ARCADIA-Studio/harness/STATE.md

Progress Ledger — first release

Updated 2026-09-11, coordination through Gitea #190. This ledger describes the Codex-956atf checkout; external-host evidence is labeled separately. The entire previous ledger is preserved byte-for-byte. Historical dates, binaries, process IDs and stopped-session instructions in that archive are evidence history, not current status or authorization.

Current objective and source

Deliver the complete, usable OA/SA first release in PROJECT_GOAL.md, including graphical/source/REST parity, constraints/calculations, quantities/units, reusable model libraries, traceability, offline guidance and Windows/Linux release acceptance. Direct behavior/scenario authoring and reduced cognitive load remain immediate priorities. Complete small and large models must be practical to develop through uncluttered text alone. The release is incomplete.

Local branch: codex/behavior-authoring-20260910. Last compiled source checkpoint is 2cc4d115d0ff28423c7ed87fd5f4065efb011f2d, including external named creation 4bb4aaf as 1516949, the isolated chooser and final baseline/library contracts. Full desktop676/676, formatting and strict Clippy pass at that source. The active native GUI remains immutable80cf197. New workspace and production chooser changes are outside those artifacts and await a coherent build. Numeric foundation files are undeclared and under independent review. The authoritative goal was copied byte-identically from attachment a49d1780-25d5-4778-ad68-6140c60e4207/goal-objective.md; SHA256 DD58F75C93D55FE16ED141A23D834898FB6EBC625AE7E7AE379DA552A8312435. See Task Ledger for knowledge/decisions and Plan for capability acceptance. Older goal paths are superseded.

Completed work and evidence

The creation bundle was fetched anonymously and verified at105022 bytes, SHA256 0DEFAE91A56430076C432758F923BC0675911DA9FA7588C9961FF15F47EAEBF6. Its coherent commit adds named Process/Chain/Scenario/Machine drafts, bounded owner search, recovery/selection guards, offline guide and measured textual examples. The duplicate Sensor correction was not reapplied. Both hosts' report additions and exact experiment bytes are retained; this host's goal and ledgers remain authoritative locally. External653-test/native evidence is attributed in its integrated report, not claimed as a local run. The separate isolated chooser at c4e1580 passed13/13 mounted tests under cfg(test); it is not a production selector yet.

Scope	Current evidence and limits
Reproducible local toolchain	Ignored .local/toolchain/env.ps1 selects Rust/Cargo 1.98.1 GNU/LLVM and verified LLVM-MinGW 20260908 (clang 23.1.1), with one Cargo job. First cold desktop test build completed in 58m03s; one unused import warning was corrected separately.
B1 process authoring	01bb52a includes interactive empty/ordered/explicit topology and presentation-only inferred paths; first conversion preserves ordered adjacency atomically. All topology regressions pass within the current full desktop suite. Native acceptance is pending.
B2 scenario authoring	Clicked root/operand/slot and optional lifeline feed Action/State/Reuse/Fragment adapters. Pointer-to-App, cancellation, stale-event, recovery and history tests pass within the current full desktop suite.
B3 caller navigation and workspace	External 3978bbe imported as 865cd55 fixes visible-owner restoration, pending-draft guards and remaining-height workspace. Root 80cf197 keys the viewport Sensor by document/scenario; a real mounted regression failed before the fix and now passes. The full suite is green; actual OA caller return and source-neutral navigation pass. Remaining native scope is explicit below.
Architecture palette	External palette commit 7d1df4f was cherry-picked as aed40d7, retaining local invariant tests. Captured Escape, stale Submit/raw retention and bounded card fixes are source-integrated. External native results do not transfer to this checkout.
Typed stored-history F1	b6b76ef fixes nullable constructor cache discrimination; 4ecad78 fixes NotApplicable seeding by checking actual enum alternatives. Thirteen focused tests passed and are included in the current full suite; core validation remains authoritative.
Lifeline lifecycle B4	eb5161b adds pure local/scope-aware lifecycle projection with ten tests, declared under cfg(test) only. It is not production Scene wiring or native acceptance. Design.
Native scenario oracle	Independent checker compiled/self-checked and verified the generated62-element/two-source baseline. Native OA Action/State/Reuse/ALT-fragment, each exact Save/Undo/Redo, and final native reopen pass at minimum size; caller identity/scroll return and source-neutral navigation pass. Remaining prescribed checkpoints are open. Native scope, oracle inputs.
Normative documentation	Strict Docutils 0.22.2 language HTML regeneration passed. semantic_schema.py --check --validate passed 72 source7 fixtures, 204 schemas, 14 families/90 scalar leaves. These are documentation/schema gates, not a Rust build or runtime pass.
Comparable-product study	Actual isolated Capella 7.1 scenario/activity/lifeline/LOOP creation was saved and XML inspected, then closed normally. Report. Current Studio native evidence is recorded separately.
New requirement audit	Independent delta report establishes that current expressions/unit labels are descriptive and source includes are not maintained cross-project libraries. Refined B/R6/R7 and exact-pin/same-origin compatibility contracts are ratified; current source7 does not implement them.
Packaging integration	External b98f135 imported as 6d2b961; all 25 checked-in exact notices match the manifest. scripts/test-package-artifacts.py passes 11/11 and scripts/test-windows-notices.ps1 passes synthetic package/runtime/source/font checks. Actual optimized distributions and clean-environment launches remain open.

Active work, ownership and unresolved findings

Root Codex-956atf owns this host's integration, serialized Cargo and native UI. The source2cc4d11 gate completed with exit0: 676 passed, zero failed/ignored, tests611.24 seconds after warning-free compilation10m47s. Command: cargo test --locked -p oa-desktop --bin arcadia-studio -- --test-threads=1. Log .local/logs/desktop-creation-chooser-20260911.log. Immutable executable .local/creation-chooser-20260911/tests-2cc4d11/arcadia-studio-tests.exe, SHA256 8D7FB35071F155B3BD07CE1885A1136B2D2EF71C4FF8C704665202499C86A721. Its manifest pins256 tracked input hashes. Format passed; strict all-target desktop Clippy passed7m18s after the isolated manual-clamp correction. Final logs are copied beside the immutable binary. These gates supersede the historical642 total, not the current uncompiled working changes. Focused counts are included, never additive.

Actual native Studio remains .local/native-qa/studio-80cf197/, GUI SHA256 A5ED6DF334B1BFB620425AFDF3E863F0B522990E6FB47C6C401A71CF4443954A. That GUI build passed3m28s. At1100x700, all four OA creation checkpoints and final reopen now pass complete Project/two-source prediction and exact UUID restoration. Preserved snapshots end in oa-reopened/; no SA or new-fix acceptance is inferred. Native report records details and two real workspace findings: new forms retain downward scroll, and opening a form can hide its clicked marker below the reduced canvas. Root has written a fresh form-session key, guarded Name focus, one-time marker reveal and three meaningful mounted/App regressions. They are not yet compiled. The isolated chooser's corrected realistic typing/paste/IME gate passes13/13; production inspector/review integration is now owned by local behavior_audit, with exact row/facet/group/attachment guards.

The baseline contract supplies a complete17-UUID proposed model,11 exact fixtures, explicit ModelBaseline/Case, property bases and shared checks. Its transport fixtures are not parser/evaluator acceptance. The library contract has120 executed finite-prototype assertions, separate-process reopen and11 complete expected/reopened file comparisons. Final81-file manifest was independently verified; preliminary output is retained and labeled superseded. All166 experiment files retain exact Git bytes. Root ratified B/R6/R7/full private capture, nominal typed compatibility alongside exact pins, explicit source normalization and OA/SA dispositions in DECISIONS.

The independent refined review passes292 fixture/identity checks but finds C1:333 stored AST nodes can expand to over2billion calls. Request-local dynamic/numeric work accounting is mandatory; static depth/node limits alone fail. C2 is a finite-prototype threshold-unit gap, not a passing production type-checker. Local requirements_delta owns undeclared analysis_numeric foundation (exact arithmetic, dimension/unit operations and terminal work meter), with semantics_critic independently reviewing correctness. Root owns dependency/module admission and all compilation. Numeric ceiling values remain provisional pending measured gates. Approximate engineering calculations, full evaluator/model/source/REST/GUI parity and actual library maintenance are open.

Codex-9e8r6t and Codex-ytr83r are separate hosts/teams. The former owns contextual prerequisite creation, source CLI and next Windows named-authoring prerelease. Its direct branch codex/release1-9e8r6t has coherent CLI checkpoint01b76a1; corrected evidence2ec72d4 and CLI are not yet imported here. External675 staged-source and CLI12/183-assertion results remain externally attributed. Codex-ytr83r owns B4 production Scene wiring, independent native oracle and local Linux route. Its rendering source has independent criticism but no passing build/native gate yet. Do not duplicate these changes. Preserve root's distinct scenario-scroll/focus hook when integrating other hosts' small App/behavior bridges.

Updated AGENTS d93cddac adds coordinated runnable milestone prereleases. Gitea user#178 explicitly authorizes programmatic reuse of existing MCP credentials for Git/REST and relevant commits/tags/prereleases, without printing or committing them. Direct push setup is in progress on this host;9e8r6t has independently succeeded. Historical verified bundle exchange remains reproducible: c6d296a bundle on codex/behavior-transfer-20260910,334830 decoded bytes, SHA256 526A80E2D32ED0E57A7B2372B5F02D4652E8751DFEBA81953C4F045225D6AACB, prerequisite1797516. Both other hosts received it. Never merge transfer branches. Older bundles/logs retain historical scope in Git and the archived ledger.

Remaining gates and highest-value next actions

1. Freeze and compile the new workspace/chooser integration, resolve focused failures, then build an immutable GUI and verify both native workspace fixes. The676-test gate is complete for2cc4d11; new changes require their own evidence.
2. Publish coherent checkpoints directly; integrate9e8r6t CLI/corrected evidence and later staged-source work, then ytr83r B4 only after its coherent handoff. Preserve ownership and distinguish every external versus local gate.
3. Use the self-checked independent oracle to exercise current Studio at minimum/wide sizes: named creation, direct process/scenario edits, nested reuse, caller return, lifeline lifecycle, invalid/cancel/recovery, Save/Undo/Redo/reopen. Compare complete model/source and exact identities, not screenshots alone.
4. Compile/review/measure the numeric foundation, then implement the ratified analysis/library contracts with shared-core/source/API/graphical parity. Use real budgets, unit/kind failures, complete text-only and two-project update workflows. Integrate the bounded chooser and run its actual keyboard/minimum-size gates.
5. Complete current teaching, transitions, semantic review, state/data, saved-view, artifact, hosted transport and migration acceptance. Historical scoped passes remain useful references, never blanket current acceptance.
6. Complete Windows/Linux build/test/native/runtime and optimized package gates, notices, provenance and clean-environment launch. WSL has no installed distro here; Linux acceptance remains open. Historical denied guest workarounds must not be retried or split to bypass review. No host restart is authorized.
7. Run fresh independent cross-disciplinary end-state review, resolve material findings and reconcile specification/help/coverage before final handoff.

Reassessment

Meaningful progress continues: concrete interaction defects have been found and corrected,676 desktop tests pass and all four OA native creation/history/reopen checks pass. Contracts have converged through independent falsification; numeric implementation now addresses the exposed dynamic-work risk. The revised semantic requirements are no longer hidden behind metadata. Passing older suites or accumulating prose would not close the user problem. Current highest information gain comes from actual build/native results and independently criticized foundational examples, while preserving existing identity, source and atomicity contracts. Replan when those results falsify a proposal.

Historical evidence access

The complete prior STATE retains exact historical binary hashes, counts, logs, process closure, bundle hashes and all report links. Links inside the byte-preserved archive retain their original harness/ context; resolve those relative report paths from harness/. Useful direct entries: source7 native, teaching, artifact native, coverage baseline, Linux gate, Linux cross-build, denied boot experiment.