Use deterministic, fail-closed tests for governance adjudication

When changing governance adjudication, validate behavior with focused pytest coverage using synthetic fixtures and monkeypatched CLI inputs. Keep schemas and runtime constants synchronized, test both successful validation and failures that close the canonicalization gate, and cover temporal/status coherence, claim-to-source grounding, graph-edge revisions, malformed inputs, and fragment/full-envelope distinctions. Reverify every frozen input pin independently with real file hashes and pinned workflow provenance. Prefer fail-first regressions for discovered gaps, then run the targeted adjudication suite and report exact results. Preserve scientific boundaries: do not inspect or adjudicate excluded source records, outcomes, transition dates, or other prohibited artifacts.