5.3 KiB
Ordinary default omission: public consumer verification
Follow-up: the completed native workflow evidence supersedes the native-workflow gap recorded at this earlier checkpoint.
The ordinary source changes at 7bb366f66a8c6f8b6f66fc9b04c9e3b39f84b78e now pass the CLI, HTTP and desktop automated consumer gates, strict workspace Clippy, and a matching Windows build. A new HTTP regression test verifies the distinction between preserving authored source and producing compact canonical export. Native workflow acceptance and integration remain incomplete.
Executed candidate and results
All four runs captured HEAD df2e1cd65ca5e56f47669e04a9d4bcb426900fcd plus the new test in crates/oa-server/tests/http/port_anchors.rs. That file is the only captured input differing from the previous full-core run; production code is unchanged. These are execution-input pins, not a claim that HEAD alone includes the new test.
| Gate | Result | UTC interval, September 13, 2026 |
|---|---|---|
production-consumers-1 |
12 CLI and 51 HTTP tests passed; no failures or ignored tests | 11:51:14–11:56:49 |
production-clippy-1 |
Workspace, all targets, -D warnings: exit 0 |
11:57:30–12:02:34 |
production-desktop-1 |
894 passed, 0 failed, 6 ignored | 12:03:15–12:10:24 |
production-windows-build-1 |
Desktop, CLI and server binaries built; exit 0 | 12:10:46–12:13:19 |
Each run recorded 613 unchanged inputs, manifest SHA-256 6bf74c1f68431a034681b754617c39d387783f71354c9106649aec9480e4f29f and source ZIP SHA-256 e20f33dc363850a9289103a7be28ad613c3fffb2039061957a0186cac72d541b. The portable archive contains original inputs, raw streams, start/end receipts and the capture helper. It records executable hashes but does not distribute runnable binaries or establish release acceptance.
The new HTTP witness imports explicit port_anchors [] with deliberate spacing and a rationale comment, then moves one placement through an actual HTTP transaction. Its expected complete model is independently constructed by changing that coordinate, and its expected source bundle changes exactly one numeric token. It checks revision progression, authored bundle retrieval, canonical model round trip and absence of the default field in canonical export. Server restart must preserve the complete stored response and both export forms. The witness passed.
Existing desktop tests also passed graphical creation with one persisted UUID and the two-file Apply, graphical edit, exact Undo, Save As and reopen workflow. They are automated application checks; native interaction remains a separate gate.
The two HTTP stderr messages reporting Windows access denied come from deliberate failed atomic-replacement tests: failed_atomic_replacement_preserves_the_accepted_model and source_projects_failed_atomic_comment_write_preserves_both_representations. Both replace an isolated storage file with a directory, expect HTTP 503 and unchanged accepted state, then restore it. Both passed; these messages do not indicate a failed consumer run.
Six existing desktop probes remain ignored: the two accepted-release timing probes, the large source-symbol probe, two font-cluster/direction probes, and annotation_projection_cost_distributions. Their exact names are retained in raw output. Incidental chooser timings in stderr are not controlled evidence of a performance improvement or regression in source omission.
Independent read-only review checked the report against raw outcomes, the sole HTTP input difference, the expected storage refusals and the final archive/verifier. Review identified imprecise HEAD-only wording in the archive README; it was corrected to name the captured HTTP addition. No substantive finding remained. Parent readback verified all 613 executed source inputs and four archive files against the staged Git blobs; the archive verifier, targeted HTTP-file rustfmt check and staged whitespace check passed. The evidence ZIP SHA-256 is 43d3483eab331b98516d2ab8969a634c7c714fb827df1acfe972dd094923a9fb.
Reproduction and remaining acceptance
Environment: native Windows, Rust/Cargo 1.98.1, stable-x86_64-pc-windows-gnullvm, LLVM-MinGW 20260826 UCRT, locked offline dependencies, CARGO_BUILD_JOBS=1, CARGO_INCREMENTAL=0. Use the established project Windows toolchain environment setup.
cargo test --locked --offline -p oa-server --test http -p oa-cli --test source_cli --no-fail-fast --message-format=json -- --nocapture
cargo clippy --locked --offline --workspace --all-targets --message-format=json -- -D warnings
cargo test --locked --offline -p oa-desktop --bin arcadia-studio --no-fail-fast --message-format=json -- --nocapture
cargo build --locked --offline --workspace --bins --message-format=json
python harness/experiments/default-consumers-20260913/verify.py
Native drag, exact model Undo/Redo, saved-source preservation and reopen still require a completed observed workflow against the matching build. No earlier diagnostic GUI run substitutes for it. Source8 composition and combined verification remain under the ownership partition in Gitea issues 7 and 11. Linux checks remain user-paused. This checkpoint does not close issue 11, deliver a new release, or establish the full OA/SA objective.