# Delivery audit — 2026-09-11

## Status: public repository and four-target CI verified; application remains a draft

MoonRestrict was created from scratch in its own CYang-dep checkout. The participant explicitly approved creation/push of **CYang-dep/moonrestrict**. Public repository: https://github.com/CYang-dep/moonrestrict ; default branch: **main**; verified owner/login: **CYang-dep** (GitHub id 327796663); viewer permission: **ADMIN**. The unrelated shared-workspace repository was not modified.

## Executed checks and immutable CI baseline

Green four-target run: https://github.com/CYang-dep/moonrestrict/actions/runs/34614016856 at **d381fb606339a751460cc4d861ab29b1771c189a**. Each matrix job's check/test/build, real-process acceptance and clean generated-file steps completed successfully. Final documentation/tag revisions must also pass CI before release; use the public Actions history and release tag to resolve their exact SHA rather than interpreting this baseline as evidence for untested future commits.

| Evidence | Windows local | Ubuntu 24.04 CI baseline |
|---|---|---|
| `moon fmt --check` | PASS | PASS |
| Strict checks: wasm-gc, wasm, js, native | PASS | PASS |
| wasm-gc tests / build / CLI / composed workflow | PASS; 53 tests | PASS; 53 tests |
| wasm tests / build / CLI / composed workflow | PASS; 53 tests | PASS; 53 tests |
| JS tests / build / CLI / composed workflow | PASS; 53 tests | PASS; 53 tests |
| native tests / build / CLI / composed workflow | Runtime attempt failed: no C compiler | PASS; 53 tests; actual native execution |
| `moon info` and generated-file diff | PASS | PASS |
| Public owner / permissions / contribution mapping | GitHub API verified | Not inferred from local author names |

The same 53 distinct tests run on four backends; this is not 212 different tests. Exhaustive small domains include 1,364 concrete strings, rotations of all 256 four-base rings and 225 IUPAC observation/pattern pairs. Scripts reproduce the checks; raw local/per-milestone logs and downloaded CI evidence are outside Git. Tests demonstrate abstract string/geometry behavior, not biological validity.

The first run (34613352990) failed workflow validation before any jobs: `runner.temp` was not permitted in job-level `env`. The installer now establishes its isolated `MOON_HOME` inside the runner step and persists it through `GITHUB_ENV`. The failed run remains public; passing YAML parsing alone had not proved valid GitHub context placement.

## Account and repository isolation

Only repository-local `user.name=CYang-dep`, `user.email=327796663+CYang-dep@users.noreply.github.com`, `user.useConfigOnly=true` were set. Global Git identity and shared GitHub login were not changed. Read-only preflights outside an isolated context observed other active accounts and correctly refused to proceed.

For remote operations, the already stored CYang-dep authorization is obtained without printing it and scoped to temporary process-only `GH_TOKEN`. `gh api user`, masked `gh auth status` and the repository/remote guard are rechecked. Git uses an explicit gh credential helper with unrelated helpers disabled for that invocation. The environment override is restored in a `finally` block; tokens/cookies/passwords are never saved to project or evidence files.

After initial upload, GitHub's commit API mapped **both author and committer to CYang-dep for all 24 uploaded commits**, including d0ae0acd2fd725e27889bfb2fe7be93d55555bc9. Final publication rechecks the complete history including subsequent fixes/docs. Matching local name alone was not used as proof. Public visibility is independently observable through unauthenticated GitHub HTML; an anonymous API attempt hit a shared-IP rate limit and is not presented as successful API verification.

## History, licensing and materials

The [history audit](history-audit.md) contains **22 substantive commits** with full SHA, purpose, changed paths and successful verification evidence. Additional audit-only commits and the two-step CI repair are conservatively excluded from that count. Raw history is larger and remains intact. The 20-commit threshold is this skill's extra requirement, not an official competition minimum.

The audited implementation has 1,505 production/example MoonBit lines and 882 test MoonBit lines (descriptive counts, not contest minimums). README, executable examples, generated API, protocol/architecture/boundaries, MIT license, Apache dependency attribution and license copy, AI disclosure, SECURITY, CONTRIBUTING, changelog and four-target CI are included. No real-organism dataset or enzyme catalogue is bundled.

## Remaining participant and release gates

The one-page Markdown application stays outside Git. Its substantive content and verified public repository/CI facts are filled, but participant name/contact are still missing. The strict checker must reject those placeholders until the participant supplies them. No personal data is invented or committed. The final release audit is kept with that application outside Git, so it can record exact final HEAD/tag/CI/release state without a self-referential documentation commit.

Release procedure: re-run CI on the final documentation commit, publish an annotated `v0.1.0` tag and non-draft GitHub Release at that tested SHA, then verify tag/default-branch/CI/author alignment. A prepared changelog is not proof that a Release has already been published; inspect the public Release state and external final audit.

The scoped [duplicate review](duplicate-check.md) inspected MoonCakes metadata and adjacent GitHub interfaces; no directly equivalent mature workflow was found in the inspected scope. `moon search` was unavailable and general web searches yielded no usable results, so exhaustive ecosystem originality is not claimed. The fingerprint remains reserved against reuse in the shared registry.

**MoonCakes has NOT been published.** Package publication is a separate manual participant action. Final application, package publication, official submission and organizer acceptance must not be inferred from engineering/CI readiness.