docs/test(evidence): propose chain-of-custody and retention boundaries

[kaveh8866]

Scope and maturity

DRAFT / review-only / proposal and synthetic tests. Extends RFC #196 and Phase 4 Draft PR #198. Base main: 89cfdcb10ab407f5c1ef84b8cda05894b5dd4a2a.

Changes

• Docs/evidence/OBSERVATION_CUSTODY_PHASE5_DRAFT.md: bounded, source-backed consumer inventory for ObservationEvidence and separately implemented research-bundle/partner-exchange paths. Explicit known/unknown outcomes, proposed chain-of-custody stages, trusted-key and domain/scope checks, and conceptual access/retention/deletion matrices. Does not modify existing normative security policy.
• tests/unit/observation-custody-trust-prototype.test.ts: four offline synthetic cases using standard Node Ed25519 and SemantIQ's existing canonical JSON helper; tests independently pinned key, content modification, unknown/revoked key, and why a self-consistent hash is not source authentication. No production signing integration.

Boundaries

The sample proves only a prospective cryptographic design property with ephemeral test-only keys; no actual key management, trusted registry, timestamp authority, live data provenance, retention, universal redaction guarantee, third-party replication or research truth is established.

Responsible disclosure: Per SECURITY.md, any implementation-specific vulnerability details discovered during audit are not published in this public PR and must follow confidential security triage. This PR is deliberately limited to general positive architectural requirements.

Governance and checks

• Fast contribution: standalone document/test, no runtime change, no schema or migration edit, no release.
• Exact-head CI/format/lint/typecheck and CODEOWNER review, pending.
• RFC RFC(DRAFT): versioned ObservationEvidence digest and safe legacy verification #196 approved for Core implementation (not approved).
• Consumer and storage inventory independently completed; external operator flows remain unknown.
• Retention/backup/deletion periods and roles explicitly approved, not assumed.
• Independent public key trust anchor and key custody specification approved, not implemented.
• No main merge, no historical rehash, no external repo code or real secrets used.

The chain proposed here is separate from existing ObservationEvidence digest issue #192; there is no assertion of an actual deployed end-to-end custody pipeline.