Epic: Conformance suite and README claims that outrun SPEC.md

[macanderson]

Purpose

Parent tracking issue for cases where the conformance suite or README asserts something SPEC.md does not actually say — found by the same implementation-drift review as the sibling epic, but a different failure shape: here a test or marketing document claims more (or something different) than the normative text supports, rather than the normative text being silent. #184's as_of check enforces a rule with no requirement anchor; #187's README states a consent guarantee as unconditional when the spec only enforces it on one of two transports.

Sub-issues

• The as_of conformance check enforces a rule the spec never states, and covers half of it #184 — the as_of conformance check enforces a rule SPEC.md never states, and only covers half of it
• The consent guarantee README states unconditionally is enforced on one of two transports #187 — the consent guarantee README states unconditionally is enforced on one of two transports

Acceptance checklist

• Every sub-issue under this epic is closed or explicitly deferred with a reason
• Neither the conformance suite nor README asserts a guarantee stronger than what SPEC.md's normative text (MUST/SHOULD) actually requires

Risks + mitigations

Both sub-issues need a maintainer decision on semantics before the suite or docs can be corrected (what as_of should bind; how strong C3 should be, and whether stdio gets a sandbox story). Risk is documenting a guarantee as weaker than desired instead of building the missing enforcement — call this out explicitly in each PR so it reads as a stated scope decision, not a quiet downgrade.

Rollback

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