Files
ARCADIA-Studio/harness/text-default-omission-native-20260913.md
T

6.2 KiB
Raw Blame History

Ordinary default omission: native editing verification

The matching Windows application completed graphical creation, drag, model Undo/Redo, Save As and reopening for a compact two-file source project. The saved files match the explicit byte oracle: creation adds one include and one compact activity declaration; moving the existing activity changes only its x coordinate from 120.0 to 190.0. The authored port_anchors [] ; field, spacing and rationale comment survive unchanged.

This is branch-specific ordinary-source acceptance evidence. Integration with source8 and the complete OA/SA release remain separate work in Gitea issues 7, 11 and 27.

Candidate and observable checks

Production code is 7bb366f66a8c6f8b6f66fc9b04c9e3b39f84b78e. The GUI and CLI were built by the published consumer gate, whose 613 captured inputs are published at 94383107153cccb32f3fdcbd8111b6381f3e15aa. The build captured HEAD df2e1cd plus the subsequently published HTTP witness, not HEAD alone. No production edits or new build occurred during the native trials.

Observation Evidence and result
Trial 1, 12:13:57–12:28:17 UTC Opened compact source, created an activity through Model Explorer, inspected its compact declaration, saved the two-file baseline, closed normally.
Trial 2, 13:10:06–13:15:23 UTC Opened that baseline, dragged Work, observed x 190.0, chose Undo model change and observed x 120.0 plus saved state, then Redo model change and observed x 190.0 plus unsaved state.
Save As and reopen Saved to a separate folder, selected New to clear the current model, then opened the saved result from Recent models. Source again showed x 190.0, the original explicit empty field/comment, and the compact included activity.
Exact saved-source comparison Both bundles contain exactly the expected two files. The included activity is byte-identical before/after; the root changes only the expected position token. Original baseline remains unchanged.
CLI checks on both saved bundles Both exit 0 with empty stderr, no proposed changes and no identities awaiting stabilization. Each reports four expected draft warnings.
GUI closure Both trials exit 0, preserve the verified executable hash, and have empty stdout/stderr.

The four draft warnings are unallocated_leaf and work_without_behavior_context for Work and New activity. These are expected for this deliberately minimal fixture; there are no parse or model errors. The created activity retains UUID 1e1da408-da79-42ab-a990-9d00b7dc2631 through both saves and reopening.

The GUI SHA-256 is 4070927256414a770c5fbcedfaab922810cea33d9f07610b81e9ab8c0465da44 (115,056,640 bytes). The CLI SHA-256 is d89aba9f10b598926dfdc5c644474ae891b753e8583a184860e941a359ed0eea (20,608,512 bytes). The baseline root SHA-256 is dad153f45038ed9ae79ea7263364e6cdedc40737fe7f63fdad7fb198ee7f65c8; the edited root is c93bd35eaaf3a19bc07b5c6a4ee5839ab565e9fd90a2605261b7618fc8a7ecef. The unchanged included file is 42efe80410d56ae9dd919bad6b0633129028339b8271f500f35d9148a058f50a.

Interpretation and limits

Native Undo/Redo is an observed UI check. The byte oracle checks persisted bundles; it does not expose or compare the complete in-memory model at the intermediate Undo step. CLI checks parse and validate both full bundles, without comparing their complete decoded models. The separately published HTTP and automated desktop tests cover their documented model equality and exact Undo invariants; those results should not be described as additional native observations.

The diagram recenters on reopening, so comparing screen coordinates across the reopen would be misleading. Persisted x is verified in Text and the saved-byte oracle. Save-dialog accessibility reported Search Box while the filename field visibly held focus. In trial 2 an initial selection attempt did not take effect and typing appended to the filename. Observation caught this before saving; a confirmed selection and replacement produced the intended full path. Original captures retain the correction. This is an automation focus limitation, not evidence of source corruption.

Trial 2 pins its runner and baseline archive contemporaneously. Trial 1's original runner and the capture snippet are retained, but the trial 1 receipt does not hash those producer files. Do not infer stronger contemporaneous provenance from later archive pins. These short trials do not establish performance, accessibility completeness, large-model usability, cross-platform acceptance or complete release readiness. Linux remains paused by the user.

Reproduction and portable evidence

The portable package contains the actual inputs, saved bundles, original captures and process/CLI receipts. Its read-only verifier checks archive integrity and the scoped machine oracles; screenshots still require human inspection against the workflow above.

Independent review inspected the 14 key original frames covering creation, movement, Undo/Redo, Save As correction, clearing the model and reopening both source files. It independently checked the saved-byte oracle and process/CLI receipts, ran the portable verifier, and compared all 160 archived original payloads against the raw local evidence. No actionable finding remained. Parent also ran the verifier. The ZIP SHA-256 is 0eb6f2a566deff1275b19b8c5718ebb56f9afde0518ac65c09d885a1d6e84d68 (5,251,658 bytes).

python harness/experiments/default-native-20260913/verify.py
arcadia check <extracted>/native/saved-before/project.arcadia --json
arcadia check <extracted>/native/saved-after/project.arcadia --json

To repeat the native task, build the identified candidate using the documented Windows GNU LLVM environment, open the archived input, create an activity through Model Explorer, inspect its generated file, save a baseline, drag Work, use model Undo/Redo in Text Actions, save to a separate folder, create a new blank model and reopen the saved result. A new activity receives a new UUID; check that it remains stable within the new trial rather than expecting this run's literal UUID.

No issue closure, integration, tag or release is claimed by this report.