Files
ARCADIA-Studio/harness/PLAN.md

8.6 KiB

First release execution plan

Updated 2026-09-11. PROJECT_GOAL.md is the authoritative current objective. STATE.md records exact execution/evidence and leases; TASK_LEDGER.md separates facts, assumptions and decisions. The complete previous plan is preserved byte-for-byte. Its earlier goal paths, active-process claims and Escape pauses are historical.

Outcome and working principles

Deliver usable, complete OA/SA engineering workflows on Windows and Linux, with one precise reST contract and equivalent graphical, textual and REST editing. Constraints/calculations, quantities/units and reusable libraries are explicit requirements alongside behavior/state modeling, traceability and navigable help. Small and large models must be practical to author entirely in uncluttered text; evaluate complete authored and stabilized models, not abbreviated grammar sketches. Later Arcadia layers remain future scope. Compilation, metadata fields and polished screens alone do not satisfy this outcome.

Retain the shared typed core, stable identity, semantic ownership/allocation distinct from geometry, validated atomic commands and exact source/history. Persistent extensions need coherent references/deletion, transition, source/migration, API/schema and reachable graphical authoring. Define observable acceptance before broad downstream changes. Do not silently promote a design proposal to a decision.

Capability milestones

Milestone Work and acceptance Dependency/current limit
1. Reliable direct authoring Exercise actual minimum/wide native creation, allocation, editing, rejection, cancellation and history without implicit draft application or placement drift. Integrate compact named behavior creation after its coherent external gate. Named creation4bb4aaf is imported; combined desktop676/676, format and strict Clippy pass at2cc4d11. Native80cf197 passes all four OA creations/history/reopen; new workspace fixes and production chooser await compilation. Workspace/B3/palette/recovery/packages are imported. Root owns this host's Cargo/native lease.
2. Coherent process/scenario work Interactive ordered/explicit process editing; clicked root/nested scenario insertion; exact reuse bindings and caller return; measured canvas/zoom/scroll; visible scoped Create/Delete lifetimes. Validate model/source/identity and no unintended edit during navigation, including stale/deleted/retargeted references and recovery. B1/B2/B3 regressions pass in the full suite. Local B4 remains cfg(test); Codex-ytr83r owns production Scene integration and its independent native gate. Use the independently self-checked oracle for native checkpoints.
3. Precise engineering analysis Compare owned properties, analysis contexts and existing-data/external alternatives; settle quantity kind/unit/value, numerical policy, calculation/binding and Boolean/error/unknown semantics through falsifying examples. Implement mixed-unit budgets and reusable calculation/constraint definitions with full UI/source/REST parity and exact history/recovery. B/shared-check/full-capture and R6/R7 architecture ratified after complete fixtures and criticism. Exact numeric/work-meter foundation is undeclared and under independent review. Approximate profile and full evaluator parity remain required.
4. Maintained reusable libraries Compare package/definition/instance alternatives. Establish pinned identity/version/dependency closure, local edits, import bindings, detach and explicit three-way updates. Demonstrate two projects, two independent instances, offline reopen, conflicting updates and stable unaffected identities. Exact pins plus verified same-origin typed compatibility, explicit source preparation/acceptance and three-way editable updates are ratified. Finite120-assertion experiment is not production contract/archive/consumer implementation.
5. Understandable complete engineering tasks Close teaching, OA↔SA transition/reconciliation, typed state/data/interface, artifact, saved-view and hosted workflows. Establish complete text-only creation/refactoring/modular maintenance with stable identity. Replace concrete long homogeneous selectors with contextual/searchable bounded choices; verify large models, keyboard access, minimum sizes and failure/recovery. Keep offline guide/examples aligned. A complete warning-free text-only model probe passes source/identity/rename/reparent/file-move checks; measure usability separately. Anchored chooser passes13 isolated mounted tests under cfg(test); production inspector/review integration is active; its compile/native gates remain open. Separate-host CLI is coherent at01b76a1 and awaits import here. Reassess current goal rather than copying historical scope exclusions.
6. Release confidence and delivery Fresh current Windows/Linux suites, strict Clippy/format, normative reST/OpenAPI/migration/examples; actual GUI/service transport, optimized builds, runtime/notices/archive/provenance/checksums and clean-environment launch. Fresh independent requirements/domain/parity/architecture/UX/runtime review; resolve material findings, then final handoff. Integrates all preceding capabilities. Linux runtime remains open; use an approved supported environment, preserving historical failed/denied experiments. Historical previews are not final distributions.

Milestones are capability gates, not a rigid serial schedule. Root can continue build/native acceptance while independent foundational investigations compare alternatives. Libraries and analysis share definition-resolution assumptions and must converge coherently before broad model/schema edits. User workflows and cross-representation evaluators guide implementation boundaries.

Immediate execution sequence

  1. Complete the coherent workspace/production chooser build and focused gates, then stage a new GUI for real engineering tasks. Full676/format/Clippy passed for2cc4d11; working changes require their own verification.
  2. Implement the user-required common canvas interactions for Process and Scenario in issue2 after the bounded shared-viewport/domain-constraint design decision. Treat this as direct-manipulation acceptance, not a tool-shelf restyle.
  3. Exchange exact verified commits with the separate host, preserving its workspace ownership. Direct authenticated publication works. CLI01b and exact evidence2ec are imported; next contextual-source and B4 handoffs retain their owners.
  4. Use the self-checked scenario oracle, follow its native procedure, and compare exact whole snapshots for Undo/Redo/reopen at minimum/wide sizes.
  5. Exact numeric/unit/work-meter foundation26/26 unit tests pass at7d14bd0; complete its strict gate while actual schema8 core work proceeds in an isolated worktree. Extend actual model/source/API/GUI under the ratified B/R6/R7/full-capture and exact-pin/typed-compatibility contracts. Preserve explicit approximate numeric, complete consumer/archive and two-project release gates.
  6. Update current ledgers and durable decisions when evidence changes the plan. Reassess unresolved assumptions before expanding downstream implementation.

Coordination and evidence rules

Use Gitea issue #1 and Wiki for current ownership/handoffs; read before resuming. Codex-956atf, Codex-9e8r6t and Codex-ytr83r are distinct hosts. Cargo/native leases are host-local; shared source ownership is coordinated.9e8r6t owns the next runnable Windows named-authoring prerelease and contextual-source/CLI; ytr83r owns B4 Scene and its platform gates. Root owns local chooser/numeric/workspace integration. Updated AGENTS requires runnable milestone prereleases. User#178 authorizes existing MCP credential reuse for Git/REST; secrets must never be printed or committed. Direct branch publication replaces bundle work when available; old exact bundles remain valid evidence and transfer branches must not be merged.

Record exact source/binary, command exit, test scope and native model/source oracle. Do not add focused reruns to full-suite totals or transfer old binary acceptance. Older denied Linux workaround instructions remain applicable to that action; do not bypass them. Older stopped native sessions do not prevent currently authorized native work. The archived plan and archived progress preserve all previous strategy and scoped evidence; archive report links resolve from their original harness/ directory.