Skip to content

PR12.2 — Governed External Evidence → Claim Interpretation Proposal v1

Purpose

PR12.2 introduces the first generic interpretation boundary after PR12.1.

It allows one exact external EvidenceRecord already retained in an EpistemicRecordSet to be proposed as relevant to one exact CapabilityConceptRef and one bounded claim statement/scope.

It does not create a CapabilityClaim and does not decide what the evidence means for that claim.

PR12.1 neutral external EvidenceRecord
        |
        v
PR11.3 retained epistemic snapshot
        |
        v
PR12.2 interpretation proposal
        |
        X no CapabilityClaim
        X no EvidenceAssessment
        X no ClaimEvaluation
        X no state
        X no progression authority

The central invariant is:

EXTERNAL EVIDENCE INTERPRETATION PROPOSAL
=
ONE EXACT RETAINED EvidenceRecord
+
ONE EXACT CapabilityConceptRef
+
ONE PROPOSED claim statement/scope
+
ONE DECLARED proposer
+
ONE FROZEN PR12.2 POLICY

Authority boundary

PROPOSAL != CLAIM TRUTH
PROPOSAL != EvidenceBearing.SUPPORTS
PROPOSAL != EvidenceReliability
PROPOSAL != EVALUATION
PROPOSAL != CAPABILITY
PROPOSAL != STATE
PROPOSAL != READINESS
PROPOSAL != MASTERY
PROPOSAL != PERMISSION

PR12.2 deliberately exposes no fields for EvidenceBearing, EvidenceReliability, evaluation conclusion, coverage/conflict status, claim id, state id, score, mastery, readiness, or permission.

PR12.3 is reserved for an explicit future review/acceptance boundary.

Exact retained-evidence binding

The caller supplies an immutable EpistemicRecordSet and an exact EvidenceId.

The proposal function finds the selected record inside that snapshot and derives:

subject_ref
evidence_id
evidence_sha256

The subject is therefore not caller-selectable.

evidence_sha256 reuses the existing PR12.1 canonical external_observation_materialized_evidence_sha256_v1 digest. A later record with the same EvidenceId but different bytes cannot satisfy verification.

The proposal does not bind the hash of the complete epistemic snapshot.

candidate for evidence A
+
unrelated later retained evidence B
->
candidate A remains valid

mutated/replaced evidence A
->
candidate A rejects

This mirrors the exact-item rather than global-container staleness discipline of PR12.0 and PR12.1.

PR12.1 external-evidence shape

PR12.2 accepts only a retained record that structurally preserves the frozen PR12.1 external-observation materialization shape:

  • EvidenceId = external_observation:<sha256>;
  • exactly one EXTERNAL_RECORD provenance source equal to that exact identity;
  • exact external_observation scope tag;
  • outcome = None;
  • exactly one external_observation_materialize provenance step;
  • materialization mechanism equals the frozen PR12.1 policy;
  • materialization step time equals EvidenceRecord.recorded_at.

This is a structural downstream admission check. It does not invent a new signature/PKI claim and does not reinterpret the evidence.

Exact concept revision

The caller supplies a CapabilityCatalog and exact CapabilityConceptRef.

PR12.2 requires the capability lineage to be present exactly at that revision.

requested concept@1
+
catalog contains concept@2
->
REJECT

There is no silent latest-revision substitution.

Proposed claim semantics

The proposal may contain:

claim_statement
ClaimScope(description, tags)

These are proposed interpretation semantics only.

PR12.2 deliberately does not allocate a CapabilityClaimId. It therefore cannot silently turn its proposed wording into an accepted person-scoped claim.

Proposer provenance

PR12.2 v1 permits only two declared proposer kinds:

HUMAN
MODEL

The proposer ref is descriptive provenance.

DECLARED HUMAN != AUTHENTICATED HUMAN IDENTITY
DECLARED MODEL != TRUSTED MODEL OUTPUT
MODEL PROPOSAL != CLAIM AUTHORITY

Time

candidate.proposed_at >= selected EvidenceRecord.recorded_at

The proposal cannot predate the evidence it interprets.

Deterministic digest

Candidate serialization is strict schema v1 and the candidate digest is domain-separated by:

capability_lab/external_evidence_claim_interpretation_candidate@1\0

The digest commits the exact:

  • proposal id;
  • frozen policy ref;
  • evidence identity and digest;
  • derived subject;
  • exact concept revision;
  • claim statement;
  • claim scope;
  • proposer;
  • proposal time;
  • rationale.

Changing any committed interpretation semantic changes the digest.

Strict serialization

The schema requires exact fields and rejects:

  • missing fields;
  • unknown fields;
  • duplicate JSON keys;
  • non-finite JSON constants;
  • malformed timestamps;
  • invalid enum/type reconstruction.

Serialization is deterministic canonical JSON with sorted keys and compact separators.

Position in the governed loop

After PR12.2:

external / HDE activity
        |
        v
PR12.0 admitted observation
        |
        v
PR12.1 human-reviewed neutral evidence
        |
        v
PR11.3 immutable epistemic retention
        |
        v
PR12.2 exact evidence -> claim interpretation proposal
        |
        X no accepted interpretation yet
        X no ClaimEvaluation yet
        X no state update yet

The next intended boundary is PR12.3: explicit review of an exact PR12.2 interpretation candidate.