ARCADIA-Studio/harness/PLAN.md

Superseded on 2026-09-11. This document is a preserved historical checkpoint. Live plans, priorities, ownership and handoffs now reside in Gitea issue 1 and issue 2. Do not resume synchronizing this ledger. See STATE.md for the integrated checkpoint.

First release execution plan

Updated 2026-09-11 for the published Codex-ytr83r B4 source checkpoint. PROJECT_GOAL.md defines the full outcome; STATE.md scopes evidence/ownership and TASK_LEDGER.md separates facts from decisions.

Immediate integration and verification

1. Preserve completed integration c81e285/e207b04/b853a1e and reviewed B4 source 52f6e17 on codex/release1-ytr83r. Do not replay equivalent3978/d114 changes. CLI-only01b76a1 was reviewed and imported as023e3a8 without core/dependency changes. Preserve the new AGENTS maintenance instruction and exact goal bytes.
2. Core433, independent lifecycle-oracle admission, focused lifecycle22 and full desktop675 passed on this host. Production freeze2908405 is unchanged by evidence checkpoint3ade397. GUI build, both-layer history/reopen and corrected workspace strict Clippy passed. Finish the affected test and Unicode/native workflows below. Do not repeat the passed suites without a source change or new finding. Root serializes writers and records exact scope.
3. Use the imported evidence2ec creation-oracle v2 for native Undo comparisons. Reject unknown/changed detached files while allowing exact files retained from an explicit pre-Undo origin. Preserve full Project and reachable SourceBundle equality and original role identities through Redo/reopen.
4. Exercise current Windows UI at minimum/wide sizes: creation name/owner/viewpoint, one-step history, stale/raw-draft rejection, first Escape/modal priority, recovery, clicked root/ALT action/observation/reuse, exact caller return, same-size Fit and natural subsequent scrolling. Use independent semantic predictions, not only screenshots or serialization of observed outputs.
5. Review current external Gitea handoffs before importing staged prerequisites, empty-state actions, contextual chooser or source CLI. Validate final combined workflows and preserve explicit ownership/qualification semantics.
6. After actual B4 test/native acceptance, prepare a distinct runnable Windows prerelease with current-toolchain notices and exact executable/source hashes. Existing MCP credentials have now completed a scoped branch push and remote readback; release attachment upload still needs concrete-artifact verification. Codex-9e8r6t retains the earlier named-authoring prerelease ownership.
7. Linux setup/downloads/builds/checking are paused by direct user instruction. Wait for the user to resolve WSL2 networking and explicitly instruct resumption. Preserve installed Ubuntu, exact VirtioProxy configuration and complete/partial offline inputs. Networking remains unresolved; preparation is not Linux acceptance. Follow the current WSL/offline records rather than old reboot plans.
8. Incorporate the owner clarification in issue2: process/scenario interaction should use consistent direct canvas workflows while preserving domain constraints. Codex-956atf owns the shared canvas design; B4 rendering and prerequisite forms support it but do not close this requirement.

Capability milestones

Outcome	Work and acceptance
Usable direct behavior/state engineering	Continuous ordered/explicit process authoring, clicked root/nested scenario insertion, exact reuse bindings/caller return, state naming/creation, recovery and source-preserving history. Complete B4 lifetime rendering with actual lifetime bounds, branch-local Delete activation clipping, incomplete lanes and self-arrow endpoints.
Precise engineering analysis	Refine property-owned baseline values/formulas plus explicit ModelBaseline/Case one-level overrides. Close quantity kind/unit/value, exact/approximate numerical policy, dependencies/cycles, false/missing/error, complete freeze and definition provenance. Implement coherent validated core/source/REST/GUI/history contracts. An exact-duration slice alone cannot satisfy the full goal.
Maintained reusable libraries	Resolve source-only normalization, pinned offline definitions/closure, provenance versus checked same-origin compatibility, source→local instance maps, explicit three-way updates/local divergence, subjectless analysis closure and OA/SA mounted-data transition policy. Demonstrate two projects/two instances, cosmetic and semantic upgrades, conflicts/detach and rollback.
Practical complete textual and graphical tasks	Measure small/large text-only creation, cross-file reference navigation, safe alias rename and lossless patch preview. Integrate source CLI and bounded contextual choices; actual keyboard/IME, repeated names, stale events, recovery and minimum geometry must pass. Keep offline guide/examples and normative source documentation aligned.
Complete engineering stories	Close OA↔SA transition/reconciliation, typed data/interfaces/state, artifacts, saved views, hosted workflows and tutorial tasks against the full current goal. Confirm parity and precise specification rather than inheriting historical exclusions.
Reproducible Windows/Linux release	Current tests/format/strict Clippy, reST/OpenAPI/migration/example gates; real GUI and REST transport/restart/ETag conflict acceptance; matching optimized desktop/server packages, complete notices/archive hashes and clean-runtime launch. Fresh independent requirements/domain/parity/architecture/UX/runtime review resolves material findings before release.

Milestones run adaptively where independent work helps. Analysis and libraries share definition/identity assumptions and must converge before broad migrations. Other-host slices remain coordinated through issue #1 and Wiki. Native/Cargo leases are host-local; source ownership and exact transfer heads are shared.

Evaluator boundaries and replanning

Use authoritative source semantics, exact UUID/source preservation and actual user workflows. Source-only experiments are not accepted new grammar; mounted widget tests are not native readability; synthetic packaging checks are not Linux runtime. Preserve old environmental failures as evidence without inventing current blockers. Continue independent work when platform or transfer work waits.

Archived prior plans preserve all previous milestone detail. Update this plan when a counterexample changes the right next action; do not lower completion criteria after substantial effort.