Match compact loop-card presentation across project and overview

[coneilen]

Summary

Loop cards currently spend a row on raw state text, truncate live metadata, and use a different presentation in the global overview. This approved slice shares a compact 250x96 card body across project and overview, with a measured title/state row and separate monospaced live, detail and metadata rows.

Refs #764. Implements only the compact-card slice of the approved plan and corrections, not the complete issue.

Current head: 156345accb529384e200e1667e1945e857784cd1, based on actual accepted main 112d458303891f87af6c1dc6beaf7e48073fd65c after #825's runner budget, #824's lease-bound capture core and E's #816 context-menu/connector changes. All five B commits remain unchanged, and the actual old-to-new delta equals the accepted nine-path base delta. Sonnet5.5/high accepted the narrow App/menu/card/gate/runner interaction review with no coupled bug or fix. This base changes product/build inputs in App.zig and GraphContextMenu.zig; the old a44/base65194 binaries and build receipts are historical witnesses only, not current-head builds. No fresh local product build is authorized or performed. Fresh exact-head CI is green: seven successful workflows,24completed contexts(20SUCCESS/4configured SKIPPED), MERGEABLE/CLEAN/draft, after one authorized main-derived capture-core job rerun. Its original failure remains separately recorded; prior a44/e3ff results are not substituted.

Visual gate: NotExecuted(harness #826). No AFTER capture or paired-image acceptance exists. The operator now authorizes the coordinator to merge this accepted-code/exact-head-green-CI slice with visual verification explicitly deferred. This PR remains draft until the coordinator marks it ready; no completed readiness transition, merge, visual acceptance, issue-box closure or parity-ledger promotion is claimed. The shared capture core is blocked by #826, a journal check-then-open race. B's own UI/harness remains stopped; no product/zmx/native execution, capture child or new desktop lease is authorized. Existing images remain held.

The operator moved the fallback authorization to now, superseding the former23:00PDT cutoff because #826 cannot merge before then. The coordinator has accepted current156345/base112d source, narrow interaction review, exact-head green CI and the preserved evidence. The coordinator may now perform the readiness/merge transition for this slice without visual capture; B's owner is authorized only for this body-status update, not to mark ready or merge. Issue#764 remains open with box1 and all other visual boxes unticked, and visual verification is deferred to one later coordinated capture batch. This is authorization to proceed, not a claim that the PR has already been marked ready, merged or visually accepted. Required checks must remain green.

Changes

• Share the compact body on project and overview cards, with semibold title, uppercase state pill, monospaced rows, six-point type marker/five-point gap and source-derived 52-percent white metadata ink. Overview pitch is260 with a10-pixel gap, lane minimum806 and content width854.
• Extend the existing process-lifetime font cache with face/weight keys, not a second cache. Cards use Consolas for live/detail/meta; existing native-control APIs keep Segoe UI behavior. Cache capacity is512, borrowed handles are never evicted, and exhaustion reports/refuses without leaking an uncached font.
• Own complete Unicode strings; prefer reported live activity/pass prefix, then type-specific fallback. Metadata order is type, measured branch, non-default backend display name and real coarse age, not raw backend/tokens/model tier. Reserve fixed fields before Unicode-safe branch middle-elision; use measured ellipsis or omit only its complete visual segment if necessary. If fixed fields cannot fit, report the specific failure and show a bounded metadata-unavailable marker without blanking other rows. Metric deltas require two real samples and preserve direction.
• Keep existing project Reply/Inspect and Reclaim/Keep actions above metadata, sharing visibility/rectangles with paint, hit testing and UIA. Correct their existing amber COLORREF channels; add no overview actions or reclaim-policy change.
• Add focused geometry/allocation/font/DIB/Unicode and never-shown UIA-data regressions. Approved extensions update only named App geometry fixtures and gate/runner card geometry/action sites; merged A prompt-ID/hit helpers and G1/G2 runner children remain intact.

Scope and limits

Issue box1 is targeted but unchecked pending paired visual review. Boxes2-8 remain unchecked: type-palette sweep; entry/origin/chips; unwired inline recovery; authoritative attention/glow/actions; topology/initial-fit; ports/group band; add control.

The final B patch still touches only six approved files: GraphCanvas.zig, DesignTokens.zig, AppFont.zig, the bounded App.zig UIA/test blocks, uia-live-gate.ps1 geometry oracles and corresponding ValidationRunner.Tests.ps1 cases. No ledger, historical250x106 manifest, GraphModel/Swift/provider-pin/sidebar/workspace-paint/jump/dialog product changes. Quick Chat is unchanged. Existing START/UNWIRED information and context-menu recovery remain deliberate extras for follow-up.

The coordinator approved the narrow existing App test literals, gate geometry/early Open and Worktrees coordinates, independent row painting and minimal branch-elision policy after the initial rejection. At the actual1028-client gate boundary, the unpanned third card and Worktrees control extend four pixels past the viewport; its visible hit target remains reachable and pan-30 exposes it fully. This is disclosed clipping, not whole-control visibility. Overview row/lane heights remain unchanged; all production overview consumers use the shared bounds/caption helpers. Unrelated Quick Chat220/760 values and historical716 negative fixtures remain intentional.

Test plan

RED: & $env:GRAPHCODE_ZIG0152 test src\GraphCanvas.zig -target x86_64-windows-msvc -lc -luser32 -lgdi32 -IC:\gcp\ib\winghostty\include --cache-dir C:\gcs\ib\card-local --global-cache-dir C:\gcs\ib\global --test-filter "loop card parity" -> 0 passed,5failed; expected project height96 found106, overview width250 found220, goal displaced activity, raw backend/tokens/tier displaced display metadata, actions overlapped metadata. Original C:\gcs\ib\red-card.log retained.
GREEN: pwsh -NoProfile -File C:\gcs\ib\verify-cards.ps1 -Tag post821 -> historical product-source-equivalent B focus passed17card/8font/3hidden-App/38adjacent tests, each with positive counts; no current-head rerun implied. B patch unchanged but merged E changes product inputs, so historical builds/tests are not current build proof; old logs retained.
REGRESSION: pwsh -NoProfile -File Tools\windows\Tests\UiScenario.Tests.ps1; git range-diff 65194b7..a44ab90 112d458..156345a; gh pr checks 814 -R scgopi/GraphCode --watch --fail-fast --interval 30 -> direct offline G1 child166executed/166passed/0failed exit0; all five own commits '='; normalized B patch unchanged and actual nine-path delta exactly accepted main. Exact156345 CI7successful workflows/24completed checks(20SUCCESS/4configured SKIPPED), MERGEABLE/CLEAN; one authorized core-job rerun after separately preserved49/47/2 sharing violation. Product/build inputs changed; no fresh local build authorized or claimed.

Current-head preparation and execution limits

postE-range-diff.txt, postE-rebase-proof.json, rebase-diff-preE.patch, rebase-diff-postE.patch, accepted-postE-base.patch and delta-postE.patch preserve the exact rebase evidence. E's App worktree/menu targets, connector-drag gate and runner additions remain present; B's card geometry/actions and existing A/G1/G2 cases were not restored from older whole files. The only local test run for this refresh is the direct offline UiScenario.Tests.ps1 data child:166executed/166passed/0failed, exit0, preserved in postE-ui-scenario.log and postE-local-proof.json. Full ValidationRunner, capture-driver child, native/UI/product/zmx execution and fresh builds are not authorized and were not run. The approved body-evidence checker is scoped to this artifact.

Narrow Sonnet5.5/high review of the new delta's App/menu/card geometry and gate/runner interactions is ACCEPT. The reviewer independently verified the five equal commits, identical B patch, exact accepted base delta and clean merge-tree equality. E's right-edge/mid-height connector point remains clear of B's action rectangles at250x96, its named card selector cannot match the action names, menu oracles contain no card geometry, and overlapping App/gate/runner changes occupy disjoint hunks. No fix or scope expansion was needed. Original review is agent2b7cb6ff-8ffb-406e-8544-8fc40266da66 turn0, preserved verbatim in postE-interaction-review-turn0.raw.txt (SHA256BAEF3667F88333B3C455BE32AF30895AB41ADEE61DB1612AB29D2EF9BB4E6760) with separate provenance record; postE-interaction-review.json is only its parent-written summary. It did not run the local combined runner, build, CI or images. Its timeout-headroom opinion is a prediction, not measured current-head capacity evidence.

Fresh exact-head required CI is green. The first hardening attempt failed a main-derived capture-core test: run38679005657, Windows2025 job116239062659, UiCaptureDriver49executed/47passed/2failed. The actual primary error is a sharing violation opening process-journal.json, reported from Invoke-UiCapture.ps1:429 during the real argument-vector case; the dependent rehearsal report guard also failed. It is in the known #826 journal-concurrency family, but the underlying mechanism is not established and the error is not literally FileNotFound, a network failure or stdin inheritance failure. Full log ci-postE-hardening2025-attempt1.log (SHA256CC514B6094CA6DA243249138AF3C96532F5BCC76337772BCAE72D3235AD0A5B5), actual job/run metadata and the original failed watch snapshot are preserved.

The failure was reported before retry. With exact unchanged local/live156345 verified and the workflow completed, the first authorized failed-job rerun requested only job116239062659; request exit0. New run attempt2's Windows2025 job116240810455 passed, with actual hosted UiCaptureDriver49executed/49passed/0failed, UiScenario166/166/0 and final ValidationRunner PASS. Its full log ci-postE-hardening2025-attempt2.log (SHA2560E42EFBBF18A3826E1F51953AC9962E6DA3D7436DE691AA71339897390F9230D) remains separate from the original failure; other attempt2 metadata may reference inherited passed/skipped jobs, not new executions. The per-head budget is one used of at most two across all affected workflows, not two per workflow; there was no automatic retry or head/source reset. postE-ci-retry-history.json preserves the actual request, exception distinction and unused remaining budget. No core/pin/timeout/source fix was made.

All seven current-head workflows succeeded: Windows shell38679005687, Windows port38679005714, hardening38679005657 attempt2, DCO38679005634, TDD38679005708, macOS classifier38679005631, and Linux classifier38679005639. ci-postE-positive-proof.json (SHA2567292BD6B17FABD7343C469AA5FB8AF633C1CC285D89D232428186D2677DDB56E) binds exact head/base, positive seven-workflow/24-context snapshot,20SUCCESS/4configured SKIPPED, MERGEABLE/CLEAN/draft and retry1/2. Actual macOS/Linux builds and opt-in full/gated hardening were configured skips, not executed cross-platform evidence.

Actual current-head logs are preserved separately for unit shards0/1/2 (jobs116239072193/116239072213/116239072335), native integration116239072268, real packaging116239072303, Swift116239061989 and both hardening attempts. Hosted Swift's runner passed G1166/166/0 and capture-driver49/49/0 without rerun, then positive163XCTest cases with1skip/0failures; E's connector/menu cases also executed in hosted runner output. These hosted executions are not local product/native runs or screenshot proof. A later body-edit TDD context may add to the snapshot; all nonzero expected check results must remain complete before any authorized transition.

Historical a44 qualification — not current-head evidence

The smallest affected offline child ran, not the full local ValidationRunner, App/unit/integration suite, native compatibility preflight or desktop capture. g2-focused-proof.json records31executed/31passed/0failed and the real no-desktop supervisor/worker/harmless-EXE launch, UTC/refusal/failure-safe release/ownership controls. The final B scene component regression is mocked, produces zero images and is not physical/UI parity evidence.

The first invocation used C:\gcs\ib\g2-focused-offline, which the accepted guard intentionally rejects as preserved historical lane-root ancestry. Its31executed/13passed/18failed log/report remain intact. The unchanged command was rerun with an explicitly approved dedicated sibling root C:\gcs\ib-r1-a44, without changing source, expectations or guards. This is an invocation error, not a product finding, capture-slot failure or behavioral RED.

Rebase proof: g2-rebase-range-diff.txt, g2-rebase-equivalence.json, rebase-diff-pre823.patch, rebase-diff-post823.patch, delta-post823.patch and accepted-823.patch. The retained user-specified Sonnet5.5/high reviewer independently accepted five equal commits, the identical aggregate B patch, exact six-file accepted G2 delta, unchanged B/A/G1 runner hunks plus the new positive-count child, and zero product/provider/build-input delta. This is rebase preservation only, not a fresh broad source audit or image acceptance.

Historical exact-source build command: pwsh -NoProfile -File C:\gcs\ib\build-r2-binaries.ps1 -> exit0 for a44 and base65194, no GUI. The script exported immutable Git sources, built with pinned Zig0.15.2 and exact Winghostty6286560d pin in separate source/cache/global/output roots, and recorded each exact command, source/built-at SHA, source-tree/archive/toolchain/host-library/executable hashes and timestamps. Aggregate receipt: C:\gcs\ib-r2-a44\build-receipts.json, SHA256C245B38CB1700996BB5BD94CFF954330E0D56A81C0F601BFC604412EB6E1091B. Neither phase was built from current156345ac/base112d; no relabelling or current admission is implied.

Fresh phase	Exact source / built-at SHA	Executable SHA256	Immutable receipt / archive / log SHA256
AFTER	a44ab904757fda0861c2b334a0798c437066ec76, source treef2f3715ada6f21a0f729f763c07393cb7b701b49	84FDFCBDA4B3B2097C802C2119F6D937ACFBB57549E3302056512D99D0B5AC68	after\build-receipt.json:C5642D120F07F19A8CE066706EB2A83EEBA3E48304A9FB25364D5845E5614012; source.tar:189C179C54A05C4A6060E518555865180DA7A582A512F3C5A38A043EDD491B17; build.log:ADE848FC7ACE442CB98B40E3F4F54BFFBC5E40D37176FA7B8F305F7CC85D0D3F
BEFORE	65194b777523406cbbfaa7db4bcf0372119c5412, source tree9c692ff30653d3de9c144a7d4a9e78672b55e804	9D2A977F0B4F65FC2734CDB67EC37EA71C902E6DF75DAEAA77E269054D9745D0	before\build-receipt.json:0DAD15EE4241EB4F34825AE3ED18FD67AB35E0E93BB9CCDB213B6CD17294C072; source.tar:76DBE1DE1CB36C1C2A211109D07206C29C1EF5884F92B0D235EB7BF3589AADB6; build.log:FEDB7068FBBEBC35D129632DA73158107B8C24A870BD958082FB0F3A60193E64

All table paths are under C:\gcs\ib-r2-a44. Executables are in each phase's out\bin\graphcode-windows.exe; source roots are each phase's exported graphcode-windows, not Git repositories. The clean original B worktree remains the source repository. These84FD/9D2A binaries, as well as older C610/83688 binaries, are now immutable historical evidence only. No shared capture scripts, provider pins, source or job identity are changed by B.

Historical exact-a44 workflows all succeeded: Windows shell38658685033, Windows port38658685069, hardening38658684996, DCO38658685189, TDD38658685049, macOS classifier38658685184, and Linux classifier38658684924. ci-a44-positive-proof.json asserts seven exact-a44 completed-success workflows and24completed contexts(20SUCCESS/4configured SKIPPED), MERGEABLE/CLEAN/draft; ci-a44-runs.json, ci-a44-checks.json and ci-a44-watch.log retain the snapshot. Its later body-edit TDD added another same-head context without changing the original snapshot; none qualifies the current156345ac head.

Actual a44 logs are retained for unit shards0/1/2 (jobs116174532021/116174532065/116174532089), native integration116174532012, real packaging116174531958, and Swift116174542627. They were retrieved directly under distinct ci-a44-*.log names without replacing older logs. Actual macOS/Linux builds and opt-in full/gated hardening were configured skips, not executed cross-platform evidence. Historical green CI and offline launch controls do not establish current-head regression or paired visual acceptance.

Preserved product and regression history

All historical body versions and failures remain separate local artifacts; no failure log is overwritten by a passing run.

Evidence	Result and honest limit
Product RED and corrections	red-card.log:0pass/5fail before implementation. red-app-old-probe.log:old probe on compact geometry expected1 found0, a dependency regression, not proof main's old actions were invisible. red-review-composite.log:fixed-fields/branch fit. red-gate-geometry.log and red-worktrees-assignment.log:actual expected1005 found959. red-meta-ink.log:old ink failed literal source-derived COLORREF. Compiler mistakes remain separate, not behavioral RED.
Initial source/CI rejection	One substantive reviewer rejection for metadata-dependent blanking and old gate geometry, and one change-caused hosted integration failure on6961/run38592395839/job115950535486:Open click used old760 floor. Full log retained. Later predicted Worktrees failure did not occur and is not counted.
Product focused evidence	Historical17card/8font/3never-shown App/38adjacent positive tests; DIBs exercise both renderers, multiple DPI/zoom/translation cases, Unicode surrogate cuts, actual action fills, finite-extreme metrics and allocation failure. Not physical keyboard, COM invocation, real glyph parity or installed persistence. Existing source/fixture visual gate and five palette tests passed, not new screenshots.
A/G1 and e3ff	Full earlier focused ValidationRunner passed394multi-project(54positive/340negative),11prompt lookup/6hit,166G1/0failures. e3ff hosted Swift job corroborates those counts; terminal/contracts does not. Current local full runner was deliberately not rerun. e3ff7workflows/24contexts20SUCCESS4SKIPPED,CLEAN, later body-edit TDD context added. Actual new-head logs were reviewed by coordinator Sonnet5.5/high; acceptance covered evidence/body, not visuals.
Historical exact CI	d12,0831,de432,e3ff run/check/job logs and positive execution summaries remain in C:\gcs\ib. They are historical, not a44 gates. Skipped macOS/Linux builds and opt-in full/gated hardening are not executed cross-platform proof. Unit root summaries are not unique-test counts; auxiliary zero-test Swift output is not a pass.
Retrieval/extraction failures	ci-e3ff-log-download-attempt1.txt retains shared gh ZIP/EOF retrieval errors, interrupted Swift attempt2 and raw terminal-escape refusal; attempt3 direct-file retrieval succeeded. The first runner-summary extraction expected a space instead of equals; corrected extraction re-read the same completed log. Neither is a test/product failure.

Capture history and remaining image gate

Attempt	Outcome
B queue2,18:10:57-18:25:57UTC	First zmx argument order failed before GraphCode; second inherited-writer EOF wait failed before GraphCode. Third captured four BEFORE images then builtin card lookup found0; a peer race was suspected, not proved. No actions or AFTER. Seven owned processes to0; native File exits0 and post-cleanup EOF. Release18:24:20.002UTC, independently verified.
B queue5,19:34:02-19:49:02UTC	Four new BEFORE images; nested Get-CaptureActionsEvidence was not exported, plus caller exit$result typo. No actions or AFTER. Two File exits0, seed0, bounded EOF, seven owned processes to0; release19:36:19.147UTC independently verified. No retry after stop.
Central queue6	Command21:34:52 refused before GUI/PNG because UTC lease conversion added seven hours. Release/cleanup independently verified21:37; lost admission time recorded separately. No product finding.
Corrected central queue10	Empty mandatory ArgumentList @() refused before app/GUI. A secondary supervisor null-error suppressed its official release JSON. Seed0/ready/EOF existed; coordinator independently recovered and verified seven journaled generations absent/products0, release/guard cleared. This is independent recovery, not a cap official receipt. No PNG/AFTER/vision; old packet stopped.
G2 repair / R2 admission / #826	G2-R1 #823 and R2 #824 merged, followed by runner-budget #825 and E #816. The current shared journal check-then-open race remains open as #826; capture is blocked and no new clock/admission is granted. B does not repair the core, run its capture child, own a desktop slot or claim a new lease. No third old-packet attempt and no screenshot success inferred from offline31/31.

Queue2 assets remain at C:\gcs\ib\captures\attempt3-before with preserved hashes; queue5 at captures\retry-final\before, indexed by queue5-partial-hashes.json/queue5-scenes.json. Release/log manifests and full identical-source Unicode fixture BFC26149... remain separate. All eight images are BEFORE only, not AFTER or paired evidence. Historical phaseFailed=false described EOF drain only; the phase failed. The corrected seedDrainFailed field also does not imply whole-phase success.

Primary read-only macOS source is 2ccf303a04f995184aae1d4b738a9eca0f71b79e, bundle C:\gc-ref\macos-groundtruth-2ccf303a. Its fitted project reference is ParityRef90% (metadata wrongly says global); restored Graph overview is45%; these are not intrinsic100% font/card measurements. The read-only macos-groundtruth-2ccf303a-gapfill-v2 companion adds independently sampled100% linked project/overview references, not a plan/scene expansion or healthy-provider/physical proof. README/index/privacy masks/exclusions precede image access;167unsampled companion images are not independently cleared. Excluded764-fit-targeted-event and held722menus are not used. No reference bundle or screenshots have been uploaded wholesale.

Original body artifacts remain unchanged: pr-body.md SHA256F22E967B110BD8D52158B06CD9CA5A3E36BCC0271D6221D936FE51E8E170D7CD, pr-body-centralcap-pending.md SHA2561E4F61788CADDA50CD507733019C87D7E69CC2C5D578C6CC60B0F7AFB53B1A52, pr-body-g1-green.md SHA2567B0FEB35169B0100EE3888B2C31D93B6B801BA7D81AC6C1AC7FF7918691DED65, pr-body-g2-green.md SHA2569DF35577694BD83B3FAF9583D09CB4D94A5D112CA0F5DFD8ABC5D45833E66590, and pr-body-post826-hold.md SHA256F47DEF7302E8E0F1B2F709305808CFBA0AFF99174F1471C98630229E9D473818. Historical job/build hashes and source/CI acceptances remain valid only for their named scope. Visual acceptance remains unexecuted; the currently authorized coordinator-only semantic merge must retain the open visual issue boxes and #826 reason, followed by one later coordinated visual batch.

Checklist

• I have read the Contributing Guidelines
• I have signed off my commits (git commit -s) per the DCO
• Tests pass locally (make test)
• Code follows the existing style (make check)
• I added the test/contract before the implementation and observed the intended RED failure

make test/make check are macOS entry points and were not run on this Windows worker. Exact Windows evidence and its limits are listed above.