5.6 KiB
Final-04 complete gate admission review
2026-09-24. Publication 206f3b969482241e01e2dfc6be955caf31747b89:harness/prerelease37-windows-gates-20260924/final04-complete; source 715036e132677dd2bc2fb37ed316d93dc9987aec.
Admission result
No material discrepancy found in the completed source/test/lint/build evidence or declared product selection. The eight-step final-04 summary records passed:true, failure:null and unchanged inputs. Every raw step result independently agrees: rustc version, Cargo version, workspace no-run compilation, literal workspace tests, strict workspace all-target Clippy, individual desktop build, individual server build and attachment_snapshot example build. The separate literal combined build also records exit0, unchanged inputs and no instrumentation failure.
This admits these build/gate receipts and their declared artifact identities. It does not admit package transport/runtime, notices, GUI workflows or server persistence; executable bytes are intentionally absent from this archive.
Exact source comparison and reused tests
Independently compared all 3,709 captured source paths against git ls-tree -r of the exact715036e commit in local .local/peerunion: path sets and Git blob IDs match. Read every immutable Git blob with git cat-file --batch and compared raw lengths and SHA-256 against the remote input manifest: all3,709 match exactly, zero EOL conversions and zero unexplained differences. This closes the prior review's unperformed full Git-blob comparison. The comparison used immutable Git objects; the locally materialized working-tree subset is not claimed as a full checkout.
All per-step before/after maps, final after map and combined-build maps are byte-identical to the top-level before map. This establishes recorded inputs against the identified source, not cryptographic proof of what a compiler executed.
407 prior test/phase members are byte-identical to the already reviewed final04-tests checkpoint at28b8b5a. Reused FINAL04-TESTS-REVIEW.md and its independent reconciliation: 61 native harnesses,2,208 names,2,200 passed/eight ignored/zero failed; all29 required names individually passed. No unnecessary test rerun or weakened success gate was used.
New lint/build evidence and selected bytes
Raw strict lint command is cargo clippy --locked --offline --workspace --all-targets -- -D warnings, exit0. Stderr reports completed dev profile and no warning/error diagnostics.
Individual commands are cargo build -p oa-desktop --bin arcadia-studio --locked --offline --message-format=json-render-diagnostics and matching oa-server/arcadia-server. Parsed their raw Cargo JSON independently: each has exactly one matching executable event; the complete event equals artifact-relocation.json's stored event, its executable path matches the original receipt, and original/preserved byte lengths/hashes agree with final summary outputs. Emitted profiles are non-test, opt_level0, debuginfo0, debug_assertions:true, overflow_checks:true.
| Declared preserved package input | Bytes | SHA-256 |
|---|---|---|
| arcadia-studio.exe | 119,396,352 | 2c03bd6cc3f1110c219e4f57ab94874a121590f83779239a93289b7170664399 |
| arcadia-server.exe | 40,228,352 | 6cebacc117e46fdab023c6402a5cebd6929fb55993ae8ff6b89cd1c6c1170237 |
| libunwind.dll | 190,464 | aa282fa47bb2c4e210cea5e470cdbdca58568f0be0bc73e404d86ab9cd95ef6a |
The attachment_snapshot example also matches its raw Cargo event and preservation receipt:19,188,224 bytes, SHA-256 6a1f438ab8f122251846f46cf1c5440ec7b0362b34fd61b97286d470bad11b74. It is a separately identified helper, not a Studio/server package identity.
Individual versus combined build boundary
The explicit supplemental command is cargo build --locked --offline -p oa-desktop -p oa-server. Its stdout is not Cargo JSON, so its success receipt does not replace the individual raw-event artifact binding. The separately captured comparison records different combined Studio/server hashes (211b03d9… / 8995332f…). The manifest and independent review explicitly retain the preserved individual outputs as package inputs; no silent substitution or byte-equivalence claim occurs.
Raw individual dependency events substantiate distinct feature graphs: GUI serde_json1.0.151 has default/std; server adds raw_value. Server Tokio1.53.1 additionally enables macros,signal,signal-hook-registry,tokio-macros compared with GUI. This supports feature unification as a plausible cause of differing combined output, but does not prove the combined graph, sole cause of differences or semantic equivalence. Release evidence should state individual commands built the distributed bytes and the combined command passed separately. Native checks must bind the selected preserved hashes above, not later shared-target files.
Transport and limitations
MCP manifest blob: 5d246cc2ca69c6aaaffaa5dc70ec1be2bdae4262. After reading manifest/README, downloaded with a6,700,000-byte cap. ZIP size 6,588,421, SHA-256 7cead0b6aa36a6af8fc9f8987186bfe0decd8fa2ebab094fc45387ae9daea632. Verified all462 member names/counts, byte lengths, SHA-256 and CRC32 pins. Local inspector/log: inspect_final04_complete.py, final04-complete-inspection.log.
No archived code, Cargo command or executable was run; no source/Gitea/release write occurred. The independent inspector reads archive data and local immutable Git objects only. Original failures/interruption remain preserved in earlier publications. Raw clock-jump timestamps are retained; no duration is inferred by subtraction. Reported package hash, runtime trials and initial server-fixture mismatch are outside this archive/review and require separate admission before release acceptance.