VALUE-02C — Verify retained multi-model decision evidence

[daniele21]

Parent: #128

Goal

Add a bounded verifier for the retained VALUE-02 artifact set. The verifier must consume canonical Performance Lab evidence and campaign read models rather than recreating ranking/comparability policy.

Owner

tests/real_runtime/verify_value02_evidence.py plus focused deterministic tests.

Acceptance

• at least two completed immutable candidate Runs exist;
• candidates share the intended target/device, use case and benchmark protocol;
• campaign decision policy is exactly strict-quality-dominance@1.0.0;
• capability compatibility is established before any recommendation claim;
• recommendation points to a retained candidate Run, or no-rank is explicitly justified by canonical results;
• Quality / Performance / Resources remain separate and unavailable evidence is not converted to zero;
• at least one exact same-case comparison has 2+ compatible retained candidates;
• verifier emits a bounded public-safe PASS/FAIL manifest without raw prompts, model outputs, credentials or private paths.

Evidence boundary

Hosted CI validates verifier logic with deterministic evidence. Final VALUE-02 acceptance remains REAL_ENVIRONMENT.

Integration gate

Development may proceed on an isolated branch, but do not merge to dev until VALUE-01D #120 passes.

[daniele21]

Software lane READY on draft PR #137. Exact head 75127587e0e814ef5c9597d8317c1197566a4235 has PASS on Repository Health, Repository Validation, Browser Acceptance and Built Product. Keep this issue open/not integrated until VALUE-01D #120 passes; then refresh against current dev before integration.

[daniele21]

VALUE-02C software is implemented and retained in canonical integration PR #141. The former lane PR #137 is closed as superseded, not merged; its commit remains provenance. #131 stays open until #141 is integrated after VALUE-01D #120 PASS.