ARCADIA-Studio/harness/TASK_LEDGER.md

Task Ledger — first release

Updated 2026-09-10. STATE.md owns current execution/evidence; PLAN.md owns capability milestones. The entire former Task Ledger is preserved byte-for-byte. Its old goal references, aggregate test claims and stopped-session directives retain historical scope. This ledger does not add new semantic design decisions.

Goal and success criteria

Deliver the usable, feature-complete OA/SA first release specified in PROJECT_GOAL.md. It is the exact copy of the user's revised 2f1f0002-e43c-45ba-80a8-b359f05f99d0/goal-objective.md, SHA256 7C1BDBD077E31439501C04DC3613E6BE47A389704DDC3A44EF8BD9AD98EBB513.

Success requires coherent explicit semantics, practical engineering workflows, equivalent graphical/textual/REST behavior, real Windows/Linux runtime evidence, recoverable authoring, relevant automated evaluators, precise documentation, reproducible packages and fresh independent cross-disciplinary end-state review. No known blocker or high-impact finding may remain unresolved. Later Arcadia layers remain future scope. Code, a passing slice or an attractive screen alone cannot establish completion.

Given facts

• Rust/Iced desktop and a deployable REST service serve system engineers. OA/SA transitions in both directions must be understandable, explicit and useful.
• The revised goal explicitly requires constraints/calculations, quantities/units, reusable model libraries, state/behavior modeling and strong standardized specification/API, as well as textual/graphical notation and artifact traceability.
• Human/AI/Git-friendly source and a concrete unambiguous reST specification must match graphical editing. The source editor needs highlighting/linting; help must be offline, concise and navigable.
• Direct behavior/scenario diagram authoring and reduced cognitive load are current priorities. Long homogeneous selectors and bloated persistent creation controls are rejected. Study actual Capella's useful domain workflows and nesting without mechanically copying its interaction problems.
• The user authorizes autonomous bounded engineering, native Capella/Studio evaluation and explicit independent delegation, exclusively with GPT-6 Astra at effort appropriate to the question. Root owns synthesis/integration.
• Project instructions require Gitea coordination, ownership preservation and durable handoffs. Current root reserves one Cargo/native operator per host. Do not infer restart, usage reset or credential disclosure authorization.

Verified facts

• Current local source is aa40345 on codex/behavior-authoring-20260910, following baseline 1797516. Local commits cover B1/B2/B3 authoring, nullable history cache (b6b76ef), external palette integration (aed40d7), cfg(test) lifecycle projection (eb5161b), history seed correction (4ecad78) and exact-identity native oracle. Latest aa40345 commits the requirements audit and an unused-import correction prompted by the active compiler.
• This fresh host has task-local Rust/Cargo 1.98.1 GNU/LLVM and verified LLVM-MinGW. The one-job desktop test no-run build is active, with no local Rust build/test or Studio native pass yet. The history suite has thirteen tests; lifecycle has ten, both awaiting local execution. Progress and evidence.
• Strict language HTML regeneration and the source7 OpenAPI checker passed locally: 72 fixtures, 204 schemas and 90 scalar leaves. Actual isolated Capella scenario, activity/lifeline and LOOP creation was saved/inspected; this is reference study, not Studio acceptance. Native benchmark.
• Core/source7 already implements substantial direct behavior, typed data/ports, contextual chain links, maintained message-operation identity, semantic reviews, artifacts, hosted source envelopes and recovery. Historical absence claims must be checked against code and newer scoped evidence, not copied forward.
• Requirements audit: Constraint and other expression strings are descriptive, with no calculation/binding/result evaluator; unit strings persist/inherit without dimensional compatibility or conversion; named types/sequence/scenario reuse and includes exist, but there is no maintained cross-project package/version/instance/update contract. These are explicit gaps.
• The same audit identifies concrete unbounded Constraint target choices and generic expanded collection editors. Explorer already has hierarchy/search/paging; the problem is targeted, not a claim that all selection is flat.
• Gitea #79 distinguishes a B3 retarget evaluator setup error from the real same-selected-ID/visible-owner navigation defect. The external scenario owner corrects owner restoration and pending-draft guards. External layout 8/8 passes precede subsequent edits and do not establish current local/native acceptance.
• Codex-9e8r6t is a separate host with independent Cargo/native leases. It owns scenario workspace/B3/local behavior creation, palette and platform/packaging; current integration is not yet imported. Both hosts report direct push auth unavailable. MCP/anonymous fetch and verified bundle exchange remain usable.
• Current Linux acceptance is open; this host has no installed WSL distribution. Historical Linux core/server/cross-build results have exact older snapshot scope. The earlier denied guest workaround did not run and must not be bypassed.

Assumptions

• Compact contextual tools and an adequately sized scenario workspace can reduce friction; only actual engineering tasks at minimum/wide sizes can validate this.
• A bounded pure analysis model and pinned reusable packages may satisfy the new requirements without full language/runtime adoption. This is a candidate, not a selected architecture or a permission to treat metadata as semantics.
• Existing shared core/source/history boundaries remain suitable for additions; foundational examples and independent criticism must test that assumption.

Open questions

• What precise owned-value, calculation, constraint-result, quantity-kind, unit and numerical contracts meet mass/power/latency and affine/dimensional edge cases?
• Which library definition/instance/pinning/binding/update design preserves stable identities, local engineering edits and offline reproducibility across projects?
• What migration and OA↔SA treatment follows from those contracts without guessing executable meaning from old strings or silently changing old models?
• Do current real gestures, exact caller return, nested scopes and lifecycle markers remain usable and correct under minimum sizing, scrolling, zoom and dirty drafts?
• Which uncovered teaching, traceability, transition, saved-view, state/data, hosted and platform cases remain after current-binary acceptance? What supported Linux environment can complete the remaining actual runtime gate?
• Does "standardized" require any external conformance beyond the project's precise interoperable spec/API? Full OMG conformance is not inferred merely from comparison to SysML v2; explicit claims require a mapped and verified contract.

Candidate hypotheses / designs

The requirements delta compares metadata/ snippets, bounded pure engineering analysis with maintained pinned libraries, external evaluators, and full standards adoption. Metadata alone fails the explicit new semantics; external tools alone still need binding/result contracts. The report recommends the bounded route for root synthesis, not as a ratified decision.

Independent maximum-reasoning tracks now compare owned engineering properties versus analysis contexts; calculation definitions/applications; numeric/unit rules; and copy, linked-definition, live-instance or pinned-definition/editable-instance library architectures. Require concrete falsifiers, source/API migration effects, identity/history/recovery behavior and independent critique before convergence.

Decisions retained

• Keep three Rust crates and one authoritative typed Project with validated atomic Commands. Separate semantic identity/ownership, allocation/reference/order and diagram geometry. Forms are closed adapters, not a second metamodel. New persistent semantics require core/source/UI/API/reference/transition parity. Decisions.
• Preserve authored source ownership/comments/includes, exact identity and accepted model/source history. Source writeback must succeed before adoption. Raw invalid, excluded or rejected drafts survive; navigation/tools must not implicitly accept another editor. Recovery validates/replans rather than trusting serialized commands.
• Semantic reconciliation uses finite typed facets, frozen baselines, explicit mappings, actual destination slots and a single atomic candidate. Do not infer identity by name or apply reviewed rows as sequential independent edits. Foundation, artifact contract.
• Layer transitions are explicit engineering decisions, not automatic propagation. Behavior-root navigation must establish the requested owner while preserving unrelated architecture presentation. Ports/allocation and contextual occurrences retain their distinct domain meaning. Source7 contract.
• Hosted revision/source envelopes and ETag checks remain atomic; credentials stay out of persisted models and repository. Linux journal publication retains the selected no-replace invariant, with measured limits, not an overwrite fallback. Journal decision.
• Repair/evaluate the compact palette informed by native Capella evidence; do not return to the rejected horizontal toolbar. Lifecycle projection remains isolated under cfg(test) until coordinated Scene integration and actual acceptance.
• Preserve historical scope decisions as history. In particular, architecture-only SVG is implemented and behavior SVG was deferred; that is not proof of satisfying the revised release objective. Reassess older exclusions against current scope.

Current strategy

Finish the active local build and exact regression/native loop while separate agents investigate uncertain foundations. Integrate only coherent, verified external checkpoints respecting ownership. Use independently constructed complete model/source oracles and exact snapshot comparison for identity/history. Ratify analysis/library contracts through alternatives and falsifying examples, then implement shared semantics and ergonomic authoring together. Close current cross-platform/service/package and independent end-state gates before handoff.

The archived Task Ledger and archived STATE preserve all older claims, decisions and evidence links exactly. Relative report links inside those byte-preserved files resolve from their original harness/ context. Their historical test totals and process holds are not live status.