Research: compare Policy A assessment guards

[camerontjs-dot]

Classification

Draft Research Infrastructure / policy discrimination experiment. No maintained policy change, merge, release, tag, promotion, Authorization, or execution.

Exact real CAL baseline

Frozen verbatim from Decision Engine Draft PR #41 decisive artifact:

• source PR: #41
• workflow run: 34171780593
• artifact ID: 10035941867
• artifact ZIP digest: sha256:1ed05eb4353405ba6422b3fe410d82de9d5e125c63ff83c5a9d84fb050d208ec
• exact Contract C SHA: sha256:c599e97fd5b4da80ae558d5d57a351fa3b2d37081432013a9dbeaae65a80b5a3
• exact prior Contract D SHA: sha256:db47ebc844c14aa28bbc02524684b1ea7e388e1f1eea7ee8cbc7153af7548200
• proposition PIPELINE_SMOKE_001:child:1
• completed / assessed / supported
• all four generic Contract C assessment stages not_performed

The frozen bytes were revalidated under exact Contract C 1.0.0 authority and maintained Policy A reproduced the prior exact CLEAR and Contract D digest.

Result

BOUNDED DISCRIMINATION COMPLETE.

Science head: 03cbb9e17f47c6b6e32585b5febca11437a09c96

Push research run 34174242810: PASS

Job 101900378557: PASS

PR-event research run 34174250720: PASS

Normal repository CI run 34174250757: PASS

Artifact policy-a-assessment-counterfactual-34174242810, ID 10036703362, ZIP digest sha256:bc06c7fd9e3a1798e0c44b2ed5d59c93946dcbad74fab53fcb035bc1fec888a6.

Contract C vocabulary constraint

Contract C 1.0.0 exposes exactly these generic stage states:

• not_performed
• performed / unknown
• performed / adverse
• not_applicable
• failed

There is no generic affirmative performed / favorable value. Therefore a DE policy cannot require a positive generic stage pass under the current Contract C vocabulary without a future contract/producer semantic change.

Policies compared

Maintained Policy A

Current semantics remain assessment-invariant once result/proposition execution are completed/assessed and reported_verdict == supported.

Candidate 1: explicit-negative guard

HOLD if any assessment is:

• performed / adverse; or
• failed.

Allow not_performed, performed / unknown, and not_applicable.

Candidate 2: explicit-unresolved-or-negative guard

HOLD if any assessment is:

• performed / unknown;
• performed / adverse; or
• failed.

Allow not_performed and not_applicable.

These are research counterfactuals only. No Contract D policy identity or maintained behavior was changed.

Observed matrix

Across all four generic assessment slots independently:

• exact real CAL not_performed -> maintained CLEAR; both candidates CLEAR
• performed / adverse -> maintained CLEAR; both candidates HOLD
• failed -> maintained CLEAR; both candidates HOLD
• not_applicable -> maintained CLEAR; both candidates CLEAR
• performed / unknown -> maintained CLEAR; candidate 1 CLEAR; candidate 2 HOLD

Combination controls agreed:

• all unknown -> only candidate 2 HOLDs
• all adverse -> both candidates HOLD
• all failed -> both candidates HOLD
• all not_applicable -> both candidates CLEAR

The only single-slot difference between the two bounded guards is exactly performed / unknown for each of eligibility, semantic validity, aperture completeness, and temporal applicability.

Interpretation

Both candidate guards close the explicit adverse/failed Policy-A surface exposed in PR #46 without changing the exact current real-CAL supported behavior where all generic assessment stages are not_performed.

The remaining material policy choice is narrow:

Should a stage that actually ran and explicitly returned unknown block knowledge.add_verified_tag@1(scope=claim) candidacy?

This experiment does not answer that normative question. It establishes that the choice can be made independently of current real CAL interoperability, because both candidates preserve the currently produced all-not_performed supported shape.

Boundary

• semantics: POLICY_COUNTERFACTUAL
• world_causal_claim: false
• no maintained source change
• no maintained policy change authorized
• upstream CAL reachability of nonbaseline assessment states is not claimed

The next high-value DE pressure test is producer-semantic authority: determine whether a Contract-C-valid object can change producer.semantic_implementation_sha and/or producer policy identity while retaining a positive maintained Decision, and whether that is intended policy scope or an unbound authority seam.

Keep Draft.