Published-archive API companion

Base 102b9e3ee35921f2289db6e87ce3182cb06946b3, branch fix/published-api-baseline, reused owned no-std-dependencies/nox checkout. The previous wrapper context is archived in audit/2026-10-17-published-api-baseline/task-honeycrisp-backend.json. Hemera 1c385e54, Strata 56aedb2d and Lens 8b26cef6 symlink targets/revisions remain unchanged. Builds use an external task cache; no production inputs change.

Authority and scope

The required exact locked v0.2.0 tag gate remains unchanged and red. Its existing research explicitly withholds policy-change authority. This task creates an executable companion diagnostic and a concrete policy proposal for owner review; no successful companion result can replace, waive or mark that gate green. No dependency/version edits, source substitutions, tags or publication actions. Scope is Nu/Rust tooling, fixture inputs, this proposal and audit evidence.

Evidence design

1. Fetch the official cyber-nox 0.2.0 index record and crate archive into the task cache; verify archive SHA256 7c807c1b53e66cea79e8e130ca938450029af3f096984a2839fee6443a171b3f. Record exact acquisition argv/status and the archive's .cargo_vcs_info.json. Compare owning Rust bodies with the peeled immutable v0.2.0 source, explaining Cargo normalization and the published lock separately. Preserve the tag and original locked-tag failure; do not regenerate its lock.
2. Produce baseline and current rustdoc JSON in distinct source/target roots with the same pinned compiler and flags. The baseline is the unmodified published archive. The current side is an exact normalized package of the committed owning source, retaining registry dependency identity; verify its Rust bodies equal the owner. The current occupied version is diagnostic only. Record all package manifests, locks, source/artifact hashes and actual commands.
3. Build default (empty crate defaults) and explicit-std API surfaces separately using cargo rustdoc --locked --offline and explicit JSON output. Use installed pinned rustup 1.95.0 with the documented rustdoc JSON mechanism, and installed cargo-semver-checks 0.51.0 with explicit --baseline-rustdoc and --current-rustdoc paths. Predeclare patch-compatible comparison; report the actual verdict, counts and omissions. Never allow automatic shared-target baseline discovery. Any tool/format failure stays a failure.
4. Independently inspect resolved feature graphs: old default Hemera/std, repaired default without it, and explicit std forwarding in both surfaces. Name transitive std-only APIs and actual dependency versions. The public API tool does not adjudicate this compatibility change; neither does a default API pass determine Nox's release version. Optional backends remain outside this companion's acceptance claim, with their existing delivery debts named.
5. Write a portable verifier for acquisition/source/artifact/raw-stream hashes, exact commands/tool identity and separate roots. Keep bulky rebuildable artifacts in the external cache; retain the minimal reproducible source, artifact identity and raw verdict evidence under the owning audit.

Proposed policy, not adopted

For a future owner decision, distinguish source-tag reproducibility from the public compatibility of bytes actually distributed by crates.io. A published archive comparison could become an additional required compatibility gate when its checksum, VCS correspondence, dependency domain and feature surfaces are verified. The exact locked-tag failure remains a separate unresolved release gate until the owner explicitly decides its policy; no retrospective repair of the tag is proposed. State this separation in the machine-readable outcome.

Acceptance for this task: genuine old/current artifacts, executable comparison, independent feature-inheritance findings, preserved historical red and a reviewable policy proposal. Root reviews this plan before tool implementation; final CODE review and source-bound evidence precede an ordinary tooling commit.