence registry / artifact manifest before changing the current qualified-target lineage.

Candidate properties:

• every exclusion-evidence artifact has a stable artifact ID, media/type identity, SHA-256, and optional schema/version;
• every ExclusionCriterionDecision references an artifact ID plus expected digest rather than an unscoped digest alone;
• receipt validation requires the referenced artifact to exist exactly once in the registry and its digest to match;
• the registry has deterministic canonical serialization + its own digest;
• study/realized evidence packages bind the registry digest;
• NotTriggered decisions remain evidence-bearing and are subject to the same artifact-presence rule as Triggered / Indeterminate decisions;
• excluded and indeterminate evidence remains preserved rather than dropped;
• exclusion-definition identity remains distinct from exclusion-decision evidence identity;
• semantic hypothesis outcomes must not become exclusion evidence unless prospectively declared by the exclusion contract;
• missing artifact, duplicate artifact ID, digest mismatch, or decision reference to an unknown artifact fails closed.

External-verification boundary

As with qualification gate evidence, typed registry membership would establish provenance consistency, not authenticate the truth of an external artifact by itself. Any external evidence source still requires appropriate independent verification by the study/evidence harness.

Suggested adversarial tests

1. well-formed random digest not present in registry is rejected;
2. correct artifact ID with wrong digest is rejected;
3. duplicate artifact IDs are rejected;
4. decision for criterion A cannot silently reference criterion B's evidence when policy requires criterion-scoped evidence;
5. NotTriggered decision without its referenced artifact is rejected;
6. removing an exclusion artifact after blinded metric freeze changes the realized evidence-root identity / fails validation;
7. excluded/indeterminate runs retain their referenced evidence artifacts in final accounting.

Lineage / timing rule

Do not implement this by mutating the current Native Interoception v0.1 qualification target solely to keep hardening indefinitely. Current exact v0.1 head is 1007949d5c60fd2d7dd650e8bb4521e2b2803c48, with CI/Showroom currently infrastructure-blocked.

First let that exact head face its qualification gates. If this issue is judged qualification-blocking after review, implement it as an explicit new evidence-schema lineage with the appropriate exclusion/study schema bump; otherwise schedule it for the next evidence-hardening tranche.

No emotion, affect, sentience, or consciousness claim is implicated by this issue.