Epic: Provenance chain precision — frame identity, range grammar, attestation scope

[macanderson]

Purpose

Parent tracking issue for a family of spec-precision gaps in the provenance/attestation chain surfaced by implementation-drift review: what a frame's stable identity means (#186), what bytes a provenance digest actually covers (#183), and what an attestation's signature actually authenticates (#185). All three sit on the same load-bearing path — a frame's (provider id, frame id, content_digest) triple, the range grammar F5 digests, and the attestation preimage built on top of both — and each is a case where the reference implementation has silently picked an answer the spec never wrote down. Grouped together because a maintainer decision on any one interacts with the others: the attestation preimage in #185 already names the identity choice from #186, and the range grammar in #183 feeds the digest that #185's commitment covers.

Sub-issues

• Frame identity names a provider id the spec never disambiguates, and three surfaces disagree #186 — frame identity's provider id is ambiguous across three surfaces
• An attestation's issued_at and attester_id are outside the signature, and the spec never says so #185 — an attestation's issued_at/attester_id are outside the signed preimage, undocumented
• F5 digests bytes addressed by a range grammar the spec never defines #183 — F5 digests a range the spec never defines a grammar for

Acceptance checklist

• Every sub-issue under this epic is closed or explicitly deferred with a reason
• SPEC.md states, for each field, what it means, what a verifier may assume, and what a conformance check enforces

Risks + mitigations

Each sub-issue needs a maintainer decision before a fix can land — a normative SPEC.md addition, and in one case (#185) a possible new major family (contextgraph/2). The risk is scope creep: land the documented decision plus minimal enforcement per issue, not a broader redesign. No urgent production breakage is currently known to depend on these; treat as pre-1.0-freeze hardening.

Rollback

N/A — this epic tracks work, it does not itself change code.