feat(lint): add rotation_pivot_drift layout check

[xuanruli]

What

New cross-sample layout check rotation_pivot_drift — flags a rotating element that should spin in place but pivots about the wrong point (e.g. a wheel whose spokes use a hardcoded px transformOrigin instead of svgOrigin/%, so they swing off-center while every existing check still passes).

Motivating prod case: a portrait ad's spoked-wheel whose #spokes rotated about transformOrigin:"250px 250px" in a resized 460px container — spokes detached from the hub, shipped clean because no rule inspects rotation.

How

• layout-audit.browser.js: window.__hyperframesRotationSample() reports each visible transformed element's bbox center + decoded rotation angle per layout sample. Skips [data-layout-allow-orbit].
• checkPipeline.ts: accumulates samples across the seek grid; detectRotationPivotDrift() (modeled on detectSweepStatic) flags an element that (a) actually spins (angle spread > 20° over ≥3 samples), (b) is size-stable (bbox width ratio ≤ 1.6), and (c) whose bbox center drifts > max(10% of its size, 2% of min viewport dim). Emits warning; not persistence-tiered (not demoted to info).

FP guards

Real rotation required, ≥3 samples, size stability, data-layout-allow-orbit exemption, min area ~2500px². Center-drift (not bbox size) is the discriminator, so a correctly-centered spinner reads drift ≈ 0.

Validation (check --json)

Fixture	Expected	Result
off-transformOrigin spoked wheel	fire	fired — 109px drift on #spokes
non-spinning comps (node diagram, device tree)	clean	clean, no FP
correctly-centered spinner (svgOrigin)	clean	clean (spins 162°, drift 0)
data-layout-allow-orbit off-origin spinner	clean	clean (exempt)
off-svgOrigin control, no opt-out	fire	fired — 251px drift

No false positives. tsc --noEmit clean, oxlint clean, check.test.ts + layout-audit.browser.test.ts = 112/112 pass.

Note

ROTATION_MAX_SIZE_RATIO is 1.6 (not 1.3): a rotating anisotropic shape's axis-aligned bbox inherently oscillates (8-spoke star ~1.32×, square 1.41×), so a tighter ratio rejects legitimate targets. Center-drift stays the real discriminator; thin swinging bars are excluded.

Follow-up: a detectRotationPivotDrift unit test via the fake driver's collectRotationSample (mirroring the sweep_static tests).

🤖 Generated with Claude Code

[xuanruli]

• feat(lint): add rotation_pivot_drift layout check #2741 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[miga-heygen]

SSOT Review: rotation_pivot_drift lint check

PR: feat(lint): add rotation_pivot_drift layout check — 6 files, +239/-1

---

Architecture

Clean addition that follows the established cross-sample pattern (detectSweepStatic precedent):

1. Browser-side (layout-audit.browser.js): __hyperframesRotationSample() collects every rotatable candidate's bbox center, size, and decoded rotation angle per seeked sample. Caps at 200 candidates, skips [data-layout-allow-orbit], filters by visibility and minimum area.

2. Driver bridge (checkBrowser.ts): collectRotationSample() evaluates the browser function, validates each field via parseRotationSample (using existing isRecord/stringValue/numberValue helpers — no new parsing invented). Returns RotationSample[].

3. Pipeline (checkPipeline.ts): samples accumulated in GridSamples.rotationSamples inside the main mergeSampleTimes loop, grouped by selector post-run, then detectRotationPivotDrift() runs the five-stage filter chain.

4. Types (checkTypes.ts): RotationSample interface, driver method signature. Issue code (layoutAudit.ts + checkBrowser.ts LAYOUT_ISSUE_CODES): both updated.

SSOT checks

1. RotationSample defined once in checkTypes.ts, imported by pipeline and browser. No duplication. ✓

2. Issue code rotation_pivot_drift registered in both places — the type union (LayoutIssueCode) and the runtime code list (LAYOUT_ISSUE_CODES). ✓

3. rotationAngleDeg extracts rotation from matrix(a,b,...) via atan2(b, a). Correct for both matrix() and matrix3d() (first two values are the 2D rotation cosine/sine in both). Returns null for non-transformed elements — browser-side skips them. ✓

4. maxAngleSpread is wrap-aware. Math.min(raw, 360 - raw) picks the shorter arc. Edge case: atan2 returns [-180°, 180°], so two angles -170° and 170° yield abs(-170-170) % 360 = 340, min(340, 20) = 20°. Correct. ✓

5. No new helpers that shadow existing ones. median() and maxCenterDrift() are local to the pipeline file. median is small and used only in detectRotationPivotDrift — not worth extracting to a shared util. ✓

6. Division-by-zero guard. minWidth <= 0 check before Math.max(...widths) / minWidth. ✓

FP guard chain (the real engineering here)

#	Guard	Purpose
1	data-layout-allow-orbit	Intentional orbits opt out (browser-side)
2	≥3 samples	Cross-sample confidence
3	angle spread > 20°	Filters static tilts from real rotation
4	size ratio ≤ 1.6	Excludes scale/entrance animations
5	median area ≥ 2500px²	Skips decorative spinners
6	drift > max(10% size, 2% viewport)	Relative threshold, not absolute

The 1.6 size ratio is well-motivated — a square's AABB naturally oscillates 1.41x between flat and 45°, and 8-spoke stars hit ~1.32x. The ceiling needs to sit above that. Documented in both code comments and PR body. ✓

Observations (non-blocking)

• Unit tests for detectRotationPivotDrift are deferred. The PR body calls this out as a follow-up ("via the fake driver's collectRotationSample"). The fake driver stub is already wired (collectRotationSample: vi.fn(async (_time: number) => [])), so the follow-up has a clean landing spot. The five-stage filter chain and the tuning constants would benefit from direct unit coverage — I'd recommend prioritizing that follow-up.

• CANDIDATE_CAP = 200 in the browser function is a reasonable safety valve. Worth noting: if a composition has >200 transformed elements, the cap could suppress legitimate findings on later-DOM-order elements. Unlikely in practice, but the cap is undocumented in the issue message — a composition hitting it would see silence, not a warning about being capped. Minor.

Blocking SSOT issues

None found.

Verdict

Approve. This follows the established cross-sample lint pattern precisely, the FP guards are thorough and well-motivated, the type surface is clean, and the browser→pipeline→issue code chain has no gaps. The deferred unit tests are called out and have a ready landing spot. Ship it.

— Miga

[miguel-heygen]

Found one blocking false-positive path not covered by the existing review.

• packages/cli/src/utils/checkPipeline.ts:580: the size-stability guard checks only w, even though the comment and PR contract say it excludes scale/entrance animations. A rotating element whose width stays at 100px while height scales 50→100→150 passes the ratio guard (1.0), has enough area and angle spread, and its AABB center can move 50px from a top-anchored scale. On a 1000px canvas that exceeds the 20px threshold, so this emits rotation_pivot_drift and recommends a pivot rewrite even though the movement is caused by height-only scaling. Since the code explicitly notes that false positives feed destructive downstream auto-fixes, check stability in both dimensions (or use a scale-invariant rigid-size test) and pin this height-only scale case as clean.

The same complex five-stage filter currently has no direct positive/negative unit coverage, and the Fallow audit is failing on its uncovered CRAP score. Please add focused tests for a real bad pivot, a centered spinner, wrap-around angles, width/height scaling, and the threshold boundaries rather than deferring them.

— Magi

Verdict: Request changes.
Reasoning: The width-only stability guard admits height-only scale/entrance motion as pivot drift, risking a destructive false-positive auto-fix; both dimensions and direct regression coverage are required before approval.

[james-russo-rames-d-jusso]

Reviewed at 341a5f3ec1b0172d9b69b406f31238a50b31180d.

Clean cross-sample finder following the detectSweepStatic precedent. Miga's SSOT / matrix-decode / wrap-arithmetic audit covers the algorithmic verification well; concurring on those. Adding four deltas — one hitting current CI, three sub-blockers.

1. Fallow CI is failing on this PR's new function's CRAP score.

detectRotationPivotDrift hits CRAP 37.1 (threshold 30, cyclomatic 11) — see the sticky Fallow comment. Since fallow audit runs with --fail-on-issues and --gate new-only, this is a genuinely new-code finding and the job is exiting 1. Two ways out (inline):

• Factor the five-stage guard into named predicates (isSpinning(group), hasStableSize(group), hasSizableMedian(group), driftExceedsFloor(group, canvas)) — drops cyclomatic below the threshold and reads a lot closer to the FP-guard table you already have in the doc-comment.
• Or bump the CRAP threshold if the team has decided this class of guard-chain function is fine as-is.

2. Concurring with Miga on the two she surfaced: unit tests deferred (the fake driver stub is already wired, so the follow-up landing spot is clean — but 6 tunable thresholds without regression coverage is a real gap for future tweaks), and CANDIDATE_CAP=200 is a silent-drop pattern that could suppress legitimate findings on late-DOM-order rotating elements without any signal to the operator. Small inline notes on each.

3. Selector stability across seeks. selectorFor() falls back to parent > tag:nth-of-type(N) when an element has no id, no data-layout-*, and no classes. N shifts as siblings render/unrender across the seek grid, so two different rotating elements at different times can alias to the same selector and get grouped. Fringe (author-crafted rotating figures almost always have ids/classes) and the FP guards can't detect it. Worth documenting the invariant "grouped selectors are assumed stable across seeks" at the top of detectRotationPivotDrift.

Verified-safe on my pass (no finding):

• matrix3d(...) values[0]/[1] correctly recover the in-plane rotation regardless of composed rotateY/rotateX (the leading two entries of the 4x4 are still cos θ, sin θ for the Z-axis rotation component).
• element.closest("[data-layout-allow-orbit]") correctly includes element itself — self-attributed opt-outs work.
• Wrap-aware maxAngleSpread handles atan2's ±180° discontinuity (-179° vs +179° reads as ~2°, not 358°).
• dataAttributes: {} on the finding matches the established pattern for cross-sample findings (sweep_static, motion_*, connector_detached all pass {}); not a regression.
• Fake driver stub (collectRotationSample: vi.fn(async () => [])) correctly satisfies the extended CheckAuditDriver interface.
• LayoutIssueCode union and runtime LAYOUT_ISSUE_CODES array both updated in the same PR — no serialization mismatch.

Nits (non-anchored):

• Cross-repo severity comment (checkPipeline.ts:538, "EF promotes this to error separately") — worth a link/issue reference so future readers can trace the severity contract without hunting in EF.
• The two size floors — browser <= 400px² and pipeline < 2500px² median — are intentional (network-noise cutoff vs the real filter) but not tied together in prose; a brief pointer in the browser comment would make it obvious the "real" cutoff is downstream.
• fixHint prescribes "use svgOrigin/percent, not hardcoded px" but the DOM sample only sees the composed transform — the check flags the symptom (bbox drift), not the specific authoring cause. Softening to "check transformOrigin/svgOrigin" (matching the message field's phrasing) would be more accurate.

LGTM from my side once Fallow is settled and the deferred unit-test follow-up has a tracker. Leaving as a comment — you can stamp when ready.

— Review by Rames D Jusso

[james-russo-rames-d-jusso]

🟠 Fallow CI is failing here — CRAP 37.1 vs threshold 30 (cyclomatic 11). The --gate new-only + --fail-on-issues combo means this is genuinely new-code complexity that blocks CI, not tool noise.

Two ways out:

1. Factor the five-stage guard into named predicates that mirror your doc-comment's FP-guard table — something like:

   for (const group of bySelector.values()) {
     if (!hasEnoughSamples(group)) continue;
     if (!isSpinning(group)) continue;
     if (!hasStableSize(group)) continue;
     if (!hasSizableMedian(group)) continue;
     const drift = maxCenterDrift(group);
     if (drift <= driftFloor(group, canvas)) continue;
     const finding = rotationDriftFinding(group, drift);
     if (finding) findings.push(finding);
   }

   That drops cyclomatic below the threshold and reads closer to the guard-table.
2. Or bump the CRAP threshold if guard-chain finders are considered fine as-is.

Not a design blocker — it's just CI arithmetic — but the current SHA won't merge until it's resolved.

— Rames D Jusso

[james-russo-rames-d-jusso]

🟡 Deferred unit tests for the detector — concurring with Miga, adding rationale.

The function has 6 tunable thresholds (ROTATION_MIN_SAMPLES, ROTATION_MIN_ANGLE_SPREAD_DEG, ROTATION_MAX_SIZE_RATIO, ROTATION_MIN_MEDIAN_AREA_PX, ROTATION_DRIFT_SIZE_FRACTION, ROTATION_DRIFT_VIEWPORT_FRACTION) whose values are load-bearing for FP suppression. Without direct unit coverage, any future tweak (e.g. tightening 1.6→1.4) lands silent — CI can't tell whether legitimate targets still fire or whether the FP-guard corners still hold.

A table-driven test covering the guard-chain corners (below-min-samples, static tilt, size-not-stable, tiny-median-area, drift-below-threshold, positive-fire) would pin the current tuning as the SSOT. The fake driver stub (check.test.ts:152) already gives you the entry point; consider prioritizing this before the check ships to enough compositions that a threshold drift becomes hard to bisect.

— Rames D Jusso

[james-russo-rames-d-jusso]

🟡 CANDIDATE_CAP = 200 is a silent-drop — concurring with Miga.

If a composition has >200 transformed visible elements, DOM order determines who survives; anyone past the cap gets no rotation sample, no warning, no telemetry. Unlikely for author-crafted comps, but the mode of failure is silent (a legitimate rotating spoked-wheel late in the DOM would just… not fire).

A couple of cheap mitigations if you want to close it in this PR:

• Sort candidates by bbox area desc before capping — largest rotating figures survive first, which is where the real prod-case lives.
• Or surface a one-line info-level finding when samples.length === CANDIDATE_CAP so the operator at least sees "rotation sampling was capped, some elements were skipped."
• Or raise the cap — 200 is arbitrary; if the whole payload is {selector, cx, cy, w, h, angle} (~50 bytes/entry) even 1000 candidates is <50KB per sample.

Nit-shaped, but leaving here so it doesn't get lost.

— Rames D Jusso

[james-russo-rames-d-jusso]

🟡 selectorFor() nth-of-type fallback isn't guaranteed stable across seeks.

When an element has no id, no data-layout-name/data-composition-id/data-start, AND no classes, selectorFor falls back to parent > tag:nth-of-type(N) (see layout-audit.browser.js:80-84). N counts live siblings at the current seek — so if a rotating element's sibling renders/unrenders across the seek grid (very common with entrance/exit animations), N shifts, and the same rotating element can collect samples under two different selectors OR two different rotating elements can alias to the same selector.

Cross-contamination case: Element A rotates from t=0-2 (nth=3). Element B rotates from t=1-3 (nth=3 after A exits). They'd be grouped together under identical parent > tag:nth-of-type(3), and the maxCenterDrift across A and B combined can trip the finding falsely.

Fringe (author-crafted rotating figures almost always have ids/classes/data-layout-name), and the FP guards can't detect it. Worth a top-of-function invariant note like:

// Assumes selector stability across seeks — rotating elements without stable ids/classes are on their own here.

Or, if you want to close it structurally, gate the browser-side sampler to only report elements where selectorFor didn't hit the nth-of-type fallback.

— Rames D Jusso

[xuanruli]

Pushed 6aff5734 addressing the blockers.

1. Size-stability FP (width-only guard) — fixed. isRotationSizeStable now bounds the bbox ratio in both width and height. The exact path you called out — fixed w=100, top-anchored height scaling 50→100→150 — now fails the height ratio (3.0 > 1.6) and is dropped before the drift test, so it no longer emits rotation_pivot_drift / recommends a destructive pivot rewrite. Direct regression test: does not fire when height scales (top-anchored) even though the AABB center moves.

2. Fallow CRAP (37.1 → under threshold) — refactored + covered. Split the guard chain into named single-decision predicates (hasEnoughRotationSamples / isActuallySpinning / isRotationSizeStable / isSizableRotation), so detectRotationPivotDrift reads as candidate filter → drift test. Exported it and added checkPipeline.rotationPivotDrift.test.ts — 11 direct positive/negative cases covering all five stages (count, spin, size-stable in each axis + degenerate zero-dim, sizable, center-drift), the viewport-floor path, multi-selector, and empty input. Every branch is exercised, so CRAP collapses toward the (now much lower) cyclomatic complexity.

Non-blocking, acknowledged as follow-ups:

• CANDIDATE_CAP=200 silent drop (Miga): agreed, will emit a debug log when the browser-side sampler truncates so a >200-transformed-element scene isn't silently under-audited.
• selectorFor nth-of-type instability across seeks (classless fringe element): good catch. Worst case is a group splitting across selectors → a missed detection, never a false positive, so it doesn't gate correctness here. Tracking separately for a stable-anchor fix.
• Deferred fake-driver end-to-end test: the new direct unit tests cover the finder logic; happy to add the collectRotationSample-driven runAuditGrid case on top if you'd prefer it in this PR.

tsc --noEmit clean, oxlint clean, 11 new + 40 existing check tests pass. Re-requesting review.

[jrusso1020]

Re-review at 6aff573 — my earlier comment was at 341a5f3. Prior blockers verified resolved; one new CI failure now gates merge.

Prior blockers — resolved

• Miguel's width-only size-stability FP (the CHANGES_REQUESTED blocker): fixed. isRotationSizeStable (checkPipeline.ts:591) now bounds the bbox ratio on both axes (maxW/minW ≤ 1.6 && maxH/minH ≤ 1.6) with a minWidth <= 0 || minHeight <= 0 degenerate guard. The exact path he called out — fixed w=100, top-anchored h=50→100→150 — now fails the height ratio (3.0 > 1.6) and is dropped before the drift test. Pinned by the does not fire when height scales (top-anchored)… regression test. ✓
• Fallow CRAP (my CI blocker at 341a5f3): resolved — Fallow audit is green on this SHA. The guard chain is split into hasEnoughRotationSamples / isActuallySpinning / isRotationSizeStable / isSizableRotation behind isRotationDriftCandidate, so detectRotationPivotDrift now reads as candidate-filter → drift-test and the cyclomatic score drops under threshold. ✓
• Deferred unit tests (raised by Miga, Miguel, and me): resolved — new checkPipeline.rotationPivotDrift.test.ts (11 cases) exercises all five guard stages, both size axes, the degenerate zero-dim case, the viewport-floor path, multi-selector independence, and empty input. detectRotationPivotDrift is exported for direct testing. ✓

New blocker — gating merge

• Preflight (lint + format) is RED on this SHA: oxfmt --check fails on packages/cli/src/utils/checkPipeline.ts ("Format issues found in above 1 files", exit 1 — job log). oxlint itself is clean (0 warnings / 0 errors), so the PR comment's "oxlint clean" is accurate but format:check was missed. Run bun run format (oxfmt) on checkPipeline.ts and push — this is the only code-side thing blocking merge.

Still pending / notes

• The CI workflow is still in_progress: Lint, Fallow audit, SDK, and File size check are green, but Test / Typecheck haven't finished, so the "11 new + 40 existing pass" claim isn't yet confirmed by CI. Worth a glance once it lands.
• preview-regression is red because preview-parity returned skipped (PREVIEW_PARITY_RESULT: skipped) and the gate fails closed — the parity job didn't execute (infra/approval), not a code defect. Should re-run green after the reformat push.
• Minor coverage gap: of Miguel's explicit test list, wrap-around angles is the one case not pinned by the new unit tests (angles used are 0/90/180 etc., no ±180° boundary). The wrap arithmetic in maxAngleSpread is verified correct by inspection (Miga's audit + my earlier pass), so this is nice-to-have — one extra it() with e.g. angle: -170 → 170 would close it. Not blocking.

Verdict: COMMENT
Reasoning: Every prior design blocker (Miguel's dual-axis FP guard, my Fallow CRAP, and the deferred unit tests) is resolved and verified at 6aff573. Not approving because a required check is red — merge is gated on oxfmt/Preflight going green (plus Miguel clearing his standing CHANGES_REQUESTED and CI Test finishing). On the code merits this is ready once formatting is fixed.

— Jerrai

[xuanruli]

Pushed e53e1a40.

Merge blocker — oxfmt/Preflight — fixed. Ran oxfmt on checkPipeline.ts; format:check is now clean. Thanks for catching that oxlint≠format:check — my earlier "oxlint clean" was accurate but incomplete.

In-file nits (@jrusso1020 / @james-russo-rames-d-jusso) — done in this commit:

• fixHint/message softened — both now phrase the finding as the symptom (bbox-center drift → "check transformOrigin/svgOrigin") rather than prescribing "use svgOrigin/percent, not px", which the composed-transform sample can't actually confirm as the cause.
• Selector-stability invariant documented — moved the detector doc-comment above detectRotationPivotDrift and spelled out that grouping assumes cross-seek selector stability; the nth-of-type(N) fallback on a classless element can mis-group (missed detection, or a rare exit-then-enter aliasing FP). Noted the structural fix = a stable-anchor gate on the browser sampler.
• Wrap-around coverage — added the two it()s @jrusso1020 flagged as the gap: a -175°↔175° boundary wobble reads as ~10° and does not fire; a real 170°→-100°→-10° spin across the seam still fires.

Still deferred (non-blocking, tracked):

• CANDIDATE_CAP=200 silent drop — I like the "sort by bbox area desc before capping" mitigation best (largest rotating figures survive, which is where the real prod case lives) + a one-line info finding when truncated. It's browser-side (layout-audit.browser.js) with its own test surface, so I'd rather land it as a focused follow-up than widen this PR — unless you'd prefer it here.
• Fake-driver runAuditGrid end-to-end test — the 13 direct unit cases now cover the finder + all six thresholds; happy to add the driver-level case on top if wanted.

Verified locally at e53e1a40: oxfmt --check clean, tsc --noEmit clean, oxlint clean, 13 rotation + 40 check tests pass. CI Test/Typecheck should confirm. @miguel-heygen — this should clear your standing changes-requested; re-requesting.

[james-russo-rames-d-jusso]

Delta from 341a5f3ec → e53e1a40886.

Every R1 open item closed:

• Fallow CRAP. Guard chain factored into 5 named predicates (hasEnoughRotationSamples / isActuallySpinning / isRotationSizeStable / isSizableRotation / isRotationDriftCandidate); Fallow audit is now green.
• Unit tests. 12-case checkPipeline.rotationPivotDrift.test.ts (137 lines) covering all five guard stages plus the degenerate zero-dim path, the viewport-floor branch, multi-selector independence, and empty input. The tuning constants are now pinned by regression coverage.
• Selector-stability invariant. Documented in the detectRotationPivotDrift doc-comment with the exact fringe (nth-of-type shift on classless elements) and the structural fix (stable-anchor gate on the browser sampler) called out for a future follow-up.
• Magi's dual-axis FP. isRotationSizeStable (checkPipeline.ts:575) now bounds both w and h ratios plus a minWidth <= 0 || minHeight <= 0 degenerate guard; the top-anchored height-scale regression is pinned by does not fire when height scales (top-anchored)….
• James's oxfmt gate. Preflight (lint + format) now green.
• James's wrap-around coverage nice-to-have. Xuanru went further and added BOTH the ±180° boundary wobble (should NOT fire) and real spin that crosses the ±180° boundary (SHOULD fire) tests.

CANDIDATE_CAP = 200 and the fixHint / EF-severity / two-tier-floor nits weren't taken — all author-choice territory; not blocking.

CI at e53e1a40886 is fully green (37 SUCCESS, 8 SKIPPED, 0 in-progress, 0 failed). mergeStateStatus is BLOCKED only because Magi's R1 CHANGES_REQUESTED is still standing — she'll want to re-review or dismiss to unblock the merge.

LGTM from my side.

— Review by Rames D Jusso

[miguel-heygen]

Re-reviewed the exact head e53e1a40886faafdda6f28f7382db9469c05b0e6 against my R1 blocker and the full follow-up delta.

The false-positive path is closed: size stability now checks both axes, so the height-only top-anchored scaling case that previously looked like pivot drift is rejected, and that exact regression is pinned in the new direct suite. The factored guard chain is substantially easier to audit, and the focused coverage now exercises the real drifting case, centered rotation, insufficient/low-angle samples, width and height scaling, zero/tiny geometry, viewport-floor behavior, selector isolation, and both near-seam wobble and a real rotation crossing ±180°.

Local exact-head verification: 125 focused tests passed (checkPipeline.rotationPivotDrift, check, and browser layout-audit suites), CLI typecheck passed, changed-file oxlint/oxfmt checks passed, and git diff --check passed. GitHub is also fully green at this head: 37 successful checks, 8 intentionally skipped, 0 failed or in progress. No new findings.

— Magi

Verdict: APPROVE
Reasoning: The dual-axis guard and direct boundary-focused tests close the prior correctness blocker, while the refactor clears the complexity gate without changing the intended detection pipeline; exact-head local and CI verification are green.