Scientific validation reports¶

This section collects per-component scientific-validation reports. Each report is produced by the scientific-validation-orchestrator workflow and is written by six reviewer lenses (construct-validity, mathematical-fidelity, empirical-realism, literature-grounding, experimental-methodology, robustness-boundaries) plus an orchestrator synthesis.

Claim cards and scientific coverage¶

ClaimCardV1 is the implemented, immutable evidence record for one scientific claim. It is frozen, extra-forbidden, and RFC 8785 self-hashed. A card names its selector, component, version, construct, equations, citations, implementation symbols, supported environment, fixtures, experiment command, thresholds, raw-artifact hashes, results, and robustness envelope. It also records the seed, positive sample count, retained figure and artifact hashes, limitations, regression outcomes, and any deferred architectural item with rationale, owner, and revisit trigger.

The methodology fields are enforced rather than merely described. Multiple comparisons use Holm correction with alpha 0.05; every recorded confidence interval is 95%; retained figures and artifacts are required; and all six lenses appear in their fixed order with PASS and zero HIGH findings. The report author and all validation gates must pass, and a recorded regression must not reproduce. A card with malformed, incomplete, nonfinite, forged, or validator-bypassed evidence is rejected when coverage is checked.

validate_scientific_coverage does not trust cards in isolation. Before it reads card selectors, it calls the existing Layer 11 validator for every production source’s status history. For a scientific source it additionally requires the exact handoff, result, latest status, and chapter manifest hashes to agree, then validates that handoff/result pair against its manifest. A non-scientific source carrying ceremonial scientific evidence is rejected. This context validation happens before card canonicalization and selector coverage.

Coverage is exhaustive over EvidenceClass.SCIENTIFIC claim sources. Every advertised scientific selector needs exactly one card. If a scientific source has no advertised selector, its module name is the required module-level card selector, so an empty selector inventory cannot create a vacuous pass. Duplicate scientific selectors across sources are invalid. Missing cards, a card with FAIL, and a DEFERRED card are all reportable failures; deferred architecture is documented but cannot satisfy an advertised release claim.

ScientificCoverageReportV1 records the bound claim-index hash, expected, passed, failed, and deferred counts, plus sorted missing selectors, failed claim IDs, and deferred selectors. The counts must partition the expected selector count. Its status is PASS only when all three identifier lists are empty; otherwise it is FAIL. The report routes and accounts for already-collected evidence. It does not rerun experiments, repair a scientific result, or convert a deferred architecture item into a passing claim.

Emitters