Files
ARCADIA-Studio/harness/STATE.md
T

14 KiB

Verified branch checkpoint — 2026-09-23 UTC

This describes Codex-956atf's contribution, not a combined release. PROJECT_GOAL.md remains authoritative. The complete OA/SA application is unfinished; published previews and finite prototypes are intermediate.

Linux/WSL builds and checks remain deferred until the user explicitly reports networking resolved. No guest setup, diagnosis, restart or cache transfer is active. The VPN remains untouched.

Current integrated candidate and evidence

Branch codex/behavior-authoring-20260910, published as codex/release1-956atf. Integrated code checkpoint 4ffff0d9d9 combines the reviewed ordinary/Text contribution 3f9e6a18af with source-editor Tab/Shift+Tab indentation. Windows desktop verification passes 1,188 tests with six existing ignores; all 1,194 names and nine targeted indentation cases are accounted for. Strict desktop Clippy and matching GUI build pass on 711 unchanged inputs. The closed native trials cover caret and forward/reverse block indentation, text undo/redo, empty drafts, continued caret typing and Apply/Save/reopen. The saved 577 bytes exactly preserve the baseline's mixed line endings and Unicode while indenting its 13 physical lines. The first trial restored the baseline and is not a save oracle. Find's Tab does not advance focus; this inherited limitation is distinct from the verified editor boundary. This is Windows contribution acceptance, not complete editor accessibility, public source8, a combined peer union, or a packaged release. Exact gates, original native captures and independent review retain both the failed initial compilation and the interrupted first trial.

Earlier code checkpoint 8abc144cf2 preserves the production inputs of feab2020d8. Its Windows desktop suite passes 1,096 tests with three existing manual ignores; all 1,099 names are admitted. Targeted navigation/rename checks, strict desktop Clippy and the matching development GUI build pass. The closed native trial verifies Details-to-Preview scroll restoration, explicit reset boundaries, and exact six-file preservation after Cancel and Save. The portable evidence, published at f95d031048, retains the initial failing test, 20 original frames and independent review. Its committed bytes and verifier were rechecked on September 22. This narrow trial does not establish Text integration, complete rename acceptance or release readiness.

Earlier production source follows; each result retains its original code scope. Verified production source 44095f78c2 includes Find focus, departure-selection and scoped Escape corrections. Its Windows desktop suite passes1095 with3 existing manual ignores; core passes471 with1 manual ignore; server passes56. All three strict Clippy scopes and the matching development GUI/server build pass. Exact source, commands and raw evidence. The integrated documentation checkpoint 3c22fe2629 preserves those production inputs and includes the updated AGENTS instructions.

The actual440 native trial closed normally with79 notes,78 images and14 complete saved-model/source comparisons. Rename/Undo/Redo/reopen, scoped selection, placement/state targeting and sampled OA/SA lifecycle/topology navigation are verified within the closed trial's explicit scope. It found a same-page Details-to-Preview scroll-loss defect; initial keyboard Delete after chooser navigation also required an explicit canvas click. Neither is hidden by passing saved-byte checks. The scroll defect is corrected and verified by the separate feab trial above; the initial chooser-to-canvas keyboard focus friction remains a limitation of the 440 trial. This remains a development checkpoint, not a combined release. Integration issue27 owns current verification, correction and release coordination. Earlier source e1240295bd combines the canvas contribution with source/diagram navigation and reviewed alias rename. Its1082-test listing,17 cross-feature tests,4 mounted preview-scroll tests and strict all-target Clippy pass on578 unchanged inputs. Combined full-suite, GUI/native and package acceptance are not established. Exact partial evidence. The matching core comparator and migrated fixtures pass their bounded preparation checks, including an independently specified six-token rename. The original migration-evaluator failure and its correction are preserved in preparation evidence; these results do not establish GUI editing or save/reopen acceptance.

Earlier source e44f07565c combines direct process/scenario authoring, allocation interaction, focused deletion and Save/Undo focus fixes, typed nested activations with lifecycle clipping, and stronger palette roles. Its custom menu/chooser target cues pass 12 and 14 actual-renderer tests, and the complete Windows desktop suite passes all 874 tests on 538 unchanged inputs. Strict Clippy exposed two test-only pixel-iterator lints, corrected by subsequent source dd39e1fdc1 after the required array-reference coercion correction. All ten affected style tests, strict Clippy and the matching GUI build pass at dd39. Its bounded native trial closes normally with 41 frames and five complete Project/exact SourceBundle comparisons: lifecycle rendering, selection, custom overlay cues, architecture/state boundaries and process previews pass within the stated scope. Native evidence and limitations. Earlier native results retain their exact source scope.

  • d61f1f3: all 836 Windows desktop tests, strict all-target Clippy and matching GUI build pass on 533 unchanged inputs. Its bounded native trial passes nested activation border continuity, exposed-shoulder versus exact-message selection in bound OA/SA and reentrant scenes, minimum-window Fit, and exact saved Project/SourceBundle preservation. Native evidence and independent review.
  • aadc639: reviewed lifecycle/typed-activation integration passes 36 lifecycle, 14 activation and one atomic Call/Reply tests on 536 unchanged inputs. The six- scenario native fixture additionally passes actual-core admission, exact source and canonical whole-Project round trips, ordering, values and operation bindings. These are targeted/core results; desktop native acceptance is separate. Integration and fixture evidence.
  • afa74a3: all ten targeted palette tests pass on 538 unchanged inputs, including actual WGPU selection/copy across an enabled-to-read-only field rebuild. Essential canvas and process boundaries, Source selections and resolved control colors are covered within the report's stated scope. Later a357b22 adds custom-overlay cues; its new paint tests exposed a missing test lifecycle step. Test-only e44f075 initializes the real overlay layout before drawing; corrected rendered tests and its 874-test full suite pass as recorded above. Palette decisions, reviews, exact results and limitations.

Earlier accepted focus, architecture/state deletion, Inspector text protection, bound scenario Save/reopen and discovered paint defects retain their exact 83f53c7 native evidence. Canvas follow-up evidence links the original failures and corrections. Earlier authoring and allocation checkpoints remain in canvas authoring evidence and allocation interaction decisions.

Isolated contributions and material limitations

Analysis/core source checkpoint a2a28f8d23, publication branch codex/analysis-core-956atf-20260911, is not integrated into production. Its private staged typed parser/lowerer has142 listed tests, 34 passing typed cases,1 passing raw-token case and strict all-target core Clippy on531 unchanged inputs. The complete library run remains126 passed/16 failed, with all16 failures at the unsupported public source8 header. The original fixture error and its test-only correction remain in exact S2b evidence. Published private writer f1d79df497 has portable S2c evidence: 152 listed, 44 typed and one token test passed, strict Clippy passed; the full library result remains 136 passed / 16 failed at unsupported public source8. Published source 1b1c3e3abb adds a private qualified source index/rename plan and the independently reviewed cleanup for the demonstrated depth129 rejected-tree stack overflow. Its S3 evidence, published at 345a42e465, records 167 listed tests, 59 typed passes (nine index/five update), one token pass, three clean stack controls and strict Clippy on 538 unchanged inputs. The original overflow and dirty diagnostic/probe results are retained. No current full-core pass is claimed. Public source8 parser, writer, source-update and persistence composition remain incomplete; these branches are not integrated production analysis.

The reviewed S1/S7 assessment distinguishes static analysis/case overlays from contextual inheritance, fixed/default bindings and executable trade studies with durable result provenance. Strengthened fixture and parity criteria remain proposals, not implemented evaluation or a replacement metamodel decision. Earlier S2a evidence retains its 129-test113/16 library result and separate7/9 language integration result, including the old V2 oracle that omitted the empty definitions catalog. Earlier bridge evidence remains scoped to6216cff. The earlier2c515cf full library run had98 passes and16 failures:14 rejected the unsupported source8 header and2 exposed the now-corrected legacy bridge mismatch. Its original failed suite and naming lint remain in historical evidence. No current full-suite or complete bundle migration acceptance is established. Its design and source reviews are preserved in that branch's harness/analysis-source8-20260913/. Public source8 support remains incomplete. At its earlier 847b1caa7e base, Windows checks pass the128 KiB-thread lifecycle subprocess, six codec tests, 28 analysis tests, 26 numeric tests and strict all-target core Clippy. The driver is included in the six; these counts are separate scopes. DEPTH_128_VALIDATED remains false. Its existing public checked-save regression now fails: saved source declares schema8, which the parser rejects. All 512 inputs remained unchanged; the JSON iteration did not run. Failure and repair boundary. Legacy/migration integration, optimized resource measurements and complete source8/transport/evaluator acceptance remain open. Exact codec evidence.

Complete direct manipulation and persistent process placement/lanes, coherent peer navigation/source integration, quantities/calculations/constraints, maintained libraries, graphical/source/REST parity, transitions, ergonomic engineering workflows, aligned reST/help, optimized packages and final independent acceptance remain required. Palette arithmetic does not establish every monitor's appearance; thin overview geometry, syntax scanability, focus, custom overlay appearance and routing-warning discoverability retain their explicit limitations. Gitea owns live priorities, ownership, dependencies and handoffs.

Authoritative records

Duplicated local ledgers were migrated once into Gitea comment412. Their original bytes and manifest preserve historical assumptions, decisions and evidence. Archived links retain their original harness/ base. Do not recreate those ledgers or repeat migration.