[kamjcx]

Summary

• define resolved_activity_value/unit as the authoritative factor-denominator-aligned application quantity
• correct resolved_quantity_kg to mean kilograms only and keep it null for non-mass activity
• normalize gCO2e/kgCO2e/tCO2e preview numerators to kgCO2e
• fail closed on denominator mismatch, inconsistent mass kilograms (including tiny values and legacy missing dimensions), and non-finite recomputed totals
• align evidence-backed reference-flow mass to /g, /kg, or /t before approval and locking
• preserve historical locked snapshots and the PR-A production response allowlist

Scope

This is an independent stacked PR based on PR #23 at 2f6b52c9effda67a73862b9dca1d3d88487cd8a7. It does not change retrieval, ranking, aliases, qualification, factor values, candidate IDs, frozen answers, or approval decisions. It does not merge PR #23.

Verification

• exact head: 24cf05884cc29358ceb57eeece42ed4585eb466a
• 572 tests passed; 87.16% branch coverage
• focused independent adversarial QA: 118 passed; P0/P1/P2 = 0
• Harness: 13 passed; 43.35% branch coverage
• evaluator critical path: 6 passed; 84.18% branch coverage
• Ruff, mypy, compileall: PASS
• FactorBench V1/V2/V3: completed
• Frozen Unit Regression: 28/28
• Sealed Unit v2: 32/32
• Sealed Unit v4: 21/21
• Portfolio and Autonomous quality gates: PASS
• public delivery scan and package build/archive isolation: PASS
• exact-head remote CI: 8/8 successful, including independent Linux container build, health and scan gate

QA history

The first independent QA pass found one reference-flow alignment issue, an inconsistent-kg lock gap, and a finite-input overflow gap. The second pass found tiny-value tolerance, missing-dimension and legacy-fallback variants. These failures remain preserved in review history. The third exact-head re-review at 24cf0588 passed with no P0/P1/P2 or residual P3 findings.