Files
ARCADIA-Studio/harness/experiments/default-consumers-20260913/README.md
T

63 lines
3.5 KiB
Markdown

# Consumer, desktop, lint and Windows build verification
Four closed gates captured the same 613 input files with HEAD
`df2e1cd65ca5e56f47669e04a9d4bcb426900fcd` plus the locally added, uncommitted
HTTP witness in `crates/oa-server/tests/http/port_anchors.rs`. That file is the
only changed captured input relative to the preceding core follow-up run. Each
gate confirms its inputs stayed unchanged during execution. The source ZIP and
input pins identify the executed candidate; HEAD alone does not contain the new
HTTP witness:
| Capture | Executed scope | Result |
| --- | --- | --- |
| `production-consumers-1` | CLI `source_cli` and server `http` integration targets | 12 CLI + 51 HTTP tests passed; none failed or ignored |
| `production-clippy-1` | Workspace, all targets, `-D warnings` | Exit 0 |
| `production-desktop-1` | `oa-desktop` / `arcadia-studio` automated tests | 894 passed, 0 failed, 6 ignored |
| `production-windows-build-1` | Workspace binaries | Exit 0; CLI, server and Studio compiler receipts |
All runs used native Windows gnullvm Rust/Cargo 1.98.1, locked offline
dependencies, one Cargo job and incremental compilation off. UTC intervals on
2026-09-13 were 11:51:14–11:56:49, 11:57:30–12:02:34,
12:03:15–12:10:24 and 12:10:46–12:13:19, respectively. Complete commands,
environment, timings and raw streams are retained.
The six ignored desktop tests are opt-in rename/layout/symbol-navigation timing,
two supported-font cluster/bidirectional-text witnesses, and diagnostic
projection performance measurements. Their exact names and reasons remain in
the raw log; the verifier checks their names. Passing automated desktop tests
does not establish an interactive native workflow result. Partial native trial
1 is intentionally outside this archive.
`evidence.zip` preserves the four original start/result/stdout/stderr sets and
the original `gate-policy.py` producer. Original source payloads were compared
byte-for-byte across all four gate directories before packaging. Identical
`inputs.json` and `inputs.zip` are stored once as `shared/inputs.json` and
`shared/inputs.zip`; each gate's independent receipts still pin these bytes:
- Manifest: `6bf74c1f68431a034681b754617c39d387783f71354c9106649aec9480e4f29f`
- Source ZIP: `e20f33dc363850a9289103a7be28ad613c3fffb2039061957a0186cac72d541b`
The internal `capture-files.json` pins all 19 physical payloads. Outer
`files.json` pins every other package file; its publishing Git commit identifies
the outer manifest. No current-source inventory or new execution claim was
created while packaging these closed captures.
Run with Python 3.10 or newer from this directory:
```text
python -B verify.py
```
This reads only. It checks memberships and duplicate ZIP entries; all file and
input hashes; fixed commit/helper/source pins; each gate's source references;
start/result and timestamp consistency; exact commands; named test outcomes,
per-suite counts and ignored names; and compiler-artifact receipts against raw
Cargo records. It does not extract files, run Cargo/tests, read current sources,
or look up executable files on the machine. Producer-host paths in the archived
runner require adaptation for a new capture.
No binaries are distributed. Compiler-artifact hashes are historical readback
receipts, not independently reverified executable files in this package. These
records establish the stated Windows automated gates, with no interactive
native, Linux, package-launch or release-publication acceptance claim.