VCP / VSO · ANNOUNCEMENT · 26 October 2026

VSO publishes a fixed checkpoint for VCP RTA verifier review

VSO has published a fixed VCP RTA review target and now checked the published instructions end to end in its own CI, from a fresh public clone through final report assessment. The fixed 809 regression and 15 separate historical tests passed per environment. The 40-event evidence verdict remains INDETERMINATE. The supplemental format remains unadopted, and independent reproduction has not been established.

Implementation extension — unadopted; independent reproduction not established

The supplemental format is an implementation extension, not an adopted VCP/VAP format. This checkpoint is not a new release, certification or independent implementation. Neither its publication nor successful first-party CI establishes independent reproduction, full conformance or production suitability.

26 October update: selected original CI records and review intake

Six selected original JSON files from the completed first-party CI (three per Python environment) are now stored in Git: the procedure-completion record, final assessment and original artifact manifest. They are byte-for-byte copies, not another execution. Full JUnit XML, the detailed 40-event replay and shell logs are not included. The selected copies no longer depend solely on the Actions artifact’s 30-day retention, but Git availability is not immutable or guaranteed permanent storage. Hashes of omitted files cannot recover their bytes.

Selected original CI records and their limits

The review intake distinguishes the RTA checkpoint (target cc695fe, guide a839d7a) from the VAP OS-isolated mock-gate kit (kit b1af0dc, target c8ccf97). Acceptance of one assignment does not authorize substituting the other. Willingness to run, receipt of a report, technical agreement, independence and publication permission are separate states. Preserve failures, incomplete runs, environment differences and assistance. Permission for results, raw logs, name, affiliation and specific attribution must be checked separately.

GitHub issues are public, including the submitting account and attachments. For a private technical report, first contact standards@veritaschain.org to agree a transfer method, without attaching secrets. A template, closed issue or successful first-party CI does not establish independent reproduction. The experimental extensions remain unadopted; independent reproduction remains unestablished.

Review targets and submission guidance

26 October update: published instructions checked in first-party CI

The additional PR #16 workflow completed in both Python 3.11 and 3.13.5 CI jobs on Ubuntu 24.04. It executes the published shell instructions without substitution: a fresh public clone, a new virtual environment, dependency installation, comparison of 29 input hashes, the fixed 809 regression and 15 historical cases, the 40-event replay, and final assessment of the saved reports.

The full procedure must complete successfully, while the evidence replay must return INDETERMINATE with exit code 2. Missing reports, setup failures or a zero shell exit without the required checks do not count as success. The launcher's 13 synthetic guard tests are separate from the fixed 809 + 15 cases. Repeating those cases in another workflow does not add new unique coverage.

The automation is maintained by the same project. These are two first-party environment runs, not two independent evaluators. Independent third-party reproduction remains unestablished; the supplemental format remains unadopted. The executed target stays cc695fe, the pinned guide stays a839d7a, and the automation is a separate revision. The workflow scope and retained-report description distinguish those revisions and the initial clone revision.

The original full Actions artifacts have 30-day retention. Selected original JSON copies are now stored in Git as described in the selected-record section; this is not a full archive or a fully locked environment. This announcement reports published records, not another verifier run, independent review, certification or full evidence acceptance.

What changed in the verifier

The merged work separates VALID, INVALID and INDETERMINATE checks. Missing material cannot erase a separately detectable mismatch. Follow-up fixes validate Ed25519 keys and signatures, Merkle proof structure and timestamp policy inputs, bound inputs, and prevent a report from overwriting source evidence, keys or earlier reports. The regression runner also rejects unregistered test modules and inconsistent or malformed JUnit reports.

Ordinary verification and supplemental verification are separate

The ordinary CLI verifies signatures against a caller-supplied trusted key; it does not authenticate batch metadata or verify external anchors. A separate opt-in command checks signed supplemental records and offline RFC 3161 timestamps using caller-selected keys, CA certificates and TSA policy. A later supplemental timestamp does not establish the original logging or anchor time. Synthetic TSA checks are not live TSA interoperability results.

What the historical 40-event replay actually establishes

The corpus contains 20 SIG and 20 EXE events. With an explicitly selected reference key, hashes, signatures and legacy Merkle comparisons match the reference material. That uses the old implementation’s encoding and odd-node duplication, not JCS or strict RFC 6962 conformance. The first PrevHash boundary, authenticated batch metadata and original external anchor remain unresolved. Key-owner identity is not established. The expected overall result is INDETERMINATE, with replay exit code 2.

Recorded first-party checks — not evidence acceptance

Reference results from the earlier PR #14 CI, not the added recipe workflow
Review target	cc695fe · 13 regression modules
Regression checks	809 cases per recorded Python environment
Separate historical checks	15 cases per recorded Python environment
Recorded environment	Ubuntu 24.04; Python 3.11.17 / 3.13.5; OpenSSL 3.0.13
CI outcomes	Zero failures, errors or skips in the recorded test runs
Historical evidence verdict	INDETERMINATE — unresolved checks remain

A fixed target and reproducible reporting

The checkout target is cc695fe. The recorded PR #14 CI executed the temporary merge 0314e01, which has the same Git tree as the target; the commit IDs are not interchangeable. The later guide commit a839d7a contains instructions, reference JSON with 29 input hashes, and an unfilled reviewer report. Record actual executor, relationships, assistance, environment differences, failures and incomplete runs. Permission to execute is separate from permission to publish results, names or affiliations.

Limits and preserved records

Full CI artifacts have 30-day retention. Six selected JSON originals are now in Git; the remaining artifact bytes are not archived there, and a manifest alone cannot restore them. Dependencies and the environment are not completely locked. This website update reports the published records and CI status; it is not another execution of the verifier. Normative specifications, existing implementations, signed evidence, fixed distributions and K3 are unchanged. The earlier VAP kit (b1af0dc, target c8ccf97) remains separate.