AI Flow

@@ -82,38 +82,6 @@ Every clip stop runs a blank check (nine sampled frames; a clip whose frames are - KiCad drops pour islands that touch nothing: a heat spreader drawn across dense routing on the other layer fills as fragments and reads small in `measure`. Put the copper where the layer is actually free (the analysis numbers say when it is not). - 0.1's analyses are conservative heuristics (IPC-2221 for tracks, presence, connection style and vias for pours and tabs); they say so in their output. 0.2 computes cross-sections through the filled copper. -## Component coverage before placement+## Silkscreen placement, complete coverage and observable decisions -The early `components` step audits component identity, library coverage and model quality before `models` and placement. Run `plan` to read its ordered sourcing workflow. Inventory every reference on the actual board and schematic, and track exact manufacturer/MPN, supplier code, package and required ratings. Reuse exact wiki component pages first; check their assets rather than assuming a page contains a complete CAD bundle. Use `adom-parts-search` next and manufacturer/distributor sites for gaps. Pup is the browser fallback for JavaScript or blocked curl/fetch downloads; read its skill for current commands.--Publish missing identified component pages with verified symbol/footprint/model assets and provenance; improve existing pages instead of duplicating them. Do not invent identities for generic land patterns, DNP parts or bare copper test pads, or silently substitute a similar part. Record unresolved identities and missing assets separately. Generated/approximate bodies must be labelled. Keep model paths portable, preserve layout, rerun `kicad_model_check`, and visibly inspect native KiCad 3D before calling model coverage complete. The AI performs this sourcing audit; these instructions do not add an automatic sourcing or component-identity gate.--Global component pages serve ALL ADOM USERS. Publish only distinct reusable manufacturer parts or independently specified reusable custom components, after duplicate checking. Keep page titles and assets part-focused and portable. Board reference mappings, unresolved identities, one-off footprints, bare copper features and project-only assemblies belong in the board project wiki page, not new component catalog entries. Do not turn project-specific placeholders into generic-looking pages merely by renaming them.--## Early component quality and optional MPN marking--Before placement, declare `step components` and run `components` with your `--run` and `--ai-thread`. It writes `components.json`, inventories every board reference and preserves review notes on rerun. This is an AI review register, not an automated geometry or licensing certification. Record wiki links, classification, dimensions, terminals, pin-1/polarity, plain/marked variants, top/bottom/oblique visual evidence and redistribution evidence. Unknowns remain unknown.--Use manufacturer drawings and permitted CAD as reference checks. Third-party downloads may be private reference-only; never redistribute them without permission, or call extracted/recolored/etched geometry independently authored. Compare independently generated geometry with dimensional and visual evidence. Publish only reusable assets on global pages; board-only definitions stay in the project.--Offer `components --etch on` or `--etch off` (initial default `ask` is not consent). Discover adom/adom-chip-laser and the current STEP engine. Keep the plain model; cache a separate MPN-marked STEP, GLB and reviewed thumbnails when enabled, with input hash, generator version/parameters, text/mode and native-render evidence. Protect pin-1, polarity and functional features. Marking is an identification aid, not a claim of real factory markings. Do not regenerate a valid shared cache, and do not claim KiCad/Fusion/Altium parity until each named viewer was checked.--## Provenance is required for shared component assets--Creating a new global component page is optional; truthful per-file provenance is required when publishing or improving one. Reuse and improve existing pages first. Keep board-only definitions in the board project. Add a readable provenance section and machine-readable asset record, linked prominently from the component README (including HTML READMEs without removing their existing content).--For every symbol, footprint, STEP, GLB and optional marked variant, record the source URL/file/revision and retrieval time, source and output SHA-256, authoring classification (manufacturer-supplied, source-derived, AI-created, or unknown), generator and version, parameters, units, transforms, modifications, redistribution evidence, checks performed and unresolved limits. A hash proves file identity, not quality or permission. Existing wiki availability alone does not establish upstream authorship or rights.--For AI-created geometry, explicitly say it is AI-created and identify the datasheet page/figure/table, dimensional numbers and assumptions used by the generator. Publish the generator or reproducible parameters, disclose simplifications, and distinguish independent geometry from a converted/extracted reference. Keep restricted reference CAD private; document what the comparison tested and did not test. Preserve the lineage from plain input to marked STEP to GLB, including full marking text/mode.--Link to the component page's issue tracker and ask reviewers to include the filename/hash, disputed dimension or pin, source evidence, EDA/version and screenshot. Never fill provenance gaps with plausible guesses. Unknown origins and failed or missing validation remain visible until supported by evidence.--## Optional live progress--Offer the Hydrogen progress widget at intake; default off unless requested. Use `widget enable`, then open AI Flow progress in Hydrogen Widgets. Reuse saved milestone PNGs through `widget event --file <inside-run image> --label <milestone>`; the service also follows saved ledger artifacts and step changes. Use it for component models, longest-axis MPN variants, symbols, placement, pours and Fields. `widget disable` stops updates and thumbnail work. No AI calls or extra captures are required; the progress bar counts planned steps, not remaining time. See widgets/README.md.--## Keep both native views current after every update--After EVERY board, footprint, silkscreen, library-binding or 3D-model update, update BOTH the native 2D board editor and its associated 3D viewer before reporting or showing the result. A file write, successful transfer, DRC result or web preview is not a refreshed native view. Verify the exact saved board path/revision in the editor, then regenerate/reload the 3D view and inspect the changed features after painting settles. Capture evidence from both exact windows; confirm models, markings and layer visibility, not only the window titles.--Use supported native refresh/reload commands through the owning EDA bridge. If an offline edit or model cache requires closing and reopening, inspect unsaved changes and dialogs first, preserve user work, close only task-owned stale windows, reopen the latest board, and open its linked 3D viewer. Never save stale editor contents over a newer disk revision. Keep one current editor/viewer pair rather than accumulating old windows. Re-discover HWNDs after reopening; preserve the user's foreground and view preferences unless showing a view was requested. Refresh a completed edit or coherent batch promptly; do not wait until the final video. If either view cannot be verified, state which one remains stale and resolve it before claiming the update is shown.+Read `docs/silkscreen.md` for complete coverage, font/locality search, both-face contact/test-point labels, values and leader rules. Read `tools/silkscreen-dashboard/SKILL.md` for optional live/replay observation and recordings. Native geometry, edit/undo and 2D/3D refresh remain bridge responsibilities. `silkscreen-layout` proposes; `silkscreen-audit` checks an independent whole-board inventory; neither claims native acceptance. The dashboard makes no AI/provider calls. Fusion and Altium data adapters still need implementation and native tests.

@@ -0,0 +1,9 @@+# Watch silkscreen placement++The optional AI Flow silkscreen dashboard displays live solver events or replays their saved history. Users can inspect sampled candidates, selected reference/value pairs and unresolved groups in 2D, then inspect fitted component geometry in 3D. It makes no AI/provider calls. Rendering and recording have CPU/GPU/storage costs; no token multiplier is assumed.++Start/show with `adom-aiflow silkscreen-dashboard show --ai-thread "<name>" --run <run>`. Prepare the explicit manifest described in `tools/silkscreen-dashboard/SKILL.md`; the ESC's exporter is currently a task adapter, not an automatic cross-EDA importer. Native bridge ownership is documented in KiCad Bridge issue #104.++Use `silkscreen-layout --events <run>/silkscreen-dashboard/events.jsonl` to emit real timestamped solver decisions. The dashboard samples candidates and explains constraints; it does not expose private reasoning or fabricate intermediate edits. Opt out by omitting `--events` and not starting the dashboard. Solver correctness does not depend on visualization.++Live/replay status and native-review-pending status remain visible. Record replay produces a labelled 2D WebM; retain the detailed walkthrough separately from the short final-video segment. No replay is presented as live native CAD footage. KiCad, Fusion and Altium bridges should supply native geometry/metrics, edits, undo, validation and linked view refresh; the dashboard and search stay shared. Only the KiCad ESC data adapter has been exercised so far.

@@ -59,3 +59,35 @@ For the requested InstaPCB profile, attempt a nearby value for EVERY resistor an After EVERY board, footprint, silkscreen, library-binding or 3D-model update, update BOTH the native 2D board editor and its associated 3D viewer before reporting or showing the result. A file write, successful transfer, DRC result or web preview is not a refreshed native view. Verify the exact saved board path/revision in the editor, then regenerate/reload the 3D view and inspect the changed features after painting settles. Capture evidence from both exact windows; confirm models, markings and layer visibility, not only the window titles.  Use supported native refresh/reload commands through the owning EDA bridge. If an offline edit or model cache requires closing and reopening, inspect unsaved changes and dialogs first, preserve user work, close only task-owned stale windows, reopen the latest board, and open its linked 3D viewer. Never save stale editor contents over a newer disk revision. Keep one current editor/viewer pair rather than accumulating old windows. Re-discover HWNDs after reopening; preserve the user's foreground and view preferences unless showing a view was requested. Refresh a completed edit or coherent batch promptly; do not wait until the final video. If either view cannot be verified, state which one remains stale and resolve it before claiming the update is shown.++## Search space before shrinking text++Treat 0.2 mm as an exceptional fallback, not a successful default. Start with the largest sensible font for the information hierarchy and search nearby positions, rotations and both sides of the component. Use the native EDA's stroke-font bounds, stroke width, justification, rotation and mirrored bottom-face transform. If only estimated bounds are available, identify them as estimates and validate every accepted batch with native plots/DRC and assembled 3D visibility.++Build obstacles from fitted component bodies, individual pads and solder-mask openings, vias/holes, board edges, existing silk and text. Do not replace a collection of separated obstacles with one enormous bounding box that discards useful gaps. Keep reference/value pairs associated but retain separate text rectangles so empty space between them remains usable. Solve each face with its own obstacles; a copied top layout is not a validated bottom layout.++Search alternative positions at each useful font size, then jointly repack neighboring label groups. Use a bounded search with backtracking or beam alternatives so an early label does not permanently consume the best space. Favor complete coverage, readable font sizes and short unmistakable association; preserve connector pin-row order. Reserve and validate curved leader paths too. Report unresolved labels rather than hiding or silently omitting them. Never move copper or components merely to fit text without a design return.++Use `adom-aiflow silkscreen-layout --input candidates.json --out layout.json --ai-thread <thread> --run <run>` for the shared bounding-box candidate search. Input carries board bounds, fixed obstacle rectangles and label groups with alternative native-derived rectangles, font sizes, positions and anchors. A candidate may provide several `boxes` for an intact ref/value pair or leader segments. This is a bounded weighted search, not proof of a global optimum; it does not edit the board. Native text measurement/insertion/refresh remains the EDA bridge's job. Apply by stable item identity, never by text alone: duplicate values and repeated pinout tables are common.++Run native clearance/overlap checks, compare findings against the inherited baseline, then inspect actual-size legibility and close-up association on both faces. Revise the candidates when validation rejects them; do not just call the search successful. Refresh both the 2D editor and its linked 3D viewer after an accepted batch. Record genuine placement and rework in the silkscreen step clip.++### Candidate geometry evidence++`tools/native-text-bounds.py --svg front.svg --labels labels.json --layer F.SilkS --out bounds.json` measures native stroke paths from a single-face KiCad SVG export. Export each face separately through the bridge. The label manifest contains `labels` with `text`, `x`, `y`, `size`, `angle` and `layer`. The helper reports its nearest-text matching method and SVG hash; inspect ambiguous duplicate/multiline mappings. It is a fallback geometry reader, not a native live text-metrics API. Preserve stable item IDs for application.++A `silkscreen-layout` input has `bounds: [xmin,ymin,xmax,ymax]`, `obstacles: [{box: [...]}]`, and `labels: [{id,anchor:[x,y],candidates:[{font,position:[x,y],box:[...]}]}]`. Optional candidate `boxes` describes separate rectangles for grouped text/leader segments; every rectangle must be clear. Optional `orderGroup` and numeric `order` preserve increasing connector-row Y. `beamWidth` and `candidatesPerLabel` bound the search. The score weights font size, distance and movement; it cannot promise the global optimum. All outputs retain `nativeVerified: false` until separate native review evidence is recorded.++For tightly coupled regions, set `solver: "milp"` to use SciPy/HiGHS mixed-integer selection with `solverSeconds`, `milpPerFont`, `preferredMinimumFont`, and `smallFontPenalty`. This optional engine needs SciPy; the default beam engine uses Python's standard library. Candidate pruning must preserve alternatives at every font size. Solver optimality only concerns the finite supplied candidate set, never all possible board artwork.++`tools/silkscreen-leaders.py --input candidates.json --out routed-candidates.json` accepts per-label `leaderTargets` and `routeBounds`, uses a bounded grid search around fixed rectangles, and appends leader segment bounds to each candidate. The final selection prevents those reserved paths crossing other selected text/routes. It retains alternatives per font size. Native application may gently round corners inside the cleared envelope; validate the actual resulting curves. Failure to route a leader must trigger repacking or an explicit unresolved finding.++### Complete obstacle coverage is required++Before placing or accepting silkscreen, enumerate every fitted model, via and drilled hole across the entire board, on both faces. Record coverage counts and stable IDs. A local search region does not excuse leaving the rest of the board unchecked. Project the actual fitted 3D model into the board plane with its scale, offset, rotation and footprint transform; include overhangs. Conservative transformed model bounds are acceptable when a tighter silhouette is unavailable. F.Fab, courtyard or pad envelopes are fallbacks, not proof of fitted-body coverage: a footprint can have no Fab drawing, and the space between separated pads can still be occupied by the body. Missing model geometry must be reported and supplied a documented conservative envelope, never a tiny placeholder that silently allows text beneath the component.++Reserve every via hole on both faces, including tented vias when visible labeling is the requirement, and use the larger of the hole keepout and actual exposed mask opening. Also reserve through holes, slots, exposed pads and existing silk. Check actual stroked text extents, not just its anchor. After each re-layout, audit ALL existing labels and leaders against this complete obstacle set. A collision-free position is not enough: keep each reference/value next to its own component or add a clear, noncrossing leader. Do not move a label beside a different component merely to clear an obstacle. Native DRC and fitted 3D visibility are separate acceptance checks; a DRC baseline does not establish readable assembled silkscreen. Refresh and inspect both native 2D and 3D views before presenting the result.++## Optional live silkscreen dashboard++Offer `silkscreen-dashboard show` with the run and ai-thread flags. The shared observer renders real timestamped solver events, sampled candidate bounds and reasons, selected proposals and unresolved labels, in 2D and a fitted-model 3D preview. `silkscreen-layout --events <run>/silkscreen-dashboard/events.jsonl` emits these without AI/provider calls or deliberate solver pauses. Read `tools/silkscreen-dashboard/SKILL.md` for the manifest, transform limits, lifecycle and recording contract. Live and replay are visibly distinct; replay exports are not raw native CAD footage. Selected proposals remain unverified until native application, DRC and assembled visibility checks pass. Fusion/Altium adapters are not implemented merely because the manifest is tool-neutral.

@@ -12,6 +12,5 @@ command -v service-kicad >/dev/null 2>&1 || echo "Hint: the offline gate needs s command -v adom-bridge >/dev/null 2>&1 || echo "Hint: the live stages need adom-bridge (the Adom Bridge CLI) and the KiCad Bridge on a desktop" echo "OK: adom-aiflow installed at ~/.local/bin/adom-aiflow. Run 'adom-aiflow --version'." -# Retain the helper beside its package so the scoped stop can verify its PID.-chmod +x "$HERE/tools/aiflow-widget.py"-ln -sf "$HERE/tools/aiflow-widget.py" "$HOME/.local/bin/adom-aiflow-widget"+chmod +x "$HERE/tools/silkscreen-dashboard/server.py"+ln -sf "$HERE/tools/silkscreen-dashboard/server.py" "$HOME/.local/bin/adom-aiflow-silkscreen"

@@ -82,32 +82,6 @@ Every clip stop runs a blank check (nine sampled frames; a clip whose frames are - KiCad drops pour islands that touch nothing: a heat spreader drawn across dense routing on the other layer fills as fragments and reads small in `measure`. Put the copper where the layer is actually free (the analysis numbers say when it is not). - 0.1's analyses are conservative heuristics (IPC-2221 for tracks, presence, connection style and vias for pours and tabs); they say so in their output. 0.2 computes cross-sections through the filled copper. -## Component coverage before placement+## Silkscreen placement, complete coverage and observable decisions -The early `components` step audits component identity, library coverage and model quality before `models` and placement. Run `plan` to read its ordered sourcing workflow. Inventory every reference on the actual board and schematic, and track exact manufacturer/MPN, supplier code, package and required ratings. Reuse exact wiki component pages first; check their assets rather than assuming a page contains a complete CAD bundle. Use `adom-parts-search` next and manufacturer/distributor sites for gaps. Pup is the browser fallback for JavaScript or blocked curl/fetch downloads; read its skill for current commands.--Publish missing identified component pages with verified symbol/footprint/model assets and provenance; improve existing pages instead of duplicating them. Do not invent identities for generic land patterns, DNP parts or bare copper test pads, or silently substitute a similar part. Record unresolved identities and missing assets separately. Generated/approximate bodies must be labelled. Keep model paths portable, preserve layout, rerun `kicad_model_check`, and visibly inspect native KiCad 3D before calling model coverage complete. The AI performs this sourcing audit; these instructions do not add an automatic sourcing or component-identity gate.--Global component pages serve ALL ADOM USERS. Publish only distinct reusable manufacturer parts or independently specified reusable custom components, after duplicate checking. Keep page titles and assets part-focused and portable. Board reference mappings, unresolved identities, one-off footprints, bare copper features and project-only assemblies belong in the board project wiki page, not new component catalog entries. Do not turn project-specific placeholders into generic-looking pages merely by renaming them.--## Early component quality and optional MPN marking--Before placement, declare `step components` and run `components` with your `--run` and `--ai-thread`. It writes `components.json`, inventories every board reference and preserves review notes on rerun. This is an AI review register, not an automated geometry or licensing certification. Record wiki links, classification, dimensions, terminals, pin-1/polarity, plain/marked variants, top/bottom/oblique visual evidence and redistribution evidence. Unknowns remain unknown.--Use manufacturer drawings and permitted CAD as reference checks. Third-party downloads may be private reference-only; never redistribute them without permission, or call extracted/recolored/etched geometry independently authored. Compare independently generated geometry with dimensional and visual evidence. Publish only reusable assets on global pages; board-only definitions stay in the project.--Offer `components --etch on` or `--etch off` (initial default `ask` is not consent). Discover adom/adom-chip-laser and the current STEP engine. Keep the plain model; cache a separate MPN-marked STEP, GLB and reviewed thumbnails when enabled, with input hash, generator version/parameters, text/mode and native-render evidence. Protect pin-1, polarity and functional features. Marking is an identification aid, not a claim of real factory markings. Do not regenerate a valid shared cache, and do not claim KiCad/Fusion/Altium parity until each named viewer was checked.--## Provenance is required for shared component assets--Creating a new global component page is optional; truthful per-file provenance is required when publishing or improving one. Reuse and improve existing pages first. Keep board-only definitions in the board project. Add a readable provenance section and machine-readable asset record, linked prominently from the component README (including HTML READMEs without removing their existing content).--For every symbol, footprint, STEP, GLB and optional marked variant, record the source URL/file/revision and retrieval time, source and output SHA-256, authoring classification (manufacturer-supplied, source-derived, AI-created, or unknown), generator and version, parameters, units, transforms, modifications, redistribution evidence, checks performed and unresolved limits. A hash proves file identity, not quality or permission. Existing wiki availability alone does not establish upstream authorship or rights.--For AI-created geometry, explicitly say it is AI-created and identify the datasheet page/figure/table, dimensional numbers and assumptions used by the generator. Publish the generator or reproducible parameters, disclose simplifications, and distinguish independent geometry from a converted/extracted reference. Keep restricted reference CAD private; document what the comparison tested and did not test. Preserve the lineage from plain input to marked STEP to GLB, including full marking text/mode.--Link to the component page's issue tracker and ask reviewers to include the filename/hash, disputed dimension or pin, source evidence, EDA/version and screenshot. Never fill provenance gaps with plausible guesses. Unknown origins and failed or missing validation remain visible until supported by evidence.--## Optional live progress--Offer the Hydrogen progress widget at intake; default off unless requested. Use `widget enable`, then open AI Flow progress in Hydrogen Widgets. Reuse saved milestone PNGs through `widget event --file <inside-run image> --label <milestone>`; the service also follows saved ledger artifacts and step changes. Use it for component models, longest-axis MPN variants, symbols, placement, pours and Fields. `widget disable` stops updates and thumbnail work. No AI calls or extra captures are required; the progress bar counts planned steps, not remaining time. See widgets/README.md.+Read `docs/silkscreen.md` for complete coverage, font/locality search, both-face contact/test-point labels, values and leader rules. Read `tools/silkscreen-dashboard/SKILL.md` for optional live/replay observation and recordings. Native geometry, edit/undo and 2D/3D refresh remain bridge responsibilities. `silkscreen-layout` proposes; `silkscreen-audit` checks an independent whole-board inventory; neither claims native acceptance. The dashboard makes no AI/provider calls. Fusion and Altium data adapters still need implementation and native tests.

@@ -0,0 +1,25 @@+from pathlib import Path+import xml.etree.ElementTree as E,re,json,math,argparse,hashlib+p=argparse.ArgumentParser(description='Measure text from a single-face native KiCad stroke SVG; layout evidence only.');p.add_argument('--svg',required=True,type=Path);p.add_argument('--labels',required=True,type=Path);p.add_argument('--out',required=True,type=Path);p.add_argument('--layer',default='F.SilkS');a=p.parse_args();root=E.parse(a.svg).getroot();groups=[]+def walk(e,sw=0):+ st=dict(re.findall(r'([\w-]+):([^;]+)',e.get('style','')));sw=float(st.get('stroke-width',e.get('stroke-width',sw)))+ if e.get('class')=='stroked-text':+  desc=next((x.text for x in e if x.tag.endswith('desc')),None);points=[]+  for p in e.iter():+   if p.tag.endswith('path'):+    vals=list(map(float,re.findall(r'[-+]?(?:\d*\.\d+|\d+)(?:[eE][-+]?\d+)?',p.get('d',''))));points+=list(zip(vals[::2],vals[1::2]))+  if points:+   xs,ys=zip(*points);groups.append(dict(text=desc,box=[min(xs)-sw/2,min(ys)-sw/2,max(xs)+sw/2,max(ys)+sw/2]))+ for c in e:walk(c,sw)+walk(root);m=json.loads(a.labels.read_text());out=[]+for i,l in enumerate(m['labels']):+ if l['layer']!=a.layer:continue+ boxes=[]+ for line in l['text'].split('\n'):+  gs=[g for g in groups if g['text']==line]+  if not gs:raise ValueError((i,line))+  def dist(g):b=g['box'];return math.hypot((b[0]+b[2])/2-l['x'],(b[1]+b[3])/2-l['y'])+  g=min(gs,key=dist);boxes.append(g['box'])+ b=[min(v[0] for v in boxes),min(v[1] for v in boxes),max(v[2] for v in boxes),max(v[3] for v in boxes)]+ out.append(dict(index=i,text=l['text'],box=b,offset=[(b[k]-(l['x'] if k%2==0 else l['y']))/l['size'] for k in range(4)]))+a.out.write_text(json.dumps({'sourceSvgSha256':hashlib.sha256(a.svg.read_bytes()).hexdigest(),'layer':a.layer,'method':'nearest matching native stroked-text group; inspect mapping before layout','bounds':out},indent=2));print('Native SVG text bounds:',len(out))

@@ -0,0 +1,23 @@+# Silkscreen dashboard++An optional, EDA-neutral observer of placement candidates and recorded solver decisions. It makes no AI/provider calls and never edits a native board. Native geometry, text metrics, undoable edits and 2D/3D refresh belong to each EDA bridge. The ESC integration currently exports KiCad geometry; Fusion and Altium adapters still need implementation and native tests. Do not claim those integrations work from schema compatibility alone.++## Displaying this app (REQUIRED — read before showing anything)++Use `adom-aiflow silkscreen-dashboard show --ai-thread "<name>" --run <run>` to open the named Hydrogen webview. Never use VS Code Simple Browser. `--surface pup` requires an explicit desktop `--target` and desktop-reachable `--url`. Keep standard workspace navigation. Reuse the same run/thread instance; `ls` shows instances and version drift.++## Inputs and live observation++Place `board.json` and an append-only `events.jsonl` in `<run>/silkscreen-dashboard/`. These are separate artifacts; never edit the measurement ledger `run.jsonl`. The board manifest has `bounds` in mm, `labels`, `obstacles`, `viaCount`, source-board identity and validation state. Each label carries text, x/y, size, angle, face and an optional existing-collision flag. Each obstacle carries a stable reference, kind, layer and conservative world-space box. Model obstacles may additionally supply a relative GLB URL, footprint `at` in KiCad XY/degrees, model offset and height. This initial 3D adapter supports top-mounted, unit-scale models without separate model rotations; bake other transforms into the supplied GLB or refuse the preview. The viewport's board slab is an envelope, not a manufactured-board export; pads, routed copper and drilled geometry are not yet native-equivalent. Keep that distinction visible.++`adom-aiflow silkscreen-layout --input candidates.json --out layout.json --events <run>/silkscreen-dashboard/events.jsonl --run <run> --ai-thread "<name>"` emits timestamped start, sampled candidate, selected proposal, unresolved and completion events with a solve ID. Events are generated by real computation, not invented for a film. Candidate samples are bounded: this is not every internal solver iteration. Pre-filtered inputs cannot reconstruct prior rejected candidates. The event reason describes the observable constraint/result, not private internal reasoning.++SSE updates do not reload the DOM. Live follows new events; replay has shared play/pause, scrub and speed controls. Playback is independent of solver speed. A selected proposal is not a native edit. Native application/DRC/assembled-visibility evidence must be recorded separately; the dashboard never promotes proposal success into native acceptance.++## Recording++Record replay exports a WebM of the 2D canvas with a persistent REPLAY label and the current recorded decision. Detailed review and a short final-video segment are separate artifacts. Preserve original event timestamps and raw native per-step clips. Do not call replay footage a live native edit or a KiCad recording. Verify the exported video actually plays. The browser-font preview does not replace KiCad's measured stroke-font geometry. Final-video pacing should not force the solver to sleep or trigger extra AI calls for every label.++## Acceptance++Audit the whole board on both faces: fitted body projections including the area between pads, every via/hole, actual mask openings, labels and leader paths. Missing Fab geometry is not empty space. Keep each label associated with its own component. Unresolved labels remain visible and block acceptance. The independent inventory in `silkscreen-audit` must be derived from the whole board, not from the same potentially incomplete obstacle list.

@@ -0,0 +1,28 @@+<!doctype html><html><head><meta charset="utf-8"><meta name="viewport" content="width=device-width"><title>AI Flow · Silkscreen</title><style>+:root{color-scheme:dark;font:14px system-ui;color:#e7edef;background:#141a20}*{box-sizing:border-box}body{margin:0;height:100vh;display:grid;grid-template-rows:48px 1fr 66px}header,footer{display:flex;align-items:center;gap:12px;padding:8px 18px;background:#1d252d;border-bottom:1px solid #35424d}header b{font-size:17px}button,select,input{font:inherit;color:inherit;background:#26333e;border:1px solid #536371;border-radius:5px;padding:7px}button{cursor:pointer}button:hover{border-color:#51cdc2}button.active{background:#24504f}main{display:grid;grid-template-columns:1fr 300px;min-height:0}#stage{position:relative;overflow:hidden;background:#111a1f}canvas{display:block;width:100%;height:100%;touch-action:none}#three{position:absolute;inset:0;display:none}aside{padding:18px;overflow:auto;background:#1a222a}h2{font-size:23px;margin:8px 0}p{line-height:1.5;color:#b5c4cd}.badge{border:1px solid #648076;border-radius:12px;padding:3px 10px;color:#7be0c7;font-size:12px}#native{color:#ffd293}.muted{color:#93a5b1;font-size:12px}#scrub{flex:1}#progress{height:5px;background:#33444f;margin:14px 0}#progress span{display:block;background:#48cdbb;height:100%;width:0}#legend{display:grid;gap:8px;margin-top:24px}.swatch{display:inline-block;width:12px;height:12px;margin-right:8px;border-radius:2px}#settings{position:absolute;right:20px;top:55px;z-index:8;background:#26313b;border:1px solid #566676;padding:18px;border-radius:8px}#settings label{display:block;margin:12px 0}#hud{position:absolute;left:16px;top:14px;pointer-events:none;background:#142028de;border:1px solid #41535e;padding:10px 14px;border-radius:6px}#error{color:#ff9999}a{color:#67dacc}footer{border-top:1px solid #35424d}#fit{margin-left:auto}@media(max-width:800px){main{grid-template-columns:1fr 220px}header{gap:5px;padding:7px}header b{font-size:14px}}+</style></head><body><header><b>Adom AI Flow · Silkscreen</b><span id="mode" class="badge">Connecting</span><select id="face" aria-label="Board face"><option value="F.SilkS">Top</option><option value="B.SilkS">Bottom</option></select><button id="two" class="active">2D layout</button><button id="threeBtn">3D inspection</button><button id="fit">Fit board</button><button id="gear" aria-label="Settings">⚙</button></header><main><div id="stage"><canvas id="board"></canvas><div id="three"></div><div id="hud"><div id="headline">Loading board geometry</div><div class="muted" id="geometry"></div></div></div><aside><div class="muted">Placement decision</div><h2 id="part">Waiting</h2><p id="reason">No solver events yet.</p><div id="progress"><span></span></div><div id="count"></div><p id="native">Native CAD review pending</p><label>Inspect component <select id="pick"><option value="">Whole board</option></select></label><div id="legend"><div><i class="swatch" style="background:#516a78"></i>Fitted body / reserved envelope</div><div><i class="swatch" style="background:#cbad4c"></i>Via / mask opening</div><div><i class="swatch" style="background:#ffb963"></i>Sampled candidate</div><div><i class="swatch" style="background:#57dcc4"></i>Selected proposal</div><div><i class="swatch" style="background:#f18389"></i>Existing collision</div></div><p class="muted">Selection is a proposal, not an edit to the EDA board. Native text bounds drive collision checks. Preview text uses the browser font.</p><div id="error"></div></aside></main><footer><button id="live">Follow live</button><button id="play">Replay</button><input type="range" id="scrub" min="-1" max="0" value="-1" aria-label="Event timeline"><span id="stamp" class="muted">0 / 0</span><button id="record">Record walkthrough</button><button id="short">Record 5s overview</button></footer><div id="settings" hidden><b>Playback & display</b><label>Events per second <input id="speed" type="number" min="0.5" max="60" step="0.5"></label><label><input id="obstacles" type="checkbox"> Show obstacles</label><label><input id="follow" type="checkbox" checked> Follow selected component</label><label>Idle shutdown, hours <input id="ttl" type="number" min="1" max="168"></label><p class="muted">No AI/provider calls. Solver runs independently of playback.</p></div>+<script type="module">+const $=s=>document.querySelector(s),canvas=$('#board'),ctx=canvas.getContext('2d');let data,history=[],state={},selected='',bounds,view,drag=null,viewer=null,scene=null,labelTexture=null,threeGroups=[],recording=false;const api=async(path,body)=>{const r=await fetch(path,{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({...body,aiThread:state.aiThread||'Silkscreen dashboard user'})});const j=await r.json();if(!r.ok)throw Error(j.error);if(j.state)await apply(j.state);return j};window.silkscreen={ready:false};+let cameraAnimation=0;function frame(b){cancelAnimationFrame(cameraAnimation);if(recording&&!state.summary&&view){const from=[...view],start=performance.now();const tick=now=>{const t=Math.min(1,(now-start)/220),ease=t*t*(3-2*t);view=from.map((v,i)=>v+(b[i]-v)*ease);draw();if(t<1)cameraAnimation=requestAnimationFrame(tick)};cameraAnimation=requestAnimationFrame(tick)}else{view=[...b];draw()}}+function fitPart(ref){const os=data.obstacles.filter(o=>o.reference===ref&&o.layer===state.layer);if(!os.length)return;const b=[Math.min(...os.map(o=>o.box[0]))-3,Math.min(...os.map(o=>o.box[1]))-3,Math.max(...os.map(o=>o.box[2]))+3,Math.max(...os.map(o=>o.box[3]))+3];frame(b);if(viewer){const cam=viewer.getCamera();cam.setTarget(new window.BABYLON.Vector3((b[0]+b[2])/2,-(b[1]+b[3])/2,0));cam.radius=Math.max(b[2]-b[0],b[3]-b[1])*1.8}}+function proposals(){const map=new Map();let start=0;for(let k=Math.min(state.cursor,history.length-1);k>=0;k--)if(history[k].kind==='start'){start=k;break}for(let k=start;k<=state.cursor&&k<history.length;k++){const e=history[k];if(e.kind==='placement')map.set(e.id,e.candidate)}return map}+function paint(c,w,h,extent,texture=false){c.clearRect(0,0,w,h);if(!texture){c.fillStyle='#122128';c.fillRect(0,0,w,h)}const bw=extent[2]-extent[0],bh=extent[3]-extent[1],scale=Math.min(w/bw,h/bh)*.93,sx=texture?w/bw:scale,sy=texture?h/bh:scale,ox=texture?-extent[0]*sx:(w-bw*scale)/2-extent[0]*scale,oy=texture?-extent[1]*sy:(h-bh*scale)/2-extent[1]*scale;c.save();c.translate(ox,oy);c.scale(sx,sy);c.lineWidth=1/scale;+function rect(b,color,fill=true){c.fillStyle=color;c.strokeStyle=color;if(fill)c.fillRect(b[0],b[1],b[2]-b[0],b[3]-b[1]);else c.strokeRect(b[0],b[1],b[2]-b[0],b[3]-b[1])}+if(!texture){rect(data.bounds,'#233b32');if(state.settings?.showObstacles)for(const o of data.obstacles){if(o.layer!==state.layer)continue;const col=o.kind.startsWith('fitted')?(o.reference===selected?'#607b87':'#394d58'):o.kind.startsWith('via')?'#bd9f43':'#889081';if(o.kind.startsWith('via')){const b=o.box;c.beginPath();c.arc((b[0]+b[2])/2,(b[1]+b[3])/2,(b[2]-b[0])/2,0,Math.PI*2);c.strokeStyle=col;c.stroke()}else rect(o.box,col,true)}}+const placed=proposals();const replaced=new Set([...placed.values()].flatMap(p=>(p.children||[]).map(x=>x.index)));+function label(l,color,b){c.save();c.translate(l.x??l.position?.[0]??0,l.y??l.position?.[1]??0);c.rotate(-(l.angle||0)*Math.PI/180);c.font=`${l.size||l.font||.5}px sans-serif`;c.textAlign='center';c.textBaseline='middle';c.fillStyle=color;(l.text||'').split('\n').forEach((s,i,all)=>c.fillText(s,0,(i-(all.length-1)/2)*(l.size||l.font||.5)*1.5));c.restore();if(!texture&&b)rect(b,color,false)}+for(let i=0;i<data.labels.length;i++){const l=data.labels[i];if(l.layer!==state.layer||replaced.has(i))continue;label(l,l.collision?'#f1838999':'#bfccc666')}+for(const p of placed.values()){if(p.layer&&p.layer!==state.layer)continue;for(const child of p.children||[p]){const original=child.index!=null?data.labels[child.index]:null;if(original&&original.layer!==state.layer)continue;label(child,'#79f5d9',texture?null:child.box)}}+const e=history[state.cursor];if(!texture&&e?.kind==='candidate'){for(const b of e.candidate.boxes||[e.candidate.box])rect(b,'#ffbd70',false);for(const child of e.candidate.children||[e.candidate])label(child,'#ffbd70')}c.restore();if(!texture&&recording){c.fillStyle='#101820df';c.fillRect(12,h-62,w-24,48);c.fillStyle='#dfece8';c.font='16px system-ui';c.fillText('REPLAY · '+(e?.id||'Silkscreen')+' · '+(e?.reason||e?.kind||''),24,h-33)} }+function draw(){if(!data||!view)return;const r=canvas.getBoundingClientRect();canvas.width=Math.max(1,r.width*devicePixelRatio);canvas.height=Math.max(1,r.height*devicePixelRatio);paint(ctx,canvas.width,canvas.height,view);if(labelTexture){const t=labelTexture.getContext();paint(t,2048,2048,data.bounds,true);labelTexture.update(false)}}+async function show3d(){if(viewer)return;$('#headline').textContent='Loading 3D geometry';try{await import('https://wiki.adom.inc/static/vendor/adom-3d-viewer-babylon9/adom-3d-viewer-babylon9.esm.js?v=20260915a');const B=window.BABYLON;viewer=window.Adom3DViewerBabylon9.init($('#three'),{zUp:true,showGround:false,showViewCube:false});scene=viewer.getScene();const b=data.bounds,cx=(b[0]+b[2])/2,cy=(b[1]+b[3])/2;const board=B.MeshBuilder.CreateBox('Board envelope',{width:b[2]-b[0],height:b[3]-b[1],depth:1.6},scene);board.position.set(cx,-cy,-.8);const bm=new B.PBRMaterial('board',scene);bm.albedoColor=B.Color3.FromHexString('#24563b');bm.metallic=0;bm.roughness=.85;board.material=bm;+for(const o of data.obstacles.filter(o=>o.kind.startsWith('fitted'))){const b=o.box;const mesh=B.MeshBuilder.CreateBox(o.reference,{width:b[2]-b[0],height:b[3]-b[1],depth:o.height||1},scene);mesh.position.set((b[0]+b[2])/2,-(b[1]+b[3])/2,(o.height||1)/2+.03);const mat=new B.StandardMaterial('reserved envelope',scene);mat.diffuseColor=B.Color3.FromHexString('#5b707d');mat.alpha=.72;mesh.material=mat;threeGroups.push({mesh,o});if(o.glb){try{const imported=await B.SceneLoader.ImportMeshAsync('',o.glb,'',scene),group=new B.TransformNode('fitted_'+o.reference,scene);for(const m of imported.meshes)if(!m.parent)m.parent=group;group.scaling.setAll(1000);const theta=(o.at[2]||0)*Math.PI/180,dx=o.offset?.[0]||0,dy=o.offset?.[1]||0;group.position.set(o.at[0]+dx*Math.cos(theta)-dy*Math.sin(theta),-o.at[1]+dx*Math.sin(theta)+dy*Math.cos(theta),o.offset?.[2]||0);group.rotation.z=theta;mesh.setEnabled(false);const buckets=new Map();for(const m of imported.meshes){if(!m.getTotalVertices())continue;m.computeWorldMatrix(true);const mat=m.material,key=JSON.stringify([mat?.albedoColor?.asArray(),mat?.diffuseColor?.asArray(),mat?.emissiveColor?.asArray(),mat?.alpha,mat?.metallic,mat?.roughness]);if(!buckets.has(key))buckets.set(key,[]);buckets.get(key).push(m)}for(const ms of buckets.values()){const mat=ms[0].material;for(const m of ms)m.material=mat;if(ms.length>1)B.Mesh.MergeMeshes(ms,true,true,undefined,false,false)}}catch(error){$('#error').textContent='Model fallback for '+o.reference+': '+error.message}}}+labelTexture=new B.DynamicTexture('silkscreen proposals',{width:2048,height:2048},scene,false);labelTexture.hasAlpha=true;const plane=B.MeshBuilder.CreatePlane('Silkscreen preview',{width:b[2]-b[0],height:b[3]-b[1]},scene);plane.position.set(cx,-cy,.015);const pm=new B.StandardMaterial('silkscreen',scene);pm.diffuseTexture=labelTexture;pm.opacityTexture=labelTexture;pm.useAlphaFromDiffuseTexture=true;pm.emissiveColor=B.Color3.White();pm.disableLighting=true;pm.backFaceCulling=false;plane.material=pm;viewer.getCamera().setTarget(new B.Vector3(cx,-cy,0));viewer.getCamera().radius=100;viewer.getCamera().beta=.65;viewer.getCamera().alpha=-1.6;viewer.getCamera().minZ=.01;viewer.getCamera().upperRadiusLimit=300;draw();$('#geometry').textContent='Fitted component models · rectangular board preview · native CAD review still required';}catch(e){$('#error').textContent=e.message}}+async function apply(s){const oldTotal=history.length;if(s.events)history=s.events;else if(s.total!==oldTotal){const full=await(await fetch('state')).json();history=full.events}state={...state,...s};window.silkscreen.state=state;const e=history[state.cursor];$('#mode').textContent=state.mode==='live'?'Live solver feed':'Recorded-event replay';$('#part').textContent=e?.id||'Whole board';$('#reason').textContent=e?.reason|| (e?.kind==='complete'?`Search complete; ${e.unresolved.length} unresolved groups.`:'Waiting for the next solver event.');$('#headline').textContent=e?.kind==='complete'?'Candidate layout · review pending':e?.id?`${e.id} · ${e.kind}`:'ESC silkscreen placement';$('#face').value=state.layer;$('#stamp').textContent=`${Math.max(0,state.cursor+1)} / ${history.length}`;$('#scrub').max=Math.max(0,history.length-1);$('#scrub').value=state.cursor;$('#count').textContent=`${proposals().size} selected groups`;$('#progress span').style.width=(history.length?(state.cursor+1)/history.length*100:0)+'%';$('#play').textContent=state.playing?'Pause':'Replay';$('#speed').value=state.settings.speed;$('#ttl').value=state.settings.ttlHours;$('#obstacles').checked=state.settings.showObstacles;if(e?.id&&selected!==e.id&&$('#follow').checked){selected=e.id;fitPart(selected)}$('#three').style.display=state.view==='3d'?'block':'none';$('#board').style.visibility=state.view==='3d'?'hidden':'visible';if(state.view==='3d'){await show3d();if(scene){const plane=scene.getMeshByName('Silkscreen preview');if(plane)plane.position.z=state.layer==='B.SilkS'?-1.615:.015;if(s.layer)viewer.getCamera().beta=state.layer==='B.SilkS'?Math.PI-.65:.65}}draw()}+try{data=await(await fetch('board.json')).json();bounds=data.bounds;view=[...bounds];const refs=[...new Set(data.obstacles.filter(o=>o.kind.startsWith('fitted')).map(o=>o.reference))].sort();for(const r of refs){const o=document.createElement('option');o.value=r;o.textContent=r;$('#pick').append(o)}$('#geometry').textContent=`${refs.length} fitted bodies · ${data.viaCount} vias · native measured text bounds`;+await apply(await(await fetch('state')).json());const stream=new EventSource('stream');stream.addEventListener('state',e=>apply(JSON.parse(e.data)).catch(e=>$('#error').textContent=e.message));stream.onerror=()=>$('#mode').textContent='Reconnecting';window.silkscreen.ready=true;+$('#live').onclick=()=>api('control',{mode:'live'});$('#play').onclick=()=>api('control',{mode:'replay',summary:false,playing:!state.playing,cursor:state.cursor>=history.length-1?-1:state.cursor});$('#scrub').oninput=e=>api('control',{mode:'replay',playing:false,cursor:+e.target.value});$('#face').onchange=e=>api('control',{layer:e.target.value});$('#two').onclick=()=>api('control',{view:'2d'});$('#threeBtn').onclick=()=>api('control',{view:'3d'});$('#fit').onclick=()=>{selected='';frame(bounds);if(viewer){const b=bounds;viewer.getCamera().setTarget(new window.BABYLON.Vector3((b[0]+b[2])/2,-(b[1]+b[3])/2,0));viewer.getCamera().radius=100}};$('#pick').onchange=e=>{selected=e.target.value;$('#follow').checked=false;selected?fitPart(selected):frame(bounds)};$('#gear').onclick=()=>$('#settings').hidden=!$('#settings').hidden;$('#speed').onchange=e=>api('settings',{speed:+e.target.value});$('#ttl').onchange=e=>api('settings',{ttlHours:+e.target.value});$('#obstacles').onchange=e=>api('settings',{showObstacles:e.target.checked});+canvas.onwheel=e=>{e.preventDefault();const f=e.deltaY>0?1.15:.87;const r=canvas.getBoundingClientRect(),px=(e.clientX-r.left)/r.width,py=(e.clientY-r.top)/r.height,w=view[2]-view[0],h=view[3]-view[1];view=[view[0]+w*px*(1-f),view[1]+h*py*(1-f),view[0]+w*px*(1-f)+w*f,view[1]+h*py*(1-f)+h*f];draw()};canvas.onpointerdown=e=>{drag={x:e.clientX,y:e.clientY,v:[...view]};canvas.setPointerCapture(e.pointerId)};canvas.onpointermove=e=>{if(!drag)return;const r=canvas.getBoundingClientRect(),dx=(e.clientX-drag.x)/r.width*(drag.v[2]-drag.v[0]),dy=(e.clientY-drag.y)/r.height*(drag.v[3]-drag.v[1]);view=[drag.v[0]-dx,drag.v[1]-dy,drag.v[2]-dx,drag.v[3]-dy];draw()};canvas.onpointerup=()=>drag=null;+async function recordReplay(summary=false){if(recording)return;const latest=history.map((e,i)=>e.kind==='start'?i:-1).filter(i=>i>=0).pop()||0;await api('control',{mode:'replay',playing:false,summary,cursor:latest-1,view:'2d'});$('#follow').checked=!summary;frame(bounds);recording=true;const chunks=[],rec=new MediaRecorder(canvas.captureStream(30),{mimeType:'video/webm'});rec.ondataavailable=e=>chunks.push(e.data);rec.onstop=()=>{recording=false;const blob=new Blob(chunks,{type:'video/webm'}),a=document.createElement('a');window.silkscreen.recordingBlob=blob;a.href=URL.createObjectURL(blob);a.download='silkscreen-replay.webm';a.click();$('#record').textContent='Record replay';draw()};rec.start();$('#record').textContent='Recording replay…';await api('control',{playing:true});const timer=setInterval(()=>{draw();if(!state.playing&&state.cursor>=history.length-1){clearInterval(timer);rec.stop()}},100)};$('#record').onclick=()=>recordReplay(false);$('#short').onclick=()=>recordReplay(true);+new ResizeObserver(draw).observe($('#stage'));draw();}catch(e){$('#error').textContent=e.stack}+</script></body></html>

@@ -0,0 +1,65 @@+#!/usr/bin/env python3+"""Optional bounded leader candidate routing for the rectangle planner.+Caller supplies per-label leaderTargets and routeBounds in board millimetres.+Routes are candidate geometry, requiring native clearance and visual review.+"""+import argparse,json,math,heapq+from pathlib import Path++def prepare(spec):+ gap=spec.get('clearance',.15);step=spec.get('leaderGrid',.1);fixed=[o['box'] for o in spec['obstacles']]+ def hit(a,b,g=0):return not(a[2]+g<=b[0] or b[2]+g<=a[0] or a[3]+g<=b[1] or b[3]+g<=a[1])+ for group in spec['labels']:+  if not group.get('leaderTargets'):continue+  rb=group['routeBounds'];nx=round((rb[2]-rb[0])/step)+1;ny=round((rb[3]-rb[1])/step)+1+  def point(n):return rb[0]+n[0]*step,rb[1]+n[1]*step+  def cell(p):return round((p[0]-rb[0])/step),round((p[1]-rb[1])/step)+  blocked=set()+  def occupy(b):+   x0=max(0,math.ceil((b[0]-gap-.03-rb[0])/step));x1=min(nx-1,math.floor((b[2]+gap+.03-rb[0])/step));y0=max(0,math.ceil((b[1]-gap-.03-rb[1])/step));y1=min(ny-1,math.floor((b[3]+gap+.03-rb[1])/step))+   return {(x,y) for x in range(x0,x1+1) for y in range(y0,y1+1)}+  for b in fixed:blocked.update(occupy(b))+  target_list=[cell(p) for p in group['leaderTargets']];target_list=[n for n in target_list if n not in blocked];targets=set(target_list[:1]);accepted=[]+  if not targets:group['candidates']=[];continue+  candidates=[c for c in group['candidates'] if not any(hit(c['box'],b,gap) for b in fixed)]+  candidates.sort(key=lambda c:-(c['font']*spec.get('fontWeight',70)-math.dist(c['position'],group['anchor'])*spec.get('distanceWeight',7)))+  bands={}+  for c in candidates:bands.setdefault(c['font'],[]).append(c)+  candidates=[c for band in bands.values() for c in band[:spec.get('leaderAttemptsPerFont',180)]]+  for c in candidates:+   if sum(q['font']==c['font'] for q in accepted)>=spec.get('leaderCandidatesPerFont',40):continue+   if c.get('angle',0)==0 and abs(c['position'][1]-group['anchor'][1])<=.3 and math.dist(c['position'],group['anchor'])<3.5:accepted.append(c);continue+   b=c['box'];occupied=blocked|occupy(b)+   def free(n):return 0<=n[0]<nx and 0<=n[1]<ny and n not in occupied+   starts=set();pad=gap+.08+   for x in range(math.floor((b[0]-pad-rb[0])/step),math.ceil((b[2]+pad-rb[0])/step)+1):starts.update([(x,cell((0,b[1]-pad))[1]),(x,cell((0,b[3]+pad))[1])])+   for y in range(math.floor((b[1]-pad-rb[1])/step),math.ceil((b[3]+pad-rb[1])/step)+1):starts.update([(cell((b[0]-pad,0))[0],y),(cell((b[2]+pad,0))[0],y)])+   starts={n for n in starts if free(n)};ends={n for n in targets if free(n)}+   if not ends or not starts:continue+   def h(n):return min(math.dist(n,z) for z in ends)+   costs={n:0 for n in starts};prev={n:None for n in starts};heap=[(h(n),0,n) for n in starts];heapq.heapify(heap);found=None+   while heap:+    _,g,n=heapq.heappop(heap)+    if g!=costs[n]:continue+    if n in ends:found=n;break+    for dx,dy in [(1,0),(-1,0),(0,1),(0,-1),(1,1),(1,-1),(-1,1),(-1,-1)]:+     z=(n[0]+dx,n[1]+dy)+     if not free(z) or (dx and dy and (not free((n[0]+dx,n[1])) or not free((n[0],n[1]+dy)))):continue+     ng=g+math.hypot(dx,dy)+     if ng<costs.get(z,1e99):costs[z]=ng;prev[z]=n;heapq.heappush(heap,(ng+h(z),ng,z))+   if found is None:continue+   path=[]+   while found is not None:path.append(point(found));found=prev[found]+   path.reverse()+   if len(path)<2:accepted.append(c);continue+   # Keep only changes in direction. Native EDA may round corners inside this envelope.+   reduced=[path[0]]+   for a,q,z in zip(path,path[1:],path[2:]):+    if abs((q[0]-a[0])*(z[1]-q[1])-(q[1]-a[1])*(z[0]-q[0]))>1e-8:reduced.append(q)+   reduced.append(path[-1]);rects=[b]+[[min(a[0],z[0])-.025,min(a[1],z[1])-.025,max(a[0],z[0])+.025,max(a[1],z[1])+.025] for a,z in zip(reduced,reduced[1:])]+   if any(hit(a,z,gap) for a in rects[1:] for z in fixed):continue+   accepted.append(dict(c,boxes=rects,leaderPoints=reduced))+  group['candidates']=accepted;print(group['id'],len(accepted),'routable candidates',flush=True)+ return spec+if __name__=='__main__':+ p=argparse.ArgumentParser();p.add_argument('--input',required=True);p.add_argument('--out',required=True);a=p.parse_args();s=prepare(json.loads(Path(a.input).read_text()));Path(a.out).write_text(json.dumps(s,indent=2))

@@ -0,0 +1,22 @@+#!/usr/bin/env python3+"""Fail-closed whole-board rectangular obstacle audit. Native CAD supplies the+transformed fitted-body/mask/hole envelopes and rendered text/leader bounds.+The inventory is independent of the obstacle export: missing geometry is an error.+"""+import argparse,json+from pathlib import Path++def overlaps(a,b):+ return a[0]<b[2] and b[0]<a[2] and a[1]<b[3] and b[1]<a[3]+def check(data):+ obs=data['obstacles'];inventory=data['inventory'];missing=[];collisions=[]+ present={(o['id'],o['kind'],o['layer']) for o in obs}+ for item in inventory:+  for layer in item['layers']:+   if (item['id'],item['kind'],layer) not in present:missing.append(dict(id=item['id'],kind=item['kind'],layer=layer))+ for label in data['marks']:+  for o in obs:+   if label['layer']==o['layer'] and any(overlaps(a,b) for a in label.get('boxes',[label['box']]) for b in o.get('boxes',[o['box']])):collisions.append(dict(mark=label['id'],obstacle=o['id'],kind=o['kind'],layer=o['layer']))+ return dict(schemaVersion=1,passed=bool(inventory) and not missing and not collisions,coverageErrors=[] if inventory else ['Whole-board inventory is empty'],missingObstacles=missing,collisions=collisions,inventoryItems=len(inventory),obstacles=len(obs),marks=len(data['marks']),nativeVerified=False,note='Conservative rectangle audit only; geometry provenance, fitted transforms, label association and native 2D/3D review remain required.')+if __name__=='__main__':+ p=argparse.ArgumentParser();p.add_argument('--input',required=True,type=Path);p.add_argument('--out',required=True,type=Path);a=p.parse_args();result=check(json.loads(a.input.read_text()));a.out.write_text(json.dumps(result,indent=2));print(('OK' if result['passed'] else 'ERROR')+f": {len(result['coverageErrors'])} coverage errors; {len(result['missingObstacles'])} missing obstacles; {len(result['collisions'])} collisions");raise SystemExit(0 if result['passed'] else 2)

@@ -0,0 +1,6 @@+import importlib.util,pathlib,tempfile,json+p=pathlib.Path(__file__).with_name('silkscreen-layout.py');s=importlib.util.spec_from_file_location('layout',p);m=importlib.util.module_from_spec(s);s.loader.exec_module(m)+spec=dict(bounds=[0,0,10,10],obstacles=[dict(box=[0,0,2,2])],labels=[dict(id='C1',anchor=[4,4],candidates=[dict(font=.5,position=[1,1],box=[.5,.5,1.5,1.5]),dict(font=.5,position=[4,4],box=[3,3,5,5])]),dict(id='C2',anchor=[1,1],candidates=[])],solver='milp')+with tempfile.TemporaryDirectory() as td:+ m.EVENT_FILE=str(pathlib.Path(td)/'events.jsonl');m.EVENT_ID='test';r=m.solve(spec);assert r['unresolved']==['C2'];events=[json.loads(l) for l in pathlib.Path(m.EVENT_FILE).read_text().splitlines()];assert any(e.get('reason')=='intersects a reserved obstacle' for e in events);assert any(e.get('reason','').startswith('clears') for e in events);assert all(e['solveId']=='test' for e in events);assert r['nativeVerified'] is False+print('PASS: real rejection/feasible events, solve identity, unresolved MILP result and native-review boundary')

@@ -0,0 +1,20 @@+import importlib.util+from pathlib import Path+sp=importlib.util.spec_from_file_location('silk',Path(__file__).with_name('silkscreen-layout.py'));m=importlib.util.module_from_spec(sp);sp.loader.exec_module(m)+def c(x,font):return dict(box=[x,0,x+1,1],position=[x+.5,.5],font=font)+# The first label's locally best choice blocks the next; beam must revisit it.+s={'bounds':[-1,-1,6,3],'clearance':0,'labels':[{'id':'a','anchor':[.5,.5],'candidates':[c(0,.8),c(2,.7)]},{'id':'b','anchor':[.5,.5],'candidates':[c(0,.8),c(0,.7)]}]}+r=m.solve(s);assert not r['unresolved'];assert next(x for x in r['placements'] if x['id']=='a')['position'][0]==2.5+s['obstacles']=[{'box':[-1,-1,6,3]}];assert len(m.solve(s)['unresolved'])==2+print('PASS: revises earlier choices, prefers larger text and reports blocked labels')++# A grouped reference/value must not reserve the empty middle between their boxes.+s={'bounds':[-1,-1,6,3],'clearance':0,'obstacles':[{'box':[1.5,0,2.5,1]}],'labels':[{'id':'pair','anchor':[1.5,.5],'candidates':[dict(box=[0,0,4,1],boxes=[[0,0,1,1],[3,0,4,1]],position=[1.5,.5],font=.6)]}]}+assert not m.solve(s)['unresolved']+# Preserve connector row order even if a larger candidate would swap labels.+s={'bounds':[-2,-2,6,6],'clearance':0,'labels':[{'id':'p1','orderGroup':'left','order':1,'anchor':[0,0],'candidates':[dict(box=[0,0,1,1],position=[.5,.5],font=.6)]},{'id':'p3','orderGroup':'left','order':3,'anchor':[0,2],'candidates':[dict(box=[0,-2,1,-1],position=[.5,-1.5],font=.8),dict(box=[0,2,1,3],position=[.5,2.5],font=.5)]}]}+r=m.solve(s);assert next(x for x in r['placements'] if x['id']=='p3')['font']==.5+print('PASS: compound bounds retain empty space and pin rows stay ordered')+# Optional MILP must trade a larger label for two readable labels rather than a tiny fallback.+s={'solver':'milp','bounds':[-1,-1,6,3],'clearance':0,'preferredMinimumFont':.3,'smallFontPenalty':1000,'distanceWeight':0,'labels':[{'id':'a','anchor':[0,0],'candidates':[c(0,.6),c(2,.4)]},{'id':'b','anchor':[0,0],'candidates':[c(0,.4),c(4,.2)]}]}+r=m.solve(s);assert not r['unresolved'];assert all(x['font']>=.3 for x in r['placements']);print('PASS: MILP avoids a tiny fallback when repacking fits both labels')

@@ -0,0 +1,16 @@+import importlib.util,pathlib,unittest+p=pathlib.Path(__file__).with_name('silkscreen-obstacle-check.py');s=importlib.util.spec_from_file_location('audit',p);m=importlib.util.module_from_spec(s);s.loader.exec_module(m)+class Audit(unittest.TestCase):+ def test_missing_body_between_pads(self):+  d=dict(inventory=[dict(id='C11',kind='body',layers=['F.SilkS'])],obstacles=[],marks=[])+  self.assertFalse(m.check(d)['passed']);self.assertEqual(len(m.check(d)['missingObstacles']),1)+ def test_via_requires_both_faces(self):+  d=dict(inventory=[dict(id='v1',kind='via',layers=['F.SilkS','B.SilkS'])],obstacles=[dict(id='v1',kind='via',layer='F.SilkS',box=[0,0,1,1])],marks=[])+  self.assertEqual(m.check(d)['missingObstacles'][0]['layer'],'B.SilkS')+ def test_text_anchor_outside_but_letters_cross_body(self):+  d=dict(inventory=[dict(id='C4',kind='body',layers=['F.SilkS'])],obstacles=[dict(id='C4',kind='body',layer='F.SilkS',box=[0,0,2,1])],marks=[dict(id='C11-label',layer='F.SilkS',box=[-1,.2,.3,.6])])+  self.assertEqual(len(m.check(d)['collisions']),1)+ def test_bottom_text_does_not_hit_top_body(self):+  d=dict(inventory=[dict(id='C4',kind='body',layers=['F.SilkS'])],obstacles=[dict(id='C4',kind='body',layer='F.SilkS',box=[0,0,2,1])],marks=[dict(id='label',layer='B.SilkS',box=[.1,.2,.3,.6])])+  self.assertTrue(m.check(d)['passed'])+unittest.main()