Files
ARCADIA-Studio/harness/experiments/scenario-native-local-20260911/README.md
T

107 lines
5.6 KiB
Markdown

# Closed local scenario native evidence — 2026-09-11
Codex-9e8r6t `/root/palette_finish` preserved the closed
`.local/native-qa/scenario-20260911-v5/` run. The authoritative workflow record is
[the root native report](../../release-scenario-native-local-20260911.md).
This directory preserves observations and existing checker results; it adds no
semantic expectations and reruns no compiler, oracle, application or native task.
## Scope and count
**80 original payload files, 2,370,037 bytes**, copied exactly and reread against
their original paths:
| Evidence | Files |
| --- | ---: |
| Two `.arcadia` sources from each of 14 checkpoints | 28 |
| One shared, identical `expected-baseline.json` | 1 |
| Original compile, self-check, model/source and identity logs | 29 |
| B2/B3 screenshots | 21 |
| Original oracle provenance JSON | 1 |
`manifest.json` records every original path, destination, byte count and SHA256.
Its SHA256 is `cd0dba7fe8031128282f41dab0aa02e2f7c4c37debddbd32843b2b94b6baef33`.
`SHA256SUMS.txt` covers all 80 copied payloads; its SHA256 is
`c819bad78484167c21dfbd9e9a3647a487ec1f33d276e865f2d39c35b117322d`.
`preserve.py` is the bounded copying/checking method and `preservation.log`
records its successful exact-byte result. Files already present must match
the original bytes; they are never overwritten with different content.
Excluded: mutable `work/`, `scenario-authoring-contract.exe`, `libunwind.dll`.
The original oracle source and baseline fixtures are already committed at
`harness/scenario-authoring-contract.rs` and
`harness/fixtures/scenario-authoring/base/`; they are not duplicated here.
## Physical checkpoint inventory and shared evidence
Every original checkpoint contained exactly three physical files:
1. `project.arcadia` — model entry, preserved in that checkpoint's directory.
2. `rationale.arcadia` — second reachable source, preserved there byte for byte.
3. `expected-baseline.json` — copied fixture evidence, not an authored model
source or a new expected model derived after an edit.
All 14 JSON copies were identical: **38,589 bytes**, SHA256
`abd8246c939ebb8a2162ef82e23d0fb322ee1a8b4d0d7c28a34037e42e86e841`.
Only `common/expected-baseline.json` is stored here. Each checkpoint's manifest
records all three original files and maps its JSON to that shared copy. Thus the
complete original **42-file physical inventory** is recorded without 13 needless
JSON duplicates. If reconstructing an original whole-directory checkpoint,
place that exact shared JSON beside its two preserved sources.
| Checkpoint group | Preserved directories |
| --- | --- |
| B3 navigation/cancel baseline | `checkpoint-b3-navigation` |
| OA Action | `checkpoint-oa-action`, `checkpoint-oa-action-undo`, `checkpoint-oa-action-redo` |
| OA State | `checkpoint-oa-state`, `checkpoint-oa-state-undo`, `checkpoint-oa-state-redo` |
| OA bound Reuse | `checkpoint-oa-reuse`, `checkpoint-oa-reuse-undo`, `checkpoint-oa-reuse-redo` |
| OA Fragment | `checkpoint-oa-fragment`, `checkpoint-oa-fragment-undo`, `checkpoint-oa-fragment-redo` |
| Final OA reopen/Save | `checkpoint-oa-reopen` |
## Existing verification and failure provenance
GUI identity from the root report is
`D4D9C2BD937E301FF1009D7CB93A34EE5AA6DFC837FF7E87D468C85371B20C31`.
The retained oracle provenance identifies executable
`BA9142DD98CC5EE386D344941216F4F7033F0B6992A4668F56C8B6A9A85D08C2`
and original source
`3BDA3EBC38E99BB658AD3EB9A5889E9C1CC77E8B1F8609079580AE27A4E8E789`.
These are recorded artifact identities; no executable is copied or executed.
`logs/oracle-compile-incompatible-json.log` preserves the meaningful first
compile failure: E0277 from mismatched serde/core build artifacts. The corrected
JSON/core dependency fingerprint compiled the unchanged oracle with warnings
denied in **6.585732 seconds**, recorded at `2026-09-11T11:30:26.4512964Z`.
`oracle-compile.log` is the original empty successful diagnostic file;
`oracle-provenance.json` supplies the successful build identity. Empty output
alone is not treated as success. Existing self-check and baseline logs follow.
The recorded native scope is complete OA B2 Action, nested State, explicitly
bound Reuse and Fragment: each Create/Save, one Undo/Save and one Redo/Save
passed independent complete Project/two-source predictions, with `*-exact.log`
comparisons pinning the original accepted identities. The final OA file-picker
reopen/Save passed the fragment prediction and exact identity/source comparison.
B3 observations remain separately bounded: both OA use-return paths at wide
and minimum size, both SA paths at minimum with focused canvas, and disabled
pending Open retaining the raw draft before explicit Cancel/Save. Screenshots
support the named native observations; the oracle does not measure focus,
selection or scroll offsets. The root report retains failed border-resize
attempts, the old misleading no-op Undo status and inherited form-scroll issue.
**No local SA B2 pass is claimed.** Separate SA creations/history/reopen, broader
raw-draft/Escape cases, later contextual-source GUI behavior and B4 lifetime
rendering remain outside this closed D4D9 evidence. Existing self-check coverage
of eight OA/SA expected stages does not establish native execution of them.
Proposed exact-byte attribute rules for root integration:
```gitattributes
harness/experiments/scenario-native-local-20260911/** -text whitespace=cr-at-eol
harness/experiments/scenario-native-local-20260911/**/*.log whitespace=-blank-at-eof,cr-at-eol
```
The log rule preserves original trailing blank lines without rewriting measured
compiler/checker output. This worker made no attribute or Git changes and did
not edit the root report, prior reports, production sources or common ledgers.