Task Ledger — first release

Updated 2026-09-11 by Codex-9e8r6t /root. Full release not accepted.

User resumed after enabling virtualization and explicitly authorized installing Ubuntu26.04 in WSL2. RESTART_HANDOFF.md is the historical pause; STATE.md owns current capacity, compiler and native status. User also explicitly permits disk cleanup including caches. Both scoped cache deletion commands were rejected before execution; nondeleting compression recovered space with all282 generated-binary hashes unchanged. Ordinary Ubuntu installation and an initial actual WSL2 root shell passed. Current VirtioProxy startups fail before guest execution, and a distinct WslService restart was denied by OS permissions. Owner328 remains the newest WSL solution after the latest issue check through348; its exact configuration is already applied. Five Rust components and60 apt packages are downloaded and hash-verified; guest signature verification/install and application acceptance remain pending. The user's latest direction pauses Linux work until they notify us that WSL networking is resolved. Continue Windows/product work; preserve existing Linux configuration, verified inputs and evidence without further setup/checks.

Goal and success criteria

PROJECT_GOAL.md, copied from attachment 815ff7b9-73c4-41a2-a809-635ecf8389d6/goal-objective.md, is authoritative. Deliver a usable, substantially complete OA/SA systems modeling product in Rust/Iced on Windows and Linux, with equivalent graphical/textual/REST editing, strict reST semantics/API, arbitrary development-artifact traceability and navigable offline help. Constraints/calculations, quantities/units and reusable model libraries and complete uncluttered text-only small/large modeling are explicit requirements. Later Arcadia layers remain future releases. Completion requires real engineering tasks, sound semantics, relevant passing checks and fresh independent cross-disciplinary review.

Given facts

User prioritizes Capella's domain capability and useful visual nesting while rejecting clutter, intrusive creation rows, static behavior diagrams and long homogeneous option lists. Reduce cognitive load in small and large models.
Source must be understandable to engineers and AI, work with Git and preserve full parity. OA/SA transitions in both directions must be clear and reviewed.
User authorizes autonomous reversible work, research, native Computer Use and parallel agents. Use GPT-6 Astra exclusively with task-appropriate effort.
Gitea coordination and shared Wiki updates are required. Identify host and worker consistently and preserve overlapping work. Explicit project messages are authorized; unrelated external messages are not.
Updated AGENTS requires runnable versioned milestone prereleases. User issue 1 comment 178 authorizes reusing existing MCP credentials programmatically for Git push and REST asset upload, without printing or committing credentials.
Remote AGENTS.md, blob 4b8465c8993107062ad2974c15a958d449530e92, adds concise harness/Gitea maintenance and preservation of useful historical knowledge (user comment 229). Root's instruction sync 07eb3cd is integrated; contextual source df6f595, CLI publication d95b1ab and OA native evidence c4bad4b are committed and pushed.

Verified facts

STATE.md records exact commits, live leases and scoped verification. The current model has one typed core, validated atomic commands and coordinated Project/SourceBundle history shared by desktop and REST. Included files share one global namespace. Current Constraint expressions are strings; numeric unit strings do not convert values or establish dimensional compatibility. There is no locally integrated package/version/instance/update contract for cross-project libraries; peer implementation is tracked separately below.

The main basic/transition/history/reopen story passed exact independent model and four/six-source checks; supplementary native cases remain. Named creation passed 7/8/9-source oracles, one Undo/Redo and byte-exact reopen at actual minimum size. Cancellation retained its six-file baseline; SA draft recovery retained fields/identities, autofocus and Escape/Save As behavior. The D4D9 blank-canvas finding motivated the newer first-action implementation. Detailed historical suite and native scopes remain in STATE.md and its linked reports. Mounted geometry and bounded package passes do not establish full release acceptance. This local host has installed Ubuntu26.04 but lacks a currently usable guest session, installed build prerequisites or Linux application pass.

Current grammar supports omitted initial IDs/defaults, lexical ownership, forward references and preserved source. The earlier desktop audit found no exposed project-wide reference navigation/alias refactor; its Find searched the active buffer. The later CLI supplies typed alias rename. The audited process example had 128 lines, 14 UUIDs and 31 empty-default lines, while meaningful shorter source is possible. Two concise snippets passed 26 real REST requests, including identity-only normalization, exact name-only edits, rejected writes and restart persistence. Counts and this small trial prove neither improved native usability nor large-project performance.

CLI tooling is implemented, integrated and published: semantic source 01b76a1, packaging assembly a0a48e8, release integration 27df4a3. Author 12 tests and independent 183 semantic assertions pass; independent clean-package review adds 156 assertions across 10 processes. This is a bounded source-editing tool, not proof of complete large-model language ergonomics.

GUI/server release 1 and standalone CLI release 2 are published with runnable Windows artifacts, full dependency notices and downloaded asset verification. Independent archive/source and clean runtime checks pass in their declared scopes. See CLI publication evidence. The newer 13-file contextual-source slice is committed/pushed as df6f595, with its independent oracle and exact raw evidence. It passed focused 23, guide 3, strict Clippy, matching GUI build and formatting. The full 675 suite covers only its predecessor; no full 676 suite has run. Root launched the exact 52DE82 GUI in isolation: Lifeline Save-adoption and Chain direct Create pass complete source, one Undo and one Redo. Machine direct Create and one Undo pass; Redo/reopen are pending for Redo. An isolated postboot real Open/SA view/Save passed the complete machine prediction and exact six-file equality; the app closed normally. Later layout/caption source was independently reviewed. Postboot focused3 ran25 passing tests and one overflowing-fixture assertion failure. The stronger test-only fixture is frozen; its execution and new GUI remain pending.

Local OA B2 now passes all four Action/State/explicitly bound Reuse/Fragment Create, one Undo/Save, one Redo/Save and final OA reopen/Save against independent complete model/two-source predictions and original exact identities. SA B2 is pending locally. Existing B3 OA wide/minimum and SA minimum return scopes remain unchanged; neither these passes nor mounted geometry tests imply broader acceptance.

Assumptions

Contextual named creation, remaining-height canvas and local source insertion can reduce actual authoring detours. Native engineer tasks must establish this.
A bounded pure calculation/quantity model and pinned library content may fit the existing typed core better than a general-purpose metamodel rewrite. Peer-ratified core staging must still be checked against independent engineering tasks and integrated across all required representations.
Existing OA/SA semantics provide a useful foundation; neither their presence nor old acceptance proves full conformance to the updated release objective.

Open questions

Which precise calculation, quantity-kind/unit and library contracts satisfy practical budgets, reuse and update workflows without ambiguous defaults or implicit execution? How should they migrate and participate in transitions?
Which behavior workflows still force cancellation/navigation to create missing context, and which local actions best remove those detours?
Which text-only assistance best preserves engineering context and identity: native language services, external-editor services, source recipes or explicit reviewed patch operations? Which grammar changes are actually necessary?
Does the new scenario pane remain readable and editable through actual native window sizes, camera changes, nested fragments and reopened caller contexts?
Which viable Linux environment can run current GUI/server acceptance? Current VM limitation does not block remaining Windows, semantics or source work.

Candidate designs and decisions

Behavior creation: compact New → three task choices → focused name/owner/ viewpoint card, explicit missing inputs, atomic creation and reserved recovery IDs. Implemented, reviewed and scoped native creation/history/reopen passed.
Contextual prerequisite creation: stage an explicit typed source inside the current parent draft, then validate/adopt source and consumer atomically. This preserves intent and one-step history; a separate Architecture excursion or early source commit loses those guarantees. Initial Lifeline/process Work/ machine-owner adapters and semantic first-action prompts are implemented, reviewed and integrated with focused gates passed; scoped native results are above, with remaining acceptance still open. See the bounded plan.
Scenario workspace: compact header, optional Tools/Selection shelves, actual remaining-height canvas/form, pinned actions, independent scrolling and explicit Fit/Focus. Mounted tests support this choice; native proof pending.
User issue2/comment250 clarifies that diagram manipulation may respect sequence/domain constraints. Codex-956atf owns a shared canvas-shell redesign; the existing controls and supporting source-card fixes do not close the issue.
Quantities/calculations: peer comment 231 records ratified schema 8/local definitions and actual shared analysis/core types under implementation in isolated codex/analysis-core-20260911. This is not a local schema admission, evaluator or graphical/source/API parity pass; the local core remains descriptive.
Libraries: compare copied source, live references, generic definition/usages and pinned typed packages with explicit mappings and reviewed updates. Falsification brief records identity, dependency, offline and conflict requirements. Refined peer baseline, Case, full-freeze and exact-pin/equivalence contracts are ratified externally; their local integration and executable lifecycle acceptance remain pending.
Text authoring: use existing omission/default/identity contracts for concise experiments; compare source navigation/refactoring/patch assistance before grammar redesign. No namespace, hidden identity or omission change ratified.
Reference chooser: peer comment 224's corrected focus ownership, nested-scroll reveal and mounted regressions are prepared, not executed. Preserve that ownership and wait for its source/test handoff.

Long-lived decisions remain in DECISIONS.md. New user scope and normative contracts supersede older prototype exclusions.

Current strategy

Finish coherent creation gates and test the rebuilt product against independent saved-source oracles. Close remaining scenario lifecycle integration boundaries. Use the native Capella study and independent criticism to address real authoring friction. In parallel, resolve foundational semantic additions and text-only authoring through bounded competing designs and executable experiments. Then implement all representations, verify real REST and Windows/Linux workflows, build actual optimized packages and conduct fresh end-state review. Do not weaken acceptance to match current implementation or stop because a narrow technical milestone passed.

Pre-consolidation content is preserved byte-for-byte in the dated archive.

12 KiB Raw Blame History