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

133 lines
7.9 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# Scenario native verification — Windows local host, 2026-09-11
Codex-9e8r6t `/root`. This record covers the native v5 trial on the verified
named-creation GUI, not the other hosts' scenario trials or later source changes.
The complete release remains unaccepted. B2 diagram creation is in progress.
## Exact inputs and evaluator
- Desktop source milestone: `4bb4aaf`, followed by evidence-only `2ec72d4`.
Running GUI SHA256 `D4D9C2BD937E301FF1009D7CB93A34EE5AA6DFC837FF7E87D468C85371B20C31`.
Native PID1284/window6161106 uses the isolated v4 profile and runtime.
- Fresh work copy: `.local/native-qa/scenario-20260911-v5/work/oa/`, copied from
the complete committed `harness/fixtures/scenario-authoring/base/`.
It contains both OA and SA baselines: 62 elements, zero relations and two
reachable source files. The third JSON file is expected evidence only.
- Independent checker: `harness/scenario-authoring-contract.rs`, SHA256
`3BDA3EBC38E99BB658AD3EB9A5889E9C1CC77E8B1F8609079580AE27A4E8E789`.
Local executable SHA256
`BA9142DD98CC5EE386D344941216F4F7033F0B6992A4668F56C8B6A9A85D08C2`.
Compilation denied warnings. The initial mismatched serde_json artifact
caused a compile-only trait error; the corrected invocation matched the
core's dependency fingerprint and passed. No oracle source was changed.
Exact core/JSON hashes are in v5 `oracle-provenance.json`; both compile logs
remain beside it. Self-check and baseline comparison passed before native work.
## B3 observed navigation
At a 1440×900 client area, selected **OA Second use of B**, opened B, selected
**OA B use of C**, and opened C. Back returned first to the selected B use and
then to the exact Second use in A (`scenario_oa_61`). The canvas retained its
offset and the Selection shelf reopened. Ordinary subsequent canvas scrolling
moved naturally; another selection did not snap back to the restored offset.
Repeated the complete A→B→C→B→A path from **OA First use of B**
(`scenario_oa_60`), initiating it through the detailed-list **Open OA B
verification →** action. A returned to the first occurrence, its canvas offset
and the scrolled detailed-list position. First and second uses were never
treated as interchangeable. Before/after screenshots are retained in v5.
Resized through Windows' native Size command to an actual 1100×700 client area
(1102×732 window capture). Repeated both complete OA paths. Caller selection,
canvas position and the first use's detailed-list position returned correctly.
The normal three-pane layout requires horizontal scrolling at 100%; the
controls remained reachable. The attempted border drags did not resize the
window and are not counted as successful sizing operations.
At the same minimum size, used **Focus canvas** for SA, which hides both side
panels. Repeated A→B→C→B→A separately for **SA Second use of B** and **SA First
use of B**. The correct use, visible selection, focus layout and canvas offset
returned each time. The fixture's original uses retained their explicit two
lifeline mappings. These are UI observations; the oracle does not measure
selection or scrolling.
## Pending navigation and preservation
In SA A, armed **Add action** and clicked the inspector lifeline's explicit
root slot before SA First use of B. Typed `unfinished native action` into the
name field. The form showed that exact destination and retained the marker.
The Selection shelf exposed **Open definition** as disabled. Clicking the
disabled action did not navigate or lose the raw name/slot. At root A no Back
button exists, so no pending Back attempt is claimed. Explicit Cancel discarded
the draft, and Save retained the complete baseline.
Native Ctrl+Z after navigation displayed `Model change undone` even though no
accepted edit was intended. The old App handler unconditionally reports that
message regardless of Document's boolean result; the message alone cannot prove
that history contained an entry. Full baseline comparison after Save passed.
The desktop owner has added a narrow return-value/status correction with an
empty-history and real-history regression; that later source remains outside
this native artifact and still needs execution/native verification.
## Evidence and remaining work
First B2 OA action completed at minimum size with focused canvas. Armed Add
action and clicked the inspector root slot before OA Root entry. The form showed
that exact slot, `OA native action`, work OA Inspect ticket and observed lane
OA A inspector. Create/Save passed full `oa-action` prediction. One Undo/Save
passed baseline and exact `same` against the preserved B3 checkpoint. One
Redo/Save passed `oa-action` and exact `same` against the original first-commit
checkpoint. All three directories and five checker logs are retained in v5.
The next Observe state click correctly targets OA Inner chosen operand before
OA Inner anchor, on OA A inspector. Its form inherited the prior lower scroll
offset, initially hiding the name; an ordinary upward form scroll exposes it.
This repeats the known form-scroll issue assigned to the external workspace
owner. No source expectation or destination was changed to accommodate it.
OA state observation then completed with literal name `OA native state`, state
OA Ready and observed lifeline OA A inspector. Create/Save passed `oa-state`;
one Undo/Save passed `oa-action` plus exact `same` to the original action
checkpoint; one Redo/Save passed `oa-state` plus exact `same` to the original
state checkpoint. The two nested ALT structures and every sibling field/source
byte matched the independent prediction. These five checker logs and three
whole-directory checkpoints are retained in v5.
OA reuse then completed at the same nested chosen-operand slot before Inner
anchor: `OA native reuse`, definition OA B verification, with both explicit
inspector/recorder mappings to the caller lifelines. Create/Save, one Undo/Save
and one Redo/Save passed `oa-reuse` / `oa-state` predictions and exact `same`
comparisons to their original first-commit checkpoints. No definition was cloned.
OA fragment completed at the root slot before OA Root exit: `OA native
alternative`, ALT operator, guards `ticket.priority > 2` and
`ticket.priority <= 2`. Create/Save passed the complete `oa-fragment`
prediction. One Undo/Save restored `oa-reuse` and its original identities;
one Redo/Save restored the original fragment checkpoint exactly. The actual
file picker reopened the same project; the Architecture page reported the
source project opened. Save then passed both the full fragment prediction and
exact comparison to the first accepted fragment. All four OA B2 creations and
individual history checks are now complete on this D4D9 artifact.
The v5 folder retains the fragment's ready-guard screenshot, five fragment
checker logs, two reopen logs, three fragment checkpoints and a reopen
checkpoint/screenshot. The GUI closed normally after the final saved check.
The v5 directory preserves oracle compile/self-check/baseline logs, before/after
OA wide/minimum and SA minimum PNGs, the disabled-action capture and the final
navigation checkpoint. Baseline prediction compares every Project field and
both reachable source byte streams; `same` additionally pins original identity.
No semantic expectation was derived from the GUI's observed output fields.
The closed v5 evidence is preserved in
[the compact experiment archive](experiments/scenario-native-local-20260911/README.md):
80 exact payload files, 2,370,037 bytes. Fourteen checkpoints retain both exact
source files; the identical non-source expected-baseline JSON is stored once.
The manifest records every original physical file and path. No executable,
runtime DLL or mutable work directory is copied into the archive.
B2's separate SA action/state/reuse/fragment creations, exact individual
Undo/Redo and SA reopen, broader raw-draft/Escape cases and later B4
lifetime rendering remain separate open gates. A successful navigation trial
does not establish those results or overall release usability.