SVG-a1: inline <svg> (path, rect, circle; fill, stroke, stroke-width) on web, proven by a strict outline differential against Chrome; native refused as SVG-a2

[thejackshelton]

This PR is SVG-a1, the compiler and web half of coverage-rank lane 16 (inline SVG, reach 395). The spec is /tmp/specs/svg-a.md, approved by the PM ruling of 2026-10-05: interior-sample pixel proof plus a strict, Chrome-proven outline differential, in two sequential unstacked PRs. Native drawing is SVG-a2, off master after this one lands.

It adds an inline <svg> laid out as a block-level replaced box (REPL-a's model), drawing <path>, <rect> and <circle> with fill, stroke and stroke-width from CSS or presentation attributes. It is proven on web against Chrome, in both directions. Native refuses <svg> as SVG-a2.

What changed

• The svg property family (css/properties/svg.ts, one line in FAMILIES): fill, stroke and stroke-width, all inherited, with Chrome's grammars (gen-css-grammar overrides):

  • <paint> takes <color>, which webref's grammar omits.
  • stroke-width is one non-negative length, percentage or (SVG quirk) number. thin and comma lists are invalid in Chrome (probed).

  A unitless stroke-width computes to px. Animation kind is UNADMITTED. The native hit test models the family, since it paints no box; without that, the UA fill on html refused every interaction case.

• Elements and the UA dataset:

  • svg joins REPLACED_TAGS; path, rect and circle join SUPPORTED_TAGS.
  • The UA capture now creates SVG-namespace elements (createElementNS). It captures svg (overflow hidden) and the shapes (transform-origin: 0 0) into the replaced-key tables.
  • A shape is its <svg>'s content and never a box: the layout lowering already skips a replaced element's children, and the computed checks give shapes only the SVG checks.

• Attributes (analysis/elements/svg.ts):

  • Geometry: d, x, y, width, height, cx, cy, r, viewBox.
  • The paint presentation attributes, and an svg's or rect's width/height, become presentational hints.
  • Hints now beat inheritance too. They are a declaration below author rules (css-cascade-5 §6.1); before this, hints were only checked for non-inherited properties.

• Path data, viewBox and numbers (analysis/elements/svg-path.ts) are parsed as Chrome builds them:

  • Numbers are summed digit by digit in float32. For example, -0.785 is -0.7849999666213989, not the nearest float.
  • Relative points and smooth reflections are float32 sums.
  • Quadratics stay quadratics.
  • A moveto right after a moveto replaces it.
  • Arcs are refused as SVG-arc, because Chrome builds them with the platform sinf, cosf and atan2f.

• Bounds (packages/layout/src/svg-geometry.ts): Skia's ComputeTightBounds, its extrema and eval helpers, TrimmedBounds for paths of lines only, and gfx::BoundingRect's next-float growth. All are ported in float32 and registered in docs/ports.json (SkGeometry.cpp and .h, SkPathPriv.cpp and .h, rect_f.cc).

• The strict outline differential (parity/src/svg-compare.ts, part of the host layout lane). Every shape's getBBox, getScreenCTM, getBoundingClientRect and computed fill/stroke/stroke-width must equal Dragon's as doubles or strings, with no allowance.

  • The viewBox and client-rect arithmetic follow Chrome's observed order:
    • the limiting axis is scaled and the other is centred in user units;
    • corners are mapped to the border box in double, rounded to float32, then the origin is added in float32.
  • Both are pinned against captured CTMs and rects.
  • Five plants each fail it: control-point bounds, viewBox ignored, meet as slice, one-step client rect, and fill and stroke swapped.
  • The capture records getBBox and getScreenCTM for SVG shapes only, so every other capture is byte-identical apart from the new longhands.

• Refusals, each with a reject fixture or a unit test naming its package:

  • An inline <svg>: RF-INL, as for img.
  • A viewBox without width and height: SVG-ratio.
  • Non-px units and percentage stroke-width: SVG-units.
  • Arcs: SVG-arc.
  • <g>, text, nesting, shapes outside <svg>, and CSS geometry on shapes: SVG-b.
  • url() and context paints: SVG-paint.
  • Dash attributes: SVG-dash.
  • Cap, join, miter and fill-rule: SVG-stroke.
  • Opacities: SVG-opacity.
  • The transform attribute and preserveAspectRatio: SVG-transform.
  • <svg> on native: SVG-a2.

  In the parity lanes, native lays out the svg box without its shapes. As with SELD-R2's lane-only interaction, its rows prove nothing for native, and device samples inside an undrawn svg box are dropped (paint-samples/image.ts).

• Fixtures (group svg, ltr and rtl):

  • svg-basic: path with cubics, quads and smooth curves; rect; circle.
  • svg-viewbox: meet on each axis, offset viewBoxes, padding and border, CSS size over the attributes, no viewBox, the 300x150 default, and float32-sensitive decimals with exponents.
  • svg-paint: inheritance, currentcolor, attribute against CSS precedence, none, and a unitless stroke-width.
  • svg-flex: svg as a flex item.
  • Ten rejects.

Review round (precomputed review of 1f834fa)

• Medium: the hit lane. hitRefusal now refuses every case with an <svg> by name: "hit testing an svg's shapes comes with SVG-a2". Chrome hits the painted shapes (visiblePainted), which the hit test doesn't model yet.
  • hit-report.test.ts pins the refused set as exactly the transform cases plus the 8 svg cases.
  • rt-vectors/hit/facts.json is regenerated without them.
  • hit-report.test passes, including the pointerEventsNotInherited plant.
• Medium: attribute values bypassed the profile. An SVG paint attribute (fill, stroke, stroke-width) now resolves with a stand-in declaration at its element's start tag (Declaration.presentationHint). usedKeys therefore keys it like a CSS declaration, and the same rows decide both.
  • A value no context proves is reported as DRAGON_UNSUPPORTED_VALUE naming the attribute, as checkValues reports the CSS value.
  • New test: fill="#ff0000", fill="red", fill="rgba(255, 0, 0, 0.5)", fill="transparent" and stroke="#00f" are each blocked with exactly the diagnostics their CSS equivalents give. Proven rgb() and number values compile from either.
  • The regen now also derives stroke:none (paint/ltr and rtl) from svg-paint's stroke="none" attribute, and the tailwind sweep shows all 6 svg utilities compiling on web.
• Low: viewBoxTransform. The engine's viewBoxTransform (svg-geometry.ts) now uses Chrome's arithmetic: it branches on the ratio and centres in user units (with a slice flag for meetOrSlice). svg-compare uses it directly, so there is one implementation, pinned by the captured CTMs.
• Passed:
  • pnpm typecheck.
  • pnpm regen via /tmp/job.sh svga1-regen3, fixed point after 3 passes.
  • 49 targeted test files, 750 tests, now including packages/parity/test/hit-report.test.ts.

Rows promoted (web only, all exact, each in paint/ltr and paint/rtl)

• fill:<rgb()>, fill:currentcolor, fill:none
• stroke:<rgb()>, stroke:none
• stroke-width:<length-px>, stroke-width:<number>

No native rows: SVG-a2 adds them after the driver's device run.

The tailwind sweep shows 5 of its 6 svg utilities now compile on web.

Music-player north-star delta

Only the supported-tags list in the <button> refusal messages changes (it now names svg, path, rect and circle). The music player has no SVG in its tree, so nothing new compiles or is refused.

What passed

• pnpm typecheck.
• pnpm regen via /tmp/job.sh svga1-regen and svga1-regen2, reaching a fixed point. Earlier attempts stopped on lane checks (hit-test facts, lane-only profile refusals and the native frames check); I fixed those in source, as the commits say.
• 48 targeted test files, 741 tests, all passing. They include the new dragon/test/svg.test.ts and parity/test/svg.test.ts, chrome-ports, both registry-claims, the identity tests, the UA and paint-role tests and the android profile test.
• The svg fixtures passed both host lanes live in Chrome, including the strict differential. They also pass it offline against the committed captures in parity/test/svg.test.ts.
• The full pnpm test was not run, per the lane contract. Device lanes are the driver's job; native-lanes.ts records them as not yet run (device step pending).

Changed tests, and why

• dragon/test/block-elements.test.ts: the element table excludes the shape tags, which read replaced keys as the replaced tags do.
• dragon/test/ua-elb2.test.ts: the replaced-key tables list svg and the shapes. A new pin checks their UA values.
• dragon/test/s4b.test.ts: stroke-width gets two width values instead of the colour pair. The "(iii) colour conversion stays in css/color.ts" check and the same scan in dragon/test/ua.test.ts exempt analysis/elements/svg-path.ts by path. That follows the fonts, forms and media/viewport.ts precedent: the file reproduces Chrome's float32 path-number arithmetic, and svg.test.ts pins it.
• dragon/src/css/math.ts: stroke-width takes the number-or-length-percentage math grammar, as its syntax gives. This is source, but math-validity.test checks it.
• parity/test/replaced-identity.test.ts: the svg fixtures postdate REPL-a and hold a replaced leaf on purpose, so they are excluded from "earlier" fixtures. The count of REPLACED fixtures is still pinned at 6.
• parity/src/hit-capture.ts, used by hit-capture-identity: the identity strip removes the svg family's keys as well as pointer-events. Every base file therefore still differs from the base only by keys a package added, and svg.test.ts proves the family's own values.

Outside the spec

• packages/parity/src/native-compare.ts: the native frames check skips SVG shapes.
• packages/parity/src/paint-samples/image.ts: samples inside an undrawn replaced box are dropped.
• packages/dragon/src/analysis/interaction.ts: HIT_MODELLED includes the svg family.
• Shared hub files with additive hunks, which open PRs also touch:
  • project.ts and internal.ts: OVFL A 2/2: scroll containers in the compiler, viewport propagation and Chrome scroll metrics #97, TXT1a-1 (v2): the text-latin registry against Chrome, engine-mode real faces and R3 vectors #103, TXT1a-2 (v2): native real faces and the text-latin cases in FIXTURES, restacked onto INL1a C2 #104, TXT-W1 (v2): font-weight and font-style, restacked onto the text stack #106, TXT-W2 (v2): the font shorthand and font-synthesis, restacked onto the text stack #107, MQ-R1 1/2 (T067): @media bands for native in the compiler, the band runtime reference and Chrome resize traces #140, OVFL A: scroll containers (overflow auto, scroll, clip), viewport propagation and Chrome scroll metrics, rebuilt on master (supersedes #96/#97) #194, CASC 1: @supports decided at build time as Chrome 145 decides it; revert and revert-layer; CSS-wide keyword rows #200
  • resolve.ts: TXT-W1 (v2): font-weight and font-style, restacked onto the text stack #106, TDEC-a1 (v2): text-decoration in the compiler, restacked onto the text stack #108, INL2a 2/2: inline-block and inline-flex in the compiler, with Chrome fixtures #115, CASC 1: @supports decided at build time as Chrome 145 decides it; revert and revert-layer; CSS-wide keyword rows #200
  • computed-checks.ts: OVFL A 2/2: scroll containers in the compiler, viewport propagation and Chrome scroll metrics #97, TXT1a-2 (v2): native real faces and the text-latin cases in FIXTURES, restacked onto INL1a C2 #104–TDEC-a1 (v2): text-decoration in the compiler, restacked onto the text stack #108, INL2a 2/2: inline-block and inline-flex in the compiler, with Chrome fixtures #115, OVFL A: scroll containers (overflow auto, scroll, clip), viewport propagation and Chrome scroll metrics, rebuilt on master (supersedes #96/#97) #194, PNT1-effects 1/2: opacity as a native group alpha and z-index with Appendix E stacking (supersedes the opacity and stacking parts of #84-#85) #196, PNT1-radius: border-radius, its corner longhands and -webkit-border-radius drawn natively on iOS and Android (supersedes #80, #81) #198
  • capture-ua-defaults.ts: TXT-W1 (v2): font-weight and font-style, restacked onto the text stack #106, TXT-W2 (v2): the font shorthand and font-synthesis, restacked onto the text stack #107
  • pipeline.ts: TXT1a-1a (v2): Ahem through HarfBuzz, and the zoomInput platform rule, restacked onto INL1a C2 #101, TXT1a-1 (v2): the text-latin registry against Chrome, engine-mode real faces and R3 vectors #103, TXT1a-2 (v2): native real faces and the text-latin cases in FIXTURES, restacked onto INL1a C2 #104, OVFL A: scroll containers (overflow auto, scroll, clip), viewport propagation and Chrome scroll metrics, rebuilt on master (supersedes #96/#97) #194
  • compare.ts and native-compare.ts: INL2a 2/2: inline-block and inline-flex in the compiler, with Chrome fixtures #115
  • attributes.ts: TDEC-a1 (v2): text-decoration in the compiler, restacked onto the text stack #108, TDEC-a2 (v2): decoration geometry and the decoration capture, restacked onto the text stack #109
  • properties.ts: TDEC-a1 (v2): text-decoration in the compiler, restacked onto the text stack #108

Notes for the PM

• The spec named element:svg-style rows, but Dragon has no element-level rows. The svg element is gated by its refusals and proven by its fixtures.
• Presentation-attribute colour syntaxes are not row-gated (hints are not declarations). Each is proven per fixture by chrome-dual.

Reviewed diff: about 151 KB. All regenerated output is under .macroscope/ignore.md shapes. The spec is at /tmp/specs/svg-a.md.

Land-prep round (2026-10-08, cloud)

• Landing failure of 2026-10-06 (parity.test.ts:303, the colour-only plant: every problem matches /color/). It was fixed in 3fb061c: SVG-a1's fill and stroke have initial value black, so the colourOnly plant moves them as well. The check now requires every problem to name a colour longhand (COLOR_LONGHANDS) or fill or stroke. That is the exact set the plant can move, so the check is still judged on every problem and keys on the longhand name instead of a substring. parity.test.ts now passes on CI (532 tests).
• Test for 2011c9b (capture.ts: an SVG element is never an inline box). A new case in parity/test/svg.test.ts checks that no SVG shape's committed Chrome capture records an <id>:line<j> fragment, across all 8 svg cases, and that at least one shape exists. Without the fix, a shape's computed display is inline, so capture.ts would push one :line node per client rect.
• Merged origin/master three times:
  • 4d4b955: the cloud migration batch. No conflicts.
  • 7a7ba63: OVFL A: scroll containers (overflow auto, scroll, clip), viewport propagation and Chrome scroll metrics, rebuilt on master (supersedes #96/#97) #194 OVFL-a. Two conflicts:
    • computed-checks.ts: the walk takes OVFL's scroller and checkPercentRelative, and still sends an svg shape only to SVG-a1's checkSvg.
    • replaced-identity.test.ts: OVFL's LATER_WITH_REPLACED now also lists the svg layout fixtures, which replaces this branch's own exclusion. Both pins are kept: REPLACED is still 6, and the excluded set equals 6 + LATER_WITH_REPLACED.length.
  • 9b32f10: CASC 1: @supports decided at build time as Chrome 145 decides it; revert and revert-layer; CSS-wide keyword rows #200 CASC-supports. No textual conflicts; GitHub's conflict was in generated outputs, which the regen rebuilt.
• Regens on CI:
  • https://github.com/compiled-run/dragoncss/actions/runs/37831539343 and https://github.com/compiled-run/dragoncss/actions/runs/37840937873, both superseded by master merges.
  • https://github.com/compiled-run/dragoncss/actions/runs/37848611592: regen commit 7787d4b.

Review round: independent review of 7787d4b

• Medium: the svg differential was one-sided (svg-compare.ts). compareSvg walked only Dragon's scene shapes. Two callers also turned null svgScenes into [], so a shape Chrome paints but Dragon drops passed.
  • compareSvg now takes readonly SvgScene[] | null.
  • Every captured node with svg geometry must be one of Dragon's scene shapes; otherwise the problem reads "Chrome captured an SVG shape that no Dragon svg scene has".
  • Null scenes with captured shapes now fail.
  • The opposite direction (a scene shape Chrome did not capture) was already reported.
  • pipeline.ts and svg.test.ts pass the scenes through without ?? [].
  • New case in parity/test/svg.test.ts. It drops one shape from the first scene of each case (exactly one problem per case), empties every scene list, and returns null scenes. Each fails the differential. All three passed as [] before the fix.

Catch-up merges (2026-10-09)

• origin/master 508db67 (@charset "utf-8"; at the start of a sheet is a no-op on every target (Chrome drops it); other encodings and positions refused #208 charset, PNT1-radius: border-radius, its corner longhands and -webkit-border-radius drawn natively on iOS and Android (supersedes #80, #81) #198 PNT1-radius-native). Three conflicts, each resolved by keeping both sides:
  • parity/src/hit-capture.ts:
    • hitRefusal refuses a transform, then an <svg> by name (SVG-a2), then a rounded corner (PNT1), then INL1a's inline box or <br>.
    • withoutPointerEvents uses master's KEYS_SINCE_BASE, which now also lists the svg family (fill, stroke, stroke-width) beside the four corner radii. This replaces this branch's inline regex.
  • parity/test/hit-report.test.ts and parity/test/lanes.test.ts: both judge the exact union of the four named refusals (transform, PNT1 radius, INL1a inline, svg). Each refused case must carry its own reason. Both RADIUS_OUT and SVG_OUT are pinned as exact sets.
  • New case in parity/test/hit-capture-identity.test.ts. It checks that the svg family's keys are stripped beside the radii, while fill-opacity, stroke-dasharray and a changed value beside a stripped key still differ.
• origin/master 51a4888 (CTX-PROOF-2: inherit proven on the flex, box-sizing, margin, padding, border and size longhands (30 alias probes) #210 CTX-PROOF-2) and 981c068 (TXT1a-1a (v2): Ahem through HarfBuzz, and the zoomInput platform rule, restacked onto INL1a C2 #101 TXT1a-1a): no textual conflicts. GitHub's conflict was in generated outputs, which the regen rebuilt.
• Regens on CI, each finished:
  • https://github.com/compiled-run/dragoncss/actions/runs/37871598420 (regen 1cc2851)
  • https://github.com/compiled-run/dragoncss/actions/runs/37876508813 (regen 34ffb6b)
  • https://github.com/compiled-run/dragoncss/actions/runs/37880889006 (regen db9c1bf, superseded)
• origin/master 8a91837 (GEN-b: content and list-style longhands, list-style shorthand, markerless list items as blocks #218 GEN-b longhands, CASC 2: @property decided at build time as Chrome 145 registers it (initial values, inherits, invalid at computed-value time) #216 CASC 2 @property). Five conflicts, each resolved by keeping both sides:
  • dragon/src/css/properties.ts: FAMILIES keeps GEN-b's lists family, then SVG-a1's svg family last. Both imports are kept.
  • scripts/gen-css-grammar.ts: SUBSET keeps GEN-b's content and list-style* entries, then ...SVG_LONGHANDS last. RADIUS and the other entries are unchanged.
  • dragon/src/analysis/computed-checks.ts: checkSvg stays, and so does the walk's svg-shape early return. checkComputed takes GEN-b's signature (faults: GenBFaults = GEN_B_FAULTS), which it passes to checkUserAgentDefaults.
  • dragon/test/s4b.test.ts and dragon/test/ua.test.ts: the rounding-boundary scans keep both single-file exemptions, CASC 2's css/chrome-number.ts and SVG-a1's analysis/elements/svg-path.ts.
  • New test in dragon/test/svg.test.ts. One document holds a shape with a percentage stroke-width and a <ul><li> with a disc marker, and both refusals fire: SVG-a1's and GEN-b's GEN-c marker refusal.
• Regen on CI:
  • https://github.com/compiled-run/dragoncss/actions/runs/37889551569 (regen commit ee4a34e)
  • https://github.com/compiled-run/dragoncss/actions/runs/37894251668 for 0c6ada6, which only adds the test: already at a fixed point, no commit.

Review round: independent review of 0c6ada6

• Medium (parity/src/hit-capture.ts): the identity strip removed the svg family at any value. withoutPointerEvents had fill, stroke and stroke-width in KEYS_SINCE_BASE, so a base file that gained a non-initial value would still hash to the base.

  • Fix: the family is now stripped only at its initial values, fill: rgb(0, 0, 0), stroke: none and stroke-width: 1px, in LONGHANDS order, as GEN-b's neutral longhands are (SVG_JSON and SVG_CSS). KEYS_SINCE_BASE is back to pointer-events and the radii.

  • hit-capture-identity.test.ts: the svg case now checks two things.

    • The neutral triple is stripped, in both JSON and CSS.
    • A non-initial fill (rgb(30, 90, 200) or none), stroke (rgb(200, 40, 40)) or stroke-width (2px) each still differs from the base.

    The old any-value strip fails this: it removed those values.

  • The base-hash case still passes on CI, so the 2471 base files gained only the neutral triple.

• Mutation check of the new dragon svg case (the svg and GEN-b refusals both fire), run locally. It fails when the shapes' checkSvg call is removed, and it fails when GEN-b's list-marker refusal is disabled. It passes when both are restored.

Catch-up merge (2026-10-09, after #196 and #231 stalled)

• origin/master b97cf50 (ANIM-b1 native runtime 1/2: animation tables as typed literals, the translated animator on the state machine, display drivers #203 ANIM-b, TXT1a-1b (v2): real-font Latin through HarfBuzz in the engine, restacked onto INL1a C2 #102 TXT1a-1b, CASC 3: @layer decided at build time as Chrome 145 orders cascade layers #233 CASC 3 layers, GEN-c: list item ordinals and marker text (counter styles), compile-time #232 GEN-c counter text). One conflict:
  • dragon/src/css/stylesheet.ts: Declaration keeps both SVG-a1's presentationHint and CASC 3's layer.
  • A hint applies only when no author declaration wins (resolve.ts), layered or not, so every author layer still ranks above it (css-cascade-5 §6.1).
  • New test in dragon/test/svg.test.ts:
    • A rect's fill="rgb(9, 9, 9)" attribute gives fill (9, 9, 9).
    • A rule in @layer base overrides it.
    • Between two layers, the later one wins.
• Regens on CI:
  • https://github.com/compiled-run/dragoncss/actions/runs/37923589554 (regen 37f1cf0)
  • https://github.com/compiled-run/dragoncss/actions/runs/37934126509 for 565cdba, which only adds the test: already at a fixed point.
• PNT1-effects 1/2: opacity as a native group alpha and z-index with Appendix E stacking (supersedes the opacity and stacking parts of #84-#85) #196 and OVFL-b: native scrolling (UIScrollView and an Android scroll view sized by the engine) #231 have not landed (the Mac driver's regen environment fault). Whichever lands after SVG-a1: inline <svg> (path, rect, circle; fill, stroke, stroke-width) on web, proven by a strict outline differential against Chrome; native refused as SVG-a2 #206 needs a catch-up.

What passed on the clean head 565cdba

• pnpm typecheck (local).
• Platform-free, local:
  • dragon block-elements, casc-property, lists, s4b, s5, seams, svg, ua-elb2, ua, registry-claims and profile-index (11 files, 143 tests).
  • The latest batch's dragon tests: casc-layer, markers-text, anim-runtime, media-wiring and state-program (5 files, 118 tests).
• pnpm ci:regen svg-a1 --head 565cdbab1a: already at a fixed point. https://github.com/compiled-run/dragoncss/actions/runs/37934126509
• pnpm ci:test-files at https://github.com/compiled-run/dragoncss/actions/runs/37934142200, all passing:
  • parity: 564 tests
  • svg: 6
  • hit-report: 9
  • hit-capture-identity: 9
  • hit-lanes: 6
  • pixel-reference: 12
  • glyph-calibration: 8
  • line-breaks: 11
  • native-host: 347
  • device-lanes: 62
  • pnt2-quads: 5
  • pnt2-samples: 7
  • replaced-paint: 13
  • p6a-dash-oracle: 330
  • replaced-identity: 3
  • chrome-ports: 13
  • parity registry-claims: 8
  • android-profile: 5
  • casc-property: 3
  • casc-layer: 3
  • list-style-parse: 1
  • ovfl-metrics: 14
  • dist: 6
  • env-engine: 8
  • fixture-reader: 7
  • fonts-fixtures: 26
  • inline: 13
  • interaction-forced: 3
  • media-runtime: 8
  • media-sweep: 19
  • phrasing-blockified: 3
  • platform: 12
  • tree-faults: 3
  • values: 72
  • lanes.test.ts: 365 passed and 1 failed. The failure is "every device lane ran", which is device step pending.
• Superseded head 0633c38: https://github.com/compiled-run/dragoncss/actions/runs/37899278032
• pnpm pr:review 206 --once: exit 0. It reports UNREVIEWED (Macroscope at its spending limit), with 0 unanswered findings. GitHub reports the PR as clean.

🤖 Generated with Claude Code

---

Generated by Claude Code

[thejackshelton]

Landing stopped at step test (pnpm land)

packages/parity/test/parity.test.ts, rerun alone on a quiet machine, exited 1 (log /tmp/land-206-test-solo-1.log)
 ❯ packages/parity/test/parity.test.ts:303:57
    301|     expect(dual.boxesEqual).toBe(dual.boxesCompared);
    302|     expect(dual.channelsEqual).toBeLessThan(dual.channelsCompared);
    303|     expect(dual.problems.every((p) => /color/.test(p))).toBe(true);
       |                                                         ^
    304|   });
    305|

⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/1]⎯


 Test Files  1 failed (1)
      Tests  1 failed | 489 passed (490)
   Start at  02:32:58
   Duration  82.84s (transform 1.17s, setup 0ms, import 1.54s, tests 81.24s, environment 0ms)

Found by bisecting the batch #206 #201 (2 proofs): master passes, adding #206 fails.

Fix the cause, then hand #206 back to the landing queue with the new clean head.

[thejackshelton]

LAND-READY
READY svg-a1:206:205b6b35eff8dc147fe9331a3e6fda6839e55403

Lane receipt (PM session A): pnpm typecheck pass; platform-free dragon {block-elements,s4b,svg,ua-elb2,ua,registry-claims,profile-index} pass (local, at 205b6b3); ci:regen https://github.com/compiled-run/dragoncss/actions/runs/37856505325 (already at a fixed point at 205b6b3); ci:test-files https://github.com/compiled-run/dragoncss/actions/runs/37860530081 (tested 205b6b3 at 205b6b3: parity 532, svg 6, hit-report 9, replaced-identity 3, chrome-ports 13, parity registry-claims 8, android-profile 5, and the fixture-wide set p6a-dash-oracle 312, dist, env-engine, fixture-reader, fonts-fixtures, inline, interaction-forced, media-runtime, media-sweep, phrasing-blockified, platform, tree-faults, values all pass; lanes.test "every device lane ran" is device step pending); pr:review 206 --once exit 0 (UNREVIEWED: Macroscope spending limit).

Independent Claude review at 205b6b3 (Medium or higher only). The first review at 7787d4b found that the SVG differential was one-sided; 205b6b3 fixes it with a test that fails without the fix, and the re-review of that fix is clean:

{"pr":206,"head":"205b6b35eff8dc147fe9331a3e6fda6839e55403","findings":[]}

---

Generated by Claude Code

[thejackshelton]

LAND-READY
READY svg-a1:206:565cdbab1af58b7712c0d4f6a647240bdc93cfa1

Lane receipt (PM session A): master merged through b97cf50 (#208, #198, #210, #101, #218, #216, #203, #102, #233, #232; resolutions in hit-capture.ts, hit-report/lanes tests, properties.ts FAMILIES, gen-css-grammar SUBSET, computed-checks.ts, s4b/ua exemptions, stylesheet.ts Declaration with both presentationHint and layer). pnpm typecheck pass; 16 platform-free files pass (local, at 565cdba); ci:regen https://github.com/compiled-run/dragoncss/actions/runs/37934126509 (already at a fixed point at 565cdba); ci:test-files https://github.com/compiled-run/dragoncss/actions/runs/37934142200 (tested 565cdba at 565cdba: 35 files incl. parity 564, svg, hit-capture-identity, hit-lanes, casc-layer, pixel-reference, native-host and the fixture-wide set, all pass; lanes.test "every device lane ran" is device step pending); pr:review 206 --once exit 0 (UNREVIEWED: Macroscope spending limit).

Note for the driver: #206 conflicts with #196 (math.ts, s4b.test.ts, hit-capture.ts, hit-refusals.ts, hit-report.test.ts, lanes.test.ts; all additive) and #231 (layout/src/index.ts). Whichever lands later needs its lane's catch-up.

Independent Claude review history (Medium or higher only): at 7787d4b (one-sided SVG differential) and 0c6ada6 (svg family stripped at any value in the identity hash), both fixed with tests that fail without the fix; re-reviews clean at 205b6b3 and 0633c38; the final merge with #233/#232/#203/#102 and the layer-vs-hint test are clean at 565cdba:

{"pr":206,"head":"565cdbab1af58b7712c0d4f6a647240bdc93cfa1","findings":[]}

---

Generated by Claude Code

[thejackshelton]

Landing stopped at step regen (pnpm land)

pnpm regen exited 1 (log /tmp/land-206-regen.log)
Error: Cannot find package '/private/tmp/dragon-land/packages/dragon/node_modules/css-tree/index.js' imported from /private/tmp/dragon-land/packages/dragon/src/css/properties/radius.ts
    at legacyMainResolve (node:internal/modules/esm/resolve:201:26)
    at packageResolve (node:internal/modules/esm/resolve:774:12)
    at moduleResolve (node:internal/modules/esm/resolve:855:18)
    at defaultResolve (node:internal/modules/esm/resolve:988:11)
    at #cachedDefaultResolve (node:internal/modules/esm/loader:697:20)
    at #resolveAndMaybeBlockOnLoaderThread (node:internal/modules/esm/loader:714:38)
    at ModuleLoader.resolveSync (node:internal/modules/esm/loader:746:52)
    at #resolve (node:internal/modules/esm/loader:679:17)
    at ModuleLoader.getOrCreateModuleJob (node:internal/modules/esm/loader:599:35)
    at ModuleJob.syncLink (node:internal/modules/esm/module_job:162:33) {
  code: 'ERR_MODULE_NOT_FOUND'
}

Node.js v24.15.0

Fix the cause, then hand #206 back to the landing queue with the new clean head.