Repository navigation
Conversation
🧪 Comprehensive Test Suite
Full-stack smoke gate runs in the CI workflow. |
🧪 Comprehensive Test Suite
Full-stack smoke gate runs in the CI workflow. |
mvalancy
left a comment
mvalancy
left a comment
There was a problem hiding this comment.
Independent source review of 2d81ac0279156ca6ca41761ef9c1a1a79fcbe31b against 4b825b12a752f54531a2263816630501396eabb8: two P2 findings block acceptance of this draft source.
- Organize receives a truncated simulation after a low-zoom rebuild. The initializer’s local transform is a D3
ZoomTransformwithk, and its existing filter can omit offscreen loaded items. The new action only arranges/savessimulation.nodes(). - Persisted Hierarchical mode can undo Organize. Deleting per-node layer pins leaves the persistent mode and its pending/reload timer active; that later replaces the radial arrangement with dependency rows.
I read the complete 15-path change, issue233, new lesson/architecture guidance and the relevant renderer/save/mode lifecycle. I independently ran the seven new pure topology/overview tests on exact-source copies with one worker: 7 passed, 0 skipped, 643ms. Connectivity deduplication, deterministic component/ring ordering, the 40,000-edge refinement budget, bounded overview swaps, truthful hidden/deferred IDs and the new clearance measurements look consistent with their stated pure contracts. The new agent note/index do not change integration authority. Browser sequence enrollment preserves the existing workers/retry/timeout policy.
The two lifecycle probes execute exact source blocks with the actual pure layout helpers and lightweight dimension/simulation adapters; they are not browser reproductions. An initial delayed-edge hypothesis was withdrawn after examining the full render gate: while either query loads, the SVG is absent and initialization returns. Its isolated-block output is explicitly marked invalid as product-failure evidence in the retained scratch README. No correction is requested for that withdrawn hypothesis.
One existing boundary must remain honest in the new docs: initialization still calls separateCards on saved overlapping positions and persists that repair. That predates this PR, but it prevents an unconditional claim that saved positions remain unchanged until Organize. Preserve the disclosed repair exception or deliberately change and test the policy.
This is source review only. I did not run a browser, database, install, build or full local matrix. The documented phone readability correction, actual failure-before/fix-after lifecycle regressions, the detailed organizer’s scale acceptance, full hosted gates/screenshots and post-dev acceptance remain pending. The existing 2,000/5,000 browser fixture uses saved positions and does not by itself execute the new explicit detailed organizer. Radial placement is not completion of obstacle-aware item routing #220.
Read-only evidence: /tmp/graphdone-resolution-20261027/pr234-review-2d81ac0/ (pure-tests.log, exact-source copies, probe.cjs, lifecycle-probe.log, probe-limits.json, README.md). No source/index edits, pushes, merge or issue closure.
…tion-20261028 # Conflicts: # .agents/issues/INDEX.md
🧪 Comprehensive Test Suite
Full-stack smoke gate runs in the CI workflow. |
mvalancy
left a comment
mvalancy
left a comment
There was a problem hiding this comment.
Independent correction/composition review of immutable 97763da249c96598416ecb89b3bc38195c2a84b6: both original P2 lifecycle findings are corrected. No additional blocker found in those corrections.
The complete composition has parents 409d28a062426a0c1e84094760e676195565dc05 and accepted PR232 dev 17acc25dc98978c64f87a5b0d6fb36b1833cdfa7, tree 25452870c699072cd75b7e3950aa7d5b9d7651dd. Its diff from accepted dev stays within the 17 topology feature paths; the accepted bus interaction implementation is retained, the shared overview component adds topology placement/clearances, and the index retains both lesson entries. No integration authority changed.
I independently ran exact-source copies of the topology, overview and camera unit files: 9 passed / 0 skipped in 616ms, one worker. A small source-level lifecycle probe confirms complete simulation membership, all five IDs reaching organization, force mode stored and notified, zero applications from an already-pending hierarchical callback, and no hierarchical timer on reload. This probe is not browser execution. The saved-position repair exception is now explicitly documented.
I inspected the author’s retained actual 4-case report: 4 passed / 0 skipped / 0 flaky in 42.860s, including native low-zoom title editing and delayed authoritative edges. Its existing phone visual result remains rejected; a passing report does not override that usability finding. I also inspected the composed 301-item desktop report: 1 passed in 51.224s. That test covers its existing card geometry and bus interaction assertions; it does not call the new inspectSchematicOverview clearance oracle. The 64/20/16 measurements are enrolled in the nested and large-graph suites, so the 301 run alone is not quantitative clearance acceptance.
No final PR acceptance is granted here. This review excludes uncommitted phone changes, which require a new immutable review and actual visual evidence. Final-source hosted gates, large/nested quantitative clearance and scale evidence, and post-dev acceptance remain required. No local browser/database/install/build/full matrix or source/index mutation was performed. Evidence: /tmp/graphdone-resolution-20261027/pr234-review-97763da/ (lifecycle.cjs, lifecycle.log, pure-tests.log, complete composition.raw).
🧪 Comprehensive Test Suite
Full-stack smoke gate runs in the CI workflow. |
🧪 Comprehensive Test Suite
Full-stack smoke gate runs in the CI workflow. |
mvalancy
left a comment
mvalancy
left a comment
There was a problem hiding this comment.
Independent review of phone/paint/lifecycle source 8516bc59e4c84f36f2cffdf21f005677e4f1bd29 and its four-path geometry correction f5640464d1b85b651863abe9b3b36024cee1ffb7: the new P2 clearance defect is corrected; no additional blocker found in this reviewed chunk. This is not final PR acceptance.
The new finding was verified with actual production layout functions: compact slots used individual measured card sizes, then length-only refinement moved larger cards into smaller slots. A five-item fixture reduced minimum clearance from 64.000027px to 9.000027px. The corrected function rejects swaps that violate any component-card clearance, with separately counted/capped rectangle work. Independent replays now preserve 64.000018px and 64.000027px for the two retained fixtures; 13 exact-source pure tests pass in 666ms. No source mutation or browser run was needed for this proof.
The explicit-organize camera fix now marks the graph framed, which the existing initial-fit callback checks before doing work; it therefore stops that competing camera owner. Scope changes still reset organization framing, and explicit Zoom to fit removes the readable-scale floor. The save-progress attribute excludes only normal progress from permanent organized framing; attention/error states remain obstacles. The bounded phone angle/aspect search and stronger focus-priority scorer match their stated limited scope. They are not an obstacle-free guarantee for every large graph or transient overlay.
I inspected the retained topology-human-painted.json report: 4 passed, 0 skipped/flaky, 45.672s. I independently checked its desktop/phone geometry receipts and viewed all four organized/reloaded PNGs. All nine desktop and five phone cards are painted; complete fixture titles remain readable, and actual recorded screen bounds have zero delta between organized and reloaded captures. Original relations remain 8/4 with zero reported page errors. The new test requires 12px screen/control margins and repeats readability after camera framing, entrance opacity and persisted camera restoration. These are the retained author-run 8516-era visuals, not a new final-head hosted run or proof for heterogeneous cross-linked fixtures; those are covered here by the pure negative/correction.
The earlier 97763da phone clipping and faded/early reload captures remain rejected evidence. The docs correctly retain the existing saved-overlap repair exception, normal-progress/attention distinction, quantitative 301-run evidence correction, and #220 detailed-wire-routing limit. The correction changes only topology source/test and their two documents; no integration authority changed.
The uncommitted overview accounting chunk is outside this review. Final immutable composition review, actual quantitative overview/scale receipts, full hosted gates and post-dev acceptance remain required. No local browser, database, install, build, matrix, source/index edit, merge or issue closure was performed. Evidence: /tmp/graphdone-resolution-20261027/pr234-review-8516bc5/ including original production-function negative inputs/logs, fixed-replay.json, and fixed-pure-tests.log.
🧪 Comprehensive Test Suite
Full-stack smoke gate runs in the CI workflow. |
mvalancy
left a comment
mvalancy
left a comment
There was a problem hiding this comment.
Independent bounded UI/visual/accounting review of 6ba55f237963af0d5c9b4f6fe7899a1dc01ebc04.
I inspected the actual topology-phone-slot-output at-rest PNG and geometry receipt, read the exact slot-phase/test delta, and independently executed the previous/current pure placement code against the measured phone obstacles. The previous source offers 2 slots; this source offers 3, matching admission capacity, with unchanged inputs, 240×96 cards, 64px row gaps and at least 20px control clearance. Candidate phases are bounded to 32 and placement/capacity share the slot calculation.
The phone image is readable: complete cards at y342/502/662, 18px titles, visible directional arrows, and useful whitespace around controls. Its four drawn directed group pairs represent 45 original external relationships out of 361; the receipt retains 316 deferred plus 9 internal relationships, totaling all 370. All three displayed groups have a drawn connection; this does not mean all eight groups or all 38 eligible pairs are displayed. The visible denominator, “5 more in group list,” and “Browse all 370” make that limitation understandable. The receipt reports zero card intersections/unrelated crossings/overlaps/diagonals, with 20px unrelated-card and 96px unrelated-wire minimum clearance. The actual local report has 1 passing phone journey, no retries/skips/failures, 40.665s total.
I also previously inspected the retained desktop 301 overview and the corrected initial/reloaded desktop-nine/phone-five star screenshots. Those sampled frames are clean and readable; the earlier clipped/faded captures were rejected and are not acceptance evidence. The desktop coverage receipt represents 101/361 external relationships and 13/38 pairs, not complete wiring coverage.
No new blocker in this reviewed UI/accounting slice. This is not whole-PR approval, final-source hosted acceptance, proof of useful coverage for every graph, or proof that mixed detailed routing is spaghetti-free. Current composed-source 2,000-item/5,000-relationship, nested, runtime/full CI and post-dev evidence remain required. Detailed item-routing/bus work remains #220. No repository edits, browser runs or merge performed by this reviewer.
🧪 Comprehensive Test Suite
Full-stack smoke gate runs in the CI workflow. |
mvalancy
left a comment
mvalancy
left a comment
There was a problem hiding this comment.
Independent exact-source review of 6ba55f2 against accepted dev 17acc25: one new P2 minimap lifecycle finding posted inline; no integration clearance yet.
The 22-path composition retains the previously reviewed organization/lifecycle and compact-ring clearance corrections. No workflow authority or dependency changes were introduced. The new overview accounting partitions the same canonical edge projection into external/internal/pending counts; deferred plus priority-summarized plus actual drawn counts equal the external denominator. The 301 journey uses the separately queried original edge list, retains attempted-write rejection/final source-state checks, and now calls the quantitative clearance helper before inspection.
The slot search considers at most 32 deterministic phases and uses the same slot function for capacity and placement. Card size, row/column spacing and obstacle margins are preserved. I independently ran 29 exact-source semantic projection/slot, topology and camera cases: all passed in 876 ms with one worker. An initial scratch-copy setup omitted graphSymbolLayout and could not collect that suite; it was corrected before this result and is not product failure evidence.
I read the retained 0b desktop and 6ba phone geometry receipts and viewed their actual at-rest PNGs. Recomputing pairs from the actual SVG routes gives desktop 101/361 external links, 13/38 directed pairs, 8/8 displayed groups connected, card/wire clearance 20/24 px; phone 45/361, 4/38, 3/3, clearance 20/96 px. All routes are painted and both receipts report zero card intersections, unrelated crossings/overlaps or diagonals. The remaining originals are honestly available through Browse; these fixture results are not a universal coverage guarantee.
This is bounded source/local-evidence review. Hosted CI 40815091344 is still running, and the minimap correction plus any PR236 composition require renewed exact-source and actual final artifact review. No local browser, service, install or full matrix was started. Detailed item routing/crossings remain #220; no main, hosted or release readiness claim.
mvalancy
left a comment
mvalancy
left a comment
There was a problem hiding this comment.
Independent bounded re-review of exact 8698e92 (four-path delta from reviewed 6ba55f2): the minimap P2 is corrected; no new source finding in this delta.
The event calls only the existing position/type publisher after organization, preserves the closed-map reopen path, and removes its listener on effect cleanup. My exact-callback probe with the real d3-force loop confirms one immediate complete snapshot and no post-cleanup publication. I independently inspected the author-run fail-before/pass-after reports, organized JSON and actual PNG: old source fails the new coordinate equality predicate; corrected desktop passes 1/1 in 19.300 s, all nine coordinates match, and existing save/reload/paint/camera assertions remain intact. The 4.5 s wait deliberately crosses the existing finite 4 s startup polling window; it is not an interaction-timeout relaxation.
Full reviewed application composition remains based on accepted dev 17acc25. This is source-correction clearance only: the previous 6ba hosted run is superseded, fresh exact-head hosted gates/artifacts and the planned PR236 composition still require review. No local browser, service, install or full matrix was started for this review. Detailed routing #220 and previously recorded release/hosted limits remain open.
🧪 Comprehensive Test Suite
Full-stack smoke gate runs in the CI workflow. |
There was a problem hiding this comment.
Independent read-only review of the dense local-item LOD at commit16690861d4b0e08f907d477416d6c3da201380c6. Two verified state-transition defects block merge: overview/member-search inspection of a nearby dense item can reactivate the retained previous local focus (inline4408976990), and a live group removal can hide the only All-groups reset while retaining an unavailable filter (inline4408985956).
The pure model/layout review found no other verified defects. Its25focused tests passed, and a deterministic scratch probe passed2,120 projections across100directed graphs plus100obstacle layouts: disjoint original-ID accounting, directed/type/parallel preservation, painted endpoints, complete unique paging,20pxbounds/obstacle margins,64pxcard gaps and frozen-input no-mutation. My separate two-case React reproduction confirmed both defects in222ms.
Source review confirms projected x/y are separate display objects, original relationship selections resolve through the current index, owner+graph scopes remount/persist independently, and all four replaced native card/edge/arrow/label groups share presentation ownership. No local browser/database suite was run; the author owns the sole headed preview. The actual107-neighbor, editor/keyboard/no-write, responsive, final-source hosted2k/5k and post-dev gates remain required. This review does not accept an unpublished delta, approve or merge the PR, or claim general arbitrary item routing is complete.
🧪 Comprehensive Test Suite
Full-stack smoke gate runs in the CI workflow. |
mvalancy
left a comment
mvalancy
left a comment
There was a problem hiding this comment.
Independent read-only re-review at4ee512005df2eb6d35930b9c98bb11beb9e85a79. Both original blockers are fixed and their threads are safe to resolve: explicit overview/search navigation replaces a nearby retained focus without changing local Open behavior, and live missing-group filters retain an accessible selector/direct reset.
One new verified blocker remains (inline4409090329): the parent's consumed navigation request survives a keyed local-layer remount and replays over the latest persisted focus/page/group when returning to that graph. My scratch React probe confirmed persisted b0 before unmount becoming old request B/1 after remount in311ms.
Independently ran33focused LocalItemNeighborhood/NodeInspector/localItemLOD/localItemLayout tests:4files passed in4.56s; jsdom's existing canvas fallback warning is explicit. Source review found the new quick editor anchor uses the displayed original-card rectangle, and inspector/editor close paths prefer connected projected original triggers without writing display coordinates into canonical data. Compact card geometry and the bounded safe-area layout control pass. I did not run a browser/database matrix; actual107-neighbor paint/keyboard/editor/no-write, responsive screenshots, final hosted2k/5k and post-dev acceptance remain separate required evidence. No approval or merge performed.
mvalancy
changed the title
🧪 Comprehensive Test Suite
Full-stack smoke gate runs in the CI workflow. |
🧪 Comprehensive Test Suite
Full-stack smoke gate runs in the CI workflow. |
1 similar comment
🧪 Comprehensive Test Suite
Full-stack smoke gate runs in the CI workflow. |
🧪 Comprehensive Test Suite
Full-stack smoke gate runs in the CI workflow. |
🧪 Comprehensive Test Suite
Full-stack smoke gate runs in the CI workflow. |
There was a problem hiding this comment.
Independent read-only composition/source review at762a9a9f52c6fee5626e6a0269accf5796454531 against live dev479edf6 (runtime is identical to frozen108f2d3; I additionally checked the committed test callback type/lesson-only delta). The three local focus/filter/request findings and older delayed-Organize minimap finding are fixed; replies identify each thread as safe to resolve. Exact-source D3 replay confirms the minimap publishes once immediately after organization after startup polling has stopped, with positions equal to the live simulation.
One verified acceptance-guard defect remains blocking (inline4409148277): the new helper's native-layer selectors miss the actual .edge-clickable and .arrow classes. This is a false-positive path in the test, not evidence that current product CSS exposes those layers.
The extracted typed presentation components preserve the controller's view-only geometry and original-item callbacks; parent navigation intent is now acknowledged once, scope-filtered/reset and monotonically sequenced. The full152px cell envelope removes the page-dependent long-title clearance gap. Independently ran49focused local/model/inspector/save/selection controls in6.89s and50topology/overview/camera/selection controls in5.85s (the two batches overlap). The only working-tree delta during these runs was the author's equivalent NodeInspector test callback type annotation, now committed in762a9a9; runtime/source was frozen108 and is identical at762. The expected jsdom canvas fallback warning remains explicit.
Independent fixture reconstruction confirms5,000relationships, all500required reverse BLOCKS, unchanged ordinary target1008degree5/incident IDs, and dense1900degree111 after107added hub connections. Existing ordinary-detail checks and both five-second zoom limits remain unchanged. The new browser guards cover actual card/wire paint, original accounting, margins, API snapshots and position-write requests; they do not themselves snapshot live simulation x/y/fx/fy. No browser was launched by this reviewer. Final corrected-source CI, actual107-neighbor responsive artifacts, hosted2k/5k artifacts and post-dev acceptance remain required; no approval or merge performed.
mvalancy
left a comment
mvalancy
left a comment
There was a problem hiding this comment.
Independent final source/delta review at41c1fa015e16ae9e8cfe54c3336ff68100df29b6: no verified source blockers remain. All four local LOD findings and the delayed-Organize minimap finding are fixed, with replies stating that their threads are safe to resolve. Product runtime is byte-identical to the reviewed108f2d3 composition; the final delta strengthens tests/docs and corrects the storage-label expectation.
The corrected native exclusion guard uses all five actual classes, requires nonempty native inventory and has isolated hitbox/arrow negatives proving old selectors0 versus actual1 and rejection. Audited the owned51.0s three-case report:3expected,0unexpected,0flaky,0errors. Both desktop/phone receipts retain all108 live x/y/fx/fy records across paging, neighbor focus, Back, pan, reload and original editing. Independently viewed desktop/phone initial and phone edited PNGs: readable focus/real neighbors, clean routed buses and full-width phone footer; these bounded fixtures do not establish arbitrary detailed routing. The offscreen regression now waits for camera readiness and requires both complete painted original cards.
The shared zoom readiness correction admits positive one-dimensional extent only for stroked local wires; card visibility still requires both dimensions plus readable typography. The retained old-phone control fails at15seconds. Existing ordinary/dense hosted zoom assertions retain their five-second limits; the small journey records initialZoomMs without enforcing five seconds. Independent MobileStorageStatus/LocalItemNeighborhood tests pass9/9 in3.39s, following the prior49/50 overlapping source-control batches.
Final41c1fa0 browser rerun, exact-head hosted2,000-item/5,000-relationship CI/artifact audit and actual post-dev acceptance remain required. This is a source review and bounded local-artifact audit, not an approval, merge, scale acceptance or hosted-parity claim.
🧪 Comprehensive Test Suite
Full-stack smoke gate runs in the CI workflow. |
mvalancy
left a comment
mvalancy
left a comment
There was a problem hiding this comment.
Independent local evidence review at41c1fa015e16ae9e8cfe54c3336ff68100df29b6: the six-case baseline and corrected three-case serial headed reruns are valid within their stated scope. I inspected the actual desktop/phone local-item and topology before/after/reload PNGs and independently recomputed the JSON geometry/identity results. The final three-case run passes in51.6s with24 valid attachments (14JSON/10PNG).
Desktop local detail has seven readable cards, six routed connections representing73 of107 original relationships, and34 explicitly in the list. Actual minimum card gap is64px, unrelated-card wire clearance20px and unrelated-wire clearance152px. Phone has the focus plus one readable original neighbor, one painted connection and106 explicitly in the list; its card gap is76px. Titles are18px/600. The390px bottom navigation has three equal130×50px tabs with zero overflow, and the connection dock spans the canvas width. These frames look clean and readable.
Verified108 original items/107 directed edges (106outgoing/one incoming); all original positions/metadata/other titles/edges remain unchanged apart from the one exact allowed title edit. Actual D3x/y/fx/fy values remain equal through paging, focusing, Back, pan, reload and editing. Corrected native inventories contain108 nodes and107 each of edges/hit paths/arrows/labels, all hidden while the local layer owns presentation. Both isolated hit-path/arrow exposure negatives show the old selectors miss one visible target and the corrected helper rejects it. The premature offscreen capture finding is corrected: readiness is true and both original cards have painted readable titles in the final PNG (recheck).
No remaining blocker in this local artifact slice. Full hosted2,000-item/5,000-relationship dense-local transitions, exact revision/runtime provenance, all13 strict reports and actual post-dev acceptance remain pending. This is a comment review, not an approval or integration claim. Scratch audit and hash inventories: /tmp/pr234-independent-artifacts-762a/.
|
Final source is frozen at41c1fa015e16ae9e8cfe54c3336ff68100df29b6 and the PR is ready for review. Independent exact-source review5693519866 is clear; every review thread has been answered and resolved. Independent bounded artifact review5693539748 accepts the corrected desktop/phone/native-isolation/canonical-coordinate evidence. Final affected headed3cases pass51.6s, one worker/zero retries, with earlier failed controls retained. The final hosted2,000-item/5,000-relationship step passed at2026-10-29 06:17:09UTC, including ordinary and111-neighbor dense phases, local→overview→local, exact source conservation and the unchanged5s browser-zoom assertions. The5m48s step duration includes fixture creation and every phase; exact individual timings must come from the pending archive, not from that duration or badge. Comprehensive40826842856 passes Node22.18/24 plus installer/config. CI40826842888 static/build/units/MCP/security and all3runtime-image contracts/scans pass; inspector/new-user/LOD/recovery/mobile/remaining smoke checks and the final13-report artifact audit are still running. Independent runtime provenance verifies synthetic9a011696 parents479edf6+41c1 with source-equivalent tree85bef78b9c7fcd7942dac1cfa4d59717afa2411b. Baseline self-host OS findings are retained under the unchanged fixable-HIGH policy, not described as zero vulnerabilities. After final complete-source CI/artifact acceptance, re-read the live dev ref and perform only the authorized exact-head native administrative merge into dev. Verify resulting parents/tree and actual post-dev gates before closing233. #220 retains the remaining native canonical-card obstacle/boundary-bus work; verified next-slice plan includes CodeGraph ideas without shared libraries. Main/release/deployment and hosted parity remain outside these claims. |
| @@ -1913,6 +1955,9 @@ export function InteractiveGraphVisualization({ onResetLayout, onNodeSelected, i | |||
| // never-placed nodes are seeded near center and left free to flow. | |||
| // (This block used to null every node's fx/fy unconditionally, which is | |||
There was a problem hiding this comment.
blocking — Preserve the first phone tap while initially organizing/framing new items. The final hosted new-user gate fails the unchanged single-tap editing journeys at320×568 and390×844; the quick editor never opens after the one real touchscreen tap. The fixture creates two unplaced originals, which this new path places and exposes before initial framing/save feedback settles. In the390 trace, the tap starts at(188.95,334.41) with the native group at scale1 and the target at translate(188.9487,227.5). During the55ms tap the group changes to translate(-7.3341,47.1184)scale1.03761 and presentation passes6→7; with the108px canvas origin, the target center moves from335.5 to391.2px. At320, the actively saving layout panel and its two wrapped disabled action buttons cover the attempted tap area. The final failure images contain no editor. Please fix initial camera/presentation/input ownership and the obstructive healthy saving UI, with a controlled first-touch regression; do not hide this by retrying the opening action or extending the5s editor timeout. Actual source41c1 / synthetic9a011696 CI40826842888, failed new-user cases13/14; archive12878350925 retains screenshots0/2 and traces1/3. This finding identifies verified interaction loss and the observed mid-touch transforms; it does not claim that detached title DOM is proven to be the sole cause.
mvalancy
left a comment
mvalancy
left a comment
There was a problem hiding this comment.
Final hosted acceptance is blocked at41c1fa015e16ae9e8cfe54c3336ff68100df29b6. CI40826842888 fails the unchanged first-tap editing journeys at320×568 and390×844: the quick editor never opens. There are two actual failing cases plus the adapter's failed execution-evidence row. Verified finding and trace details. The remaining semantic/nested/recovery/mobile/performance sequences did not run, so this archive cannot satisfy the13-report acceptance contract.
The separately passing2,000-item/5,000-edge case has useful new evidence. I independently parsed its raw files and recomputed both dense geometries: seven cards/six routes around111 neighbors,101 original relationships shown/10 in the list,64px card gap,20px wire-to-card clearance and104px unrelated-wire clearance. Native cards, arrows, hit paths and labels are hidden; all2,000 original D3x/y/fx/fy snapshots remain equal. The actual dense→overview→dense PNGs are clean and readable. Load is26.609s; ordinary browser zooms are4.0073/4.4168/2.8601s, and dense zooms1.5559/2.2120/2.2566s, all below the unchanged5s browser budget. Bridge wall times remain separately reported, including5.241/5.996s for the first two ordinary calls.
Exact provenance is clear: tested synthetic9a011696955e80ae765a96b8e0eafd43655e0791 has parents479edf6+41c1 and the source-equivalent tree85bef78b9c7fcd7942dac1cfa4d59717afa2411b. All seven available reports have a clean, unchanged fingerprint;140 referenced attachments and68 JSON files are valid. All three fresh packaged runtime contracts/scans and both Comprehensive Node variants pass. Those valid slices do not override the first-tap failures or missing later sequences.
Retain this failed archive and fix/review the interaction regression before another exact-source run. No approval or merge recommendation. Independent audit, hashes and extracted touch frames are under /tmp/pr234-independent-artifacts-762a/; the failed hosted archive is12878350925.
Items mode around a high-degree focus now shows readable original neighbors and named overflow groups with bounded schematic buses. EPIC2215 previously painted107 wires while every neighbor was offscreen; the local display cut keeps useful endpoints on screen without changing original positions. Paging/filtering, directed original-edge inspection, original editing, Back and reload retain context. Small direct graphs keep their native view.
Connectivity-aware Organize places central hubs, chains and components with measured card clearance and bounded refinement. Overview buses retain64px card gutters,20px unrelated-card/16px unrelated-wire clearance and honest original-link coverage. Phone status/navigation spans the bottom edge with equal app tabs and44/48px touch controls; measured three-line item titles remain18px.
Closes #233. General arbitrary detailed item routing remains #220.
Final source:
41c1fa015e16ae9e8cfe54c3336ff68100df29b6, composed with accepted live dev479edf6/PR236. Independent exact-source review is clear; all findings answered/resolved.Validation: coherent six-case headed organization/local journeys pass90.16s; final-head affected three cases pass51.6s after strengthening native-class isolation negatives, live D3 x/y/fx/fy equality and settled offscreen paint.107-neighbor desktop/phone fixtures check actual paint,64/20/16clearance, exact directed original IDs, hover/keyboard, paging/filter reset, focus/Back, small pan, cold reload, one exact original-title edit and unchanged remaining source data. Phone tabs are130/130/130px across390px,50px high, no overflow; status dock is full-width/flush.44focused model/layout/selection/controller/inspector controls, the compact/expanded storage-label controls, types and scoped lint pass (legacy inspector-any warnings explicit). Earlier failures and rejected screenshots are retained in canonical lessons.
Pending gates: CI40826842888, Comprehensive40826842856. The existing2,000-item/5,000-relationship gate now also exercises a111-neighbor hub (107added), local→overview→local under the unchanged5s budget, and actual canonical conservation; ordinary detail/priority/input budgets remain. Final hosted13-report/runtime/package artifact audit, dev-only exact-head integration and actual post-dev acceptance remain required before closure. No original localhost changes, heavy local matrix, main/deployment or hosted-parity claims.