Evidence contract
CA-A1 architecture record. Updated: 2026-10-05. AI never becomes an evidence authority — CA-REQ-084 boundary applies structurally.
EvidenceItem (CA-OBJ-016)
id,cycle_id,title,content_ref(blob/document — external stores allowed; canonical record holds digest)source— declared origin (academic / practitioner / advocacy / public submission / commissioned)contributor— person/org +contributor_class(expert / stakeholder / member of public / member / official)submission_channel(portal / email / postal / oral evidence session / member-requested)conflict_of_interestdeclarationsselection_rationale— why this item was included/commissioned (expert selection is governance — CA-REQ-042)status(submitted / screened / presented / contested / withdrawn)visibility(public / members-only / protected)member_requestedflag — members’ evidence requests are a representable class
EvidenceAssessment (CA-OBJ-017)
id,cycle_id,assessor(named human role, not anonymous),assessed_atcoverage_statement— what the register does/doesn’t coverbalance_judgement— the recorded human assessment that the evidence spread is adequate/contestedcontestations— objections raised, by whom, resolutionreview_path— appeal/review route for excluded or contested items
The assessment is a record of a human judgement — the system holds the artefact; it does not produce the judgement.
Rules
- Evidence balance and expert selection are governance functions recorded as artefacts; software must not silently curate (CA-REQ-043).
- Every EvidenceItem carries provenance end-to-end: submission → screening decision → presentation.
- AI-generated summaries/syntheses entering evidence flows carry an
ai_artefactflag and are structurally distinct from submitted testimony. - Public visibility is per-item declared — protected submissions (whistleblower-class, small-group-identifying) are representable.