This is useful as a fixture PR. One alignment point before these get treated as conformance vectors for #2787:

As far as I can tell, these vectors target the proposed revised envelope shape: planner_declared, issuer_asserted, payload_derived, explicit args commitment kinds, JCS canonicalization, and float rejection.

The SEP text in #2787 still defines the flatter v1 shape: top-level iss / sub / intent, toolCalls[*].args as a string, sorted-key/no-whitespace canonicalization, and optional ack.

So I would mark this vector set as provisional until the SEP text lands the matching schema and canonicalization rules, or name it clearly as proposed-shape vectors rather than conformance for the current draft.

Otherwise a second implementation could correctly implement the SEP text as written and fail these fixtures.

Fair catch and accurate read. Renaming the set as proposed-shape rather than conformance for the current draft. README and PR body updated to call out the divergence between the SEP text as written (flat iss / sub / intent, toolCalls[*].args as string, sorted-key canonicalisation) and the revised shape these vectors target (the four-proposal bundle from #issuecomment-4544033186).

If useful, I can add a sibling v1-current vector set against the SEP text as written so conformance works for both surfaces until the SEP shape settles. Say the word.

v0.40.0 shipped on vaaraio/vaara@1ac1d75; vaaraio/vaara#154 lands the streamable-HTTP proxy that emits the v2 envelope this PR targets.

The v2 envelope changes are a clear improvement. Self-discriminating args without the kind enum is cleaner, and the canonical_signing_input.bin convention for tampered cases is exactly the right primitive for independent implementers.

One fixture-determinism point on case 10: I would keep an explicit verify_at_epoch or equivalent verifier time in expected.json.

Using wall clock in _check_independent.py makes the policy case depend on when the walker runs. It happens to reject now because the pinned iat is already in the past, but test vectors are more useful if a second implementation can replay the exact verifier inputs without ambient time. Concretely: a second implementation that runs the walker in 2027 should get identical pass/fail to one that runs it tomorrow. With wall clock, that's not guaranteed for case 10.

So the case can still be "TTL expired past skew," but the expiry decision should be driven by fixture data rather than datetime.now(). That keeps the vector set boring and reproducible, which is exactly what we want for independent implementations.

Pushed 2147293: case 10's expected.json now carries verify_at_epoch (1577923200, i.e. 2020-01-02T00:00:00Z, comfortably past iat+exp+skew with the 2020-01-01 iat). The walker reads that field instead of datetime.now(), so case 10 is replay-deterministic. Walker still reports PASS across all 13 cases. README updated to document the convention for time-sensitive policy fixtures.

Vaara v0.42.0 ships the post-execution complement to these vectors: the execution receipt layer (vaara.attestation.receipt), which takes the attestation wire bytes as backLink input and records what actually executed. Reference implementation at vaaraio/vaara@4608c36, conformance vectors at tests/vectors/execution_receipt_v0/.