ARCADIA-Studio/harness/release-windows-gap-review-ytr83r-20260911.md

Windows release gap review — 2026-09-11

Reviewer: Codex-ytr83r /root/windows_release_gap_review. Bounded read-only review for this host's root; only this new report is owned. No production edits, compiler/tests/native processes, Git mutation or Linux work. Read AGENTS, the complete current PROJECT_GOAL, STATE/TASK_LEDGER/PLAN, Gitea issues 1/2 through comment362, open PRs (none), and current Home, analysis, library, canvas, contextual-source and server Wiki handoffs. Root's supplied production freeze is 29084050a7c020e0fb488be37c6e1ab235769cc0; later peer work is explicitly separate.

Verdict: finish the bounded Windows native gate, then concentrate on complete engineering tasks and the missing analysis/library capabilities. First-release completion is not supported. More generic release audits, environment preparation or repeated evidence inventories would have lower value now. The user explicitly paused Linux setup/build/checking until they resolve networking and instruct resumption (owner350); Linux remains an eventual requirement, not an active next action or a blocker to Windows work.

What exists, and what the evidence actually establishes

Class	Finding and exact evidence	Limit
Integrated, automated verification	Core all-target gate reports 433 tests, exit0, one optional ignored probe in experiments/lifecycle-native-ytr83r-20260911/core-gate.json; that probe was separately run per STATE. This core gate records source tree a9c71e2c3fc9260d06f191c0d65b1e63e41edbec, before the desktop-only corrections. Reviewer read current desktop raw logs: lifecycle 22/22, and full desktop 675/675, zero ignored/failed, terminal0 at15:32:41UTC in desktop-full-api-correction-status.json.	The22 are included in675, not additional tests. These gates do not establish native usability, current service transport, optimized packaging or peer source acceptance. Root's GUI build95641 was active at assignment; no reviewer launch or build-result inference.
Integrated capabilities, earlier scoped native evidence	OA/SA architecture, explicit allocation/containment, data/interfaces/ports, state/behavior, traces/artifact links, bidirectional reviewed transitions, source-preserving atomic commands/history, named behavior creation, clicked scenario slots and caller return are present. STATE links exact earlier palette, behavior and creation reports. The source7 contract and REST share typed Project/Commands.	Earlier native passes retain their binary/host attribution. Do not revive historical claims that these whole capabilities are absent; do not promote their existence into full current engineering-story acceptance.
Implemented, current-host native unverified	B4 lifetime bounds, conditional Delete activation clipping, incomplete lanes, markers/self messages and dense text have reviewed source plus focused tests (release-lifecycle-render-ytr83r-20260911.md). Twelve independent OA/SA semantic scenarios and exact source/name predictions exist in fixtures/scenario-lifecycle-native/README.md and lifecycle-native-contract.rs. Integrated named creation/history uses the corrected creation v2 oracle imported via b853a1e.	Real draw/hit/resize/Fit/selection and source-preserving Undo/Redo/reopen still require root's current binary. Oracle admission is preparation, not UI acceptance. No additional oracle framework is needed before attempting these cases.
Missing production capability	crates/oa-core/src/model.rs:651 stores Constraint expression as String; docs/language.rst:227 explicitly makes expressions descriptive, and :976 says unit strings neither convert values nor establish dimensional compatibility. Project at model.rs:1621 contains ordinary model collections, not maintained library/analysis environments.	This does not satisfy the current goal's calculations, quantities/units and reusable libraries (PROJECT_GOAL.md:651). Existing constraints, unit labels and source includes are useful foundations, not implementations of those outcomes.
Implemented text foundations, missing/unmeasured task support	Source7 alias/UUID identity, modular bundles, atomic Apply, highlighting/lint, source CLI and literal current-buffer navigation exist (source_navigation.rs:24 uses source.match_indices). The40-identity source experiment and26-request REST exercise are linked in STATE; the text usability audit identifies cross-file context, references and safe rename as the significant burden.	No large-model human/AI authoring acceptance. Navigation prototype does not yet deliver a complete production definition/reference/refactor workflow; code admission/capacity limits cannot prove practical usability. Optional folding/completion should be justified by task failures, not assumed mandatory features.
Current-host release evidence unverified	Independent server review found no concrete P0/P1/P2 persistence/API defect; it checked24 route paths/489 schema references and existing concurrency/restart tests (release-server-independent-ytr83r-20260911.md). The complete teaching example already has explicit OA→SA replay inputs and core/source checks (examples/release1/README.md:33,75). Packaging checks/notices and published Studio/CLI prereleases exist in Wiki Home.	Current service execution, exact frozen optimized Windows desktop/server packages, clean launch, complete integrated teaching workflow and final cross-disciplinary review remain. These are evidence gates, not new proof of broken service code. Published previews do not close the current first release.

Peer-owned work that must not be duplicated

• Codex-956atf: chooser and shared behavior canvas. Issue2 owner clarification 250 permits domain-constrained movement; arbitrary dragging is not the acceptance criterion. Corrected phase1 source 360 offers a shared viewport/palette/details surface, separate camera lifetime, actual focus controls and prepared task/pixel gates, but was uncompiled/unaccepted and independently under review. It explicitly leaves authored placement/full body/endpoint/fragment gestures and broader editing to later phases. The peer's prior full gate was716/720 with four actual failures; corrections are in progress (354). Obtain a corrected coherent transfer, including form-session reset and focus regressions. Preserve B4 pan+scale on both clip rectangles and clipped child text during integration.
• Codex-956atf: schema8 analysis and libraries. Published static core cdebb49d894952a189b144cc52deaf444e91d343 has Windows28 analysis and26 numeric selected tests plus strict Clippy, following a corrected real stack overflow. AnalysisCoreSchema820260911 still lists codec/lifecycle, source8, evaluator, command/REST/CLI/GUI, history/transition and libraries open. The library compatibility candidate's120 assertions are an isolated prototype, with symbolic IDs/restricted evaluator/one package per import (LibraryCompatibility20260911). Do not create a competing metamodel or claim complete quantities from the static checker.
• Codex-9e8r6t: staged prerequisites, CLI and text navigation. GUI5657 source-card correction passed26 affected tests, strict Clippy/fmt/build; newest 356 reports native minimum-size visibility and parent/child recovery passed, cancellation/creation/history still continuing. This is beyond local imported creation evidence but not yet a coherent accepted transfer here. Bounded source-outline/query patches 359 were unapplied/uncompiled and are under independent review362. Source CLI01b76a1 is already imported as023e3a8; do not reimport it.

Five prioritized next actions

1. Close this freeze's Windows native behavior gate — owner ytr83r root, dependency GUI95641 terminal success and immutable binary. Execute the already prepared lifecycle cases at minimum/wide size, plus bounded named creation/history/recovery with the corrected v2 oracle. Require visible correct Create/Delete bounds, branch-local activation, self-marker/stroke selection and full accepted Project/reachable-source equality across edit/Undo/Redo/reopen. Fix real failures; stop extending renderer/test infrastructure once the defined cases pass. This turns the considerable existing implementation into usable accepted work.
2. Integrate one coherent peer authoring slice and complete an engineer's OA→SA story — root integration with956atf/9e8r6t handoffs. Once peers publish reviewed corrected source, use their combined canvas/chooser/staged prerequisites to create missing actors/work in place, allocate explicitly, build/reuse process+scenario, return to the caller, add state behavior, review transition/interfaces and link an artifact. Continue the same project through text and conditional REST edit, stale-write refusal and reopen/history. Use existing release1 lesson/replay inputs; record semantic outcome, discoverability/context switches, keyboard operation and minimum-window visibility. Phase1 pan/zoom alone cannot close issue2. GUI guards must preserve raw drafts rather than implicitly accepting them.
3. Drive the existing analysis/library owner toward a complete usable vertical slice —956atf semantic/core owner; root owns eventual integration/acceptance. Keep codec/resource/lifecycle work bounded to protecting the actual contract, then deliver an engineering calculation with exact quantity kind/unit rules, property baseline, explicit case override and a constraint result distinguishable from missing/error. Extend it through ordinary source/API/GUI/save/history, then reuse pinned definitions in two projects and explicitly update/diverge/detach with exact provenance/rollback. This acceptance slice tests architectural coherence; it is not a reduction of the full selected contract. Source8 and library semantics must converge before broad migration/integration; do not parallel-design a replacement here.
4. Close practical text-only tasks using the coordinated navigation work —9e8r6t owner; dependency reviewed/executed outline seam and production adapter. Run the already proposed small-model task and100-module/1,000-declaration repeated-name task. Locate a referenced definition and uses, rename an alias and move its declaration while retaining UUIDs/comments/unrelated drafts; preview and apply one atomic bundle with stale-base refusal. Measure task errors and navigation/latency. Keep alias rename, display-name rename and identity change distinct. The outline is lexical evidence with scope/revision/ambiguity, not proof of all typed dependencies. Use findings to select the next UI/CLI improvement rather than starting new language research.
5. Publish a meaningful Windows prerelease, then retain final-release gates —ytr83r release owner, coordinated with9e8r6t's existing authoring releases. After the accepted current behavior milestone, build and validate matching optimized Windows desktop/server artifacts, notices/hashes and clean launch; exercise actual HTTP/source/ETag/restart on that build, and publish precise limitations. Do not hold this usable prerelease for all future semantics, but do not label it feature-complete. Refresh guide/reST against accepted controls during the engineering story and perform the final cross-disciplinary review when all required capability slices are realized. Linux remains explicitly deferred until the user resumes it.

Focus correction for root

The project is measurably progressing: integrated tests pass, peer source closes identified native workflow failures, and analysis has moved beyond proposals into a reviewed static core. The main risk is optimizing evidence volume rather than task completion. Prior setup and exact provenance were justified where they enabled execution; repeatedly auditing archived hashes or expanding OS diagnosis now will not improve the user's modeling work. Current B4 native acceptance is valuable because it closes an implemented user-facing gap. After it, judge each work block by a new engineer-visible outcome, a resolved semantic contract blocker or a falsified material defect.

Keep this report as a bounded handoff, not a new authoritative ledger. Root should update the existing short STATE/PLAN at its normal checkpoint: PLAN immediate step7 still proposes overlapping Linux preparation despite the newer explicit pause, and its desktop-active wording predates the now-read675/675 terminal result. These are small maintenance corrections; they should not displace actions1–4. No goal-complete, fresh native, peer full-suite, Linux or final package claim is made here.