## Summary
Standalone WebGL2 rendering engine for client-side GLSL shader preview,
with utility functions that mirror the backend `nodes_glsl.py` detection
logic.
## Changes
- **What**: New `GLSLPreviewEngine` class (OffscreenCanvas + WebGL2),
`glslUtils.ts` (detectOutputCount, detectPassCount,
hasVersionDirective), and unit tests
- **GLSLPreviewEngine**: Fullscreen triangle via `gl_VertexID` (no VBO),
ping-pong FBOs for multi-pass rendering, MRT via `gl.drawBuffers()`,
blob output via `canvas.convertToBlob()`
- **glslUtils**: Pure functions ported from backend Python to
TypeScript, regex-based detection matching `_detect_output_count()` and
`_detect_pass_count()`
## Review Focus
- WebGL2 resource lifecycle (context loss, texture cleanup, FBO teardown
in `dispose()`)
- Ping-pong FBO logic for multi-pass shaders
- Engine tests are WebGL2-gated (`describe.skip` in happy-dom) — they
run in real browser environments
## Stacked PR
PR 2 of 3. Stacked on #9198 (fix: GLSLShader preview promotion).
PR 3: `feat/glsl-live-preview` (composable + LGraphNode.vue integration)
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9200-feat-GLSLPreviewEngine-GLSL-utility-functions-3126d73d3650812fadc6df4a26387d0e)
by [Unito](https://www.unito.io)
## Summary
Adds handling for entering app mode with an empty graph prompting the
user to load a template as a starting point
## Changes
- **What**:
- app mode handle empty workflows, disable builder button, show
different message
- fix fitView when switching from app mode to graph
## Review Focus
Moving the fitView since the canvas is hidden in app mode until after
the workflow is loaded and the mode has been switched back to graph, I
don't see how this could cause any issues but worth a closer eye
## Screenshots (if applicable)
<img width="1057" height="916" alt="image"
src="https://github.com/user-attachments/assets/2ffe2b6d-9ce1-4218-828a-b7bc336c365a"
/>
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9393-feat-App-mode-empty-graph-handling-3196d73d3650812cab0ce878109ed5c9)
by [Unito](https://www.unito.io)
## Summary
When ComfyUI Manager is disabled, OSS local users see a hint in the
Missing Nodes panel explaining how to install and enable it.
## Changes
- **What**: Added an inline hint in `MissingNodeCard` that renders when
the user is on OSS (non-Cloud) and Manager is not active
(`showInfoButton` is false). The hint shows the pip install command and
the `--enable-manager` startup flag, formatted as inline `<code>`
snippets via `i18n-t` interpolation.
## Review Focus
- The `showManagerHint` computed is intentionally simple: `!isCloud &&
!props.showInfoButton`. `showInfoButton` is the existing signal for
whether Manager is available/enabled.
- Styling uses existing semantic tokens (`bg-comfy-menu-bg`,
`text-comfy-input-foreground`) to match the rest of the panel.
## Screenshot
<img width="642" height="452" alt="image"
src="https://github.com/user-attachments/assets/d08280d3-b4a0-4613-b092-1baa49f0b091"
/>
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9377-feat-add-manager-enable-hint-for-OSS-local-users-3196d73d365081a19037c8f55f11d1eb)
by [Unito](https://www.unito.io)
## Summary
Enable `better-tailwindcss/no-deprecated-classes` lint rule and auto-fix
all 103 violations across 65 files. First PR in a stacked series for
#9300.
## Changes
- **What**: Replace deprecated Tailwind v3 classes with v4 equivalents:
- `rounded` → `rounded-sm` (85)
- `flex-shrink-0` → `shrink-0` (16)
- `flex-grow` → `grow` (2)
- Enable `no-deprecated-classes` as `'error'` in eslint config
- Update one test asserting on `'rounded'` class string
## Review Focus
Mechanical auto-fix PR — all changes produced by `eslint --fix`. No
visual or behavioral changes (Tailwind v4 aliases these classes
identically).
Fixes#9300 (partial — 1 of 3 rules)
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9417-fix-enable-no-deprecated-classes-tailwind-lint-rule-31a6d73d3650819eaef4cf8ad84fb186)
by [Unito](https://www.unito.io)
Co-authored-by: Alexander Brown <drjkl@comfy.org>
## Summary
Perf report workflow fails on fork PRs because `GITHUB_TOKEN` is
read-only for forks, causing "Resource not accessible by integration" on
the PR comment step.
## Changes
- **What**: Split `ci-perf-report.yaml` into a data-collection workflow
+ a `workflow_run`-triggered reporter (`pr-perf-report.yaml`), matching
the existing `ci-size-data`/`pr-size-report` pattern. Added fork PR
permissions guidance to `.github/AGENTS.md`.
- **ci-perf-report.yaml**: Removed the `report` job and `pull-requests:
write` permission. Added PR metadata (number + base branch) artifact
upload.
- **pr-perf-report.yaml** (new): Triggered by `workflow_run` on the perf
workflow. Downloads metrics + metadata artifacts, generates report,
posts PR comment with write permissions from the default-branch context.
## Review Focus
- The two-workflow split follows the same pattern as `ci-size-data.yaml`
→ `pr-size-report.yaml`, which already works for fork PRs.
- The `workflow_run` trigger runs in the base repo context per [GitHub
Security Lab
guidance](https://securitylab.github.com/resources/github-actions-preventing-pwn-requests/),
so it safely has write permissions even for fork PRs.
- AGENTS.md guidance documents this pattern to prevent recurrence.
Fixes the failure seen in
https://github.com/Comfy-Org/ComfyUI_frontend/actions/runs/22684230751/job/65763595989?pr=9380
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9382-fix-split-perf-report-workflow-for-fork-PR-support-3196d73d365081b29b35ed354e7789e2)
by [Unito](https://www.unito.io)
## Summary
Fix deterministic DOM widget clip-path rendering to resolve the flaky
"Can drag node" screenshot test.
## Root Cause
`useDomClipping.updateClipPath()` schedules clip-path calculation in a
`requestAnimationFrame`, but `DomWidget.vue`'s watcher reads
`clippingStyle.value` synchronously before the RAF fires. The stale
clip-path gets baked into `style.value` and never updated when the RAF
completes, causing the textarea DOM widget to non-deterministically
render in front of or behind the canvas-drawn node selection border.
## Fix
- Extract `composeStyle()` function and add a dedicated watcher on
`clippingStyle` that recomposes the final inline style whenever the
RAF-deferred clip-path updates
- Add `enableDomClipping` to the main watcher dependency array so
toggling the clipping setting immediately recomposes the style
- Add `moveMouseToEmptyArea()` call in the test as a secondary
stabilizer against hover highlight non-determinism
- Delete stale snapshot so CI regenerates it with correct clip-path
behavior
- Fixes#4658
---------
Co-authored-by: github-actions <github-actions@github.com>
## Summary
Adds a collapse/expand all toggle button to all parameter and error tabs
in the right side panel, letting users quickly collapse or expand all
accordion sections at once.
## Changes
- **New component**: `CollapseToggleButton.vue` — a reusable icon button
(list-collapse / list-tree icon) with tooltip, bound via `v-model`
- **Error tab**: Toggle collapses/expands all error groups; per-group
state is now managed through `isSectionCollapsed` /
`setSectionCollapsed` helpers
- **Nodes tab** (`TabNodes.vue`): Per-node `collapseMap`; toggle
overrides per-section state; defaults to collapsed (nodes tab default
was already collapsed)
- **Normal inputs tab** (`TabNormalInputs.vue`): Per-node `collapseMap`
+ `advancedCollapsed`; toggle covers both normal and advanced sections;
defaults to collapsed when multiple nodes selected
- **Subgraph inputs tab** (`TabSubgraphInputs.vue`): Toggle covers both
main and advanced inputs sections
- **Global parameters tab** (`TabGlobalParameters.vue`): Toggle bound to
single `SectionWidgets`
- **i18n**: Added `g.collapseAll` and `g.expandAll` keys
## Review Focus
- `isAllCollapsed` getter in each tab: reads from the same per-section
state so the toggle accurately reflects current state rather than being
independently tracked
- `TabNormalInputs`: multi-node selection default collapse behaviour is
preserved through `isSectionCollapsed` fallback logic
## Screenshots
<img width="778" height="643" alt="image"
src="https://github.com/user-attachments/assets/04d07f32-5135-47f9-b029-78ca78a996fb"
/>
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9333-feat-add-collapse-expand-all-toggle-to-right-panel-tabs-3176d73d36508123ba22d6e81983bb1b)
by [Unito](https://www.unito.io)
## Summary
Add a reusable `SearchInput` component with theme-aware styling, clear
button, loading state, and configurable sizes.
## Changes
- **SearchInput.vue**: Composable search input wrapping Reka UI Combobox
with search/clear/loading icon states
- **searchInput.variants.ts**: CVA-based size variants (`sm`, `md`,
`lg`) using semantic theme tokens (`bg-secondary-background`,
`text-base-foreground`)
- **SearchInput.stories.ts**: Storybook coverage for all sizes, loading,
custom icon/placeholder, and background override
## Review Focus
- Clear button alignment with search icon (`left-3.5` for `icon-sm`
button vs `left-4` for `size-4` icon)
- Theme token choices for light/dark compatibility
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9168-feat-Add-reusable-SearchInput-component-3116d73d365081309290fe84a46852e4)
by [Unito](https://www.unito.io)
---------
Co-authored-by: Alexander Brown <drjkl@comfy.org>
## Summary
- Add `quickRegister()` mappings for models being added in open cloud
PRs so the "Use" button works when those node packs ship
## Mappings added
| Cloud PR | Node Pack | Model Directories | Loader Node |
|----------|-----------|-------------------|-------------|
| [#2645](https://github.com/Comfy-Org/cloud/pull/2645) |
ComfyUI-HunyuanVideoWrapper | `LLM/llava-llama-3-8b-*` |
`DownloadAndLoadHyVideoTextEncoder` |
| [#2598](https://github.com/Comfy-Org/cloud/pull/2598) |
comfyui-cogvideoxwrapper | `CogVideo/GGUF`, `CogVideo/ControlNet` |
`DownloadAndLoadCogVideoGGUFModel`, `DownloadAndLoadCogVideoControlNet`
|
| [#2594](https://github.com/Comfy-Org/cloud/pull/2594) |
ComfyUI-DynamiCrafterWrapper | `checkpoints/dynamicrafter{,/controlnet}`
| `DownloadAndLoadDynamiCrafterModel`,
`DownloadAndLoadDynamiCrafterCNModel` |
| [#2537](https://github.com/Comfy-Org/cloud/pull/2537) |
ComfyUI_LayerStyle_Advance | `BEN`, `BiRefNet/pth`, `onnx/human-parts`,
`lama` | `LS_LoadBenModel`, `LS_LoadBiRefNetModel`,
`LS_HumanPartsUltra`, `LaMa` |
## Safe to merge before backend
Unknown node classes are silently skipped by `registerNodeProvider()` —
the mapping becomes a no-op until the node pack is deployed. Zero risk.
## Test plan
- [ ] Verify no runtime errors on load (unknown classes are skipped
gracefully)
- [ ] After backend PRs merge, verify "Use" button creates correct node
for each model type
🤖 Generated with [Claude Code](https://claude.com/claude-code)
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9411-feat-add-model-to-node-backlinks-for-upcoming-custom-nodes-31a6d73d3650811bb129e557450f263a)
by [Unito](https://www.unito.io)
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
## Summary
Follow-up to #9380. Replaces local `clone()` with shared util and
centralizes output snapshotting.
## Changes
- **What**: Replaced local `JSON.parse(JSON.stringify)` clone in
`changeTracker.ts` with shared `clone()` from `@/scripts/utils` (prefers
`structuredClone` with JSON fallback). Added `snapshotOutputs()` to
`useNodeOutputStore` as symmetric counterpart to existing
`restoreOutputs()`, and wired `changeTracker.store()` to use it.
- **Breaking**: None
## Review Focus
Symmetry between `snapshotOutputs()` and `restoreOutputs()` in the node
output store.
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9387-refactor-changeTracker-use-shared-clone-util-and-centralize-nodeOutputs-snapshot-3196d73d365081a289c3cb414f57929e)
by [Unito](https://www.unito.io)
## Summary
Spin out workflow tab/load stability fixes from the share-by-url branch
so they can merge independently and reduce regression risk.
## Changes
- **What**: Fixes duplicate tabs on repeated same-workflow loads by
making active-workflow reload idempotent in `afterLoadNewGraph`; fixes
tab flicker on save/rename by removing async detach/attach gaps in
`workflowStore`; hardens duplicate workflow path by loading before clone
and assigning a new workflow `id`.
## Review Focus
Please review the idempotency gate in `afterLoadNewGraph`
(`activeState.id === workflowData.id`) and the save/rename path update
sequencing in `workflowStore` to confirm behavior remains correct for
restoration and re-import flows.
## Screenshots (if applicable)
N/A (workflow logic and tests only)
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9345-fix-spin-out-workflow-tab-load-stability-regressions-3186d73d365081fe922bdc61dcf8d8f8)
by [Unito](https://www.unito.io)
---------
Co-authored-by: Amp <amp@ampcode.com>
## Summary
Add missing `v-if="!compact"` guards to painter label divs so all labels
are hidden consistently in compact mode.
## Changes
- **What**: Added `v-if="!compact"` to the Color, Hardness, Width,
Height, and Background label divs in `WidgetPainter.vue`, matching the
existing guards on Tool and Size labels.
## Review Focus
Straightforward consistency fix — all 7 label divs now use the same
compact-mode guard.
Fixes#9235
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9397-fix-hide-all-painter-labels-in-compact-mode-consistently-31a6d73d3650811a9d3af8dd290e2bca)
by [Unito](https://www.unito.io)
## Summary
Adds model-to-node backlinks in `modelToNodeStore.ts` for all
cloud-deployed custom node models that were missing mappings. Without
these, clicking "Use" on a model in the model browser throws an error.
**17 new backlinks added** covering ~340 models across deployed node
packs:
| Category | Directories | Node | Models |
|----------|-------------|------|--------|
| Vision-Language | LLM/Qwen-VL/* (12 specific paths) | AILab_QwenVL /
AILab_QwenVL_PromptEnhancer | ~186 |
| TTS | qwen-tts/* | FB_Qwen3TTSVoiceClone | ~68 |
| Video | SEEDVR2, liveportrait/*, mimicmotion, rife | various | ~33 |
| Depth | depthanything3 | DownloadAndLoadDepthAnythingV3Model | 7 |
| Segmentation | face_parsing, sam3 | various | 4 |
| Diffusers | diffusers/* (Kolors) | DownloadAndLoadKolorsModel | 16 |
| Other | clip/*, dwpose, onnx, detection, UltraShape, sharp | various |
~26 |
**Key fix:** Replaced the top-level `LLM` fallback with specific
`LLM/Qwen-VL/*` paths. The old fallback incorrectly mapped `LLM/llava-*`
models to `AILab_QwenVL`.
Models without deployed node packs (llava/HyVideo, latentsync, sam3d,
sam3dbody, inpaint, vae_approx) are excluded — those are being removed
from `supported_models.json` in Comfy-Org/cloud#2652.
## Test plan
- [ ] Verify "Use" button works for QwenVL models in model browser
- [ ] Verify "Use" button works for TTS, video, depth, segmentation
models
- [ ] Verify no `No node provider registered for category` errors for
deployed models
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: GitHub Action <action@github.com>
## Summary
Replace fixed 10%/20% perf delta thresholds with dynamic σ-based
classification using z-scores, eliminating false alarms from naturally
noisy duration metrics (10-17% CV).
## Changes
- **What**:
- Run each perf test 3× (`--repeat-each=3`) and report the mean,
reducing single-run noise
- Download last 5 successful main branch perf artifacts to compute
historical μ/σ per metric
- Replace fixed threshold flags with z-score significance: `⚠️
regression` (z>2), `✅ neutral/improvement`, `🔇 noisy` (CV>50%)
- Add collapsible historical variance table (μ, σ, CV) to PR comment
- Graceful cold start: falls back to simple delta table until ≥2
historical runs exist
- New `scripts/perf-stats.ts` module with `computeStats`, `zScore`,
`classifyChange`
- 18 unit tests for stats functions
- **CI time impact**: ~3 min → ~5-6 min (repeat-each adds ~2 min,
historical download <10s)
## Review Focus
- The `gh api` call in the new "Download historical perf baselines"
step: it queries the last 5 successful push runs on the base branch. The
`gh` CLI is available natively on `ubuntu-latest` runners and
auto-authenticates with `GITHUB_TOKEN`.
- `getHistoricalStats` averages per-run measurements before computing
cross-run σ — this is intentional since historical artifacts may also
contain repeated measurements after this change lands.
- The `noisy` classification (CV>50%) suppresses metrics like `layouts`
that hover near 0 and have meaningless percentage swings.
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9305-feat-add-statistical-significance-to-perf-report-with-z-score-thresholds-3156d73d3650818d9360eeafd9ae7dc1)
by [Unito](https://www.unito.io)
## Summary
Add `eslint-plugin-better-tailwindcss` to the ESLint toolchain for
Tailwind CSS v4 class linting.
## Changes
- **What**: Integrate `eslint-plugin-better-tailwindcss` (v4.3.1) with
the recommended config, pointed at the design-system CSS entry point for
v4 theme resolution. Five rules are enabled initially:
`enforce-canonical-classes`, `no-deprecated-classes`,
`no-conflicting-classes`, `no-duplicate-classes`,
`no-unnecessary-whitespace`. Three rules are disabled pending follow-up:
`no-unknown-classes` (needs PrimeIcon/custom class whitelisting),
`enforce-consistent-line-wrapping` (oxfmt conflict risk),
`enforce-consistent-class-order` (large batch change).
- **Dependencies**: `eslint-plugin-better-tailwindcss` ^4.3.1
- Fix conflicting `outline outline-1` classes in
`FormDropdownMenuActions.vue` (caught by the new
`no-conflicting-classes` rule).
## Review Focus
- Is the rule severity/enablement strategy appropriate for incremental
adoption?
- The 700 warnings (mostly `enforce-canonical-classes` and
`no-deprecated-classes`) are all auto-fixable via `eslint --fix` —
should we batch-fix them in this PR or a follow-up?
Fixes COM-15518
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9245-feat-add-eslint-plugin-better-tailwindcss-for-Tailwind-v4-linting-3136d73d365081df8a64dd55962d073f)
by [Unito](https://www.unito.io)
---------
Co-authored-by: GitHub Action <action@github.com>
Co-authored-by: Alexander Brown <drjkl@comfy.org>
## Summary
Replace hardcoded color and spacing values with semantic design tokens
and cache a computed variant class in StatusBadge.
## Changes
- **What**: Use Tailwind 4 CSS spacing variables in FormDropdownMenu
layout configs, replace zinc color utilities with semantic
`node-component-border` tokens in FormDropdownInput, wrap
`statusBadgeVariants()` in a `computed` for caching in StatusBadge.
## Review Focus
Straightforward token replacements and a computed caching change -- no
behavioral differences expected.
Fixes#9087Fixes#9086Fixes#7910
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9349-fix-replace-hardcoded-styles-with-design-tokens-and-cache-StatusBadge-variants-3186d73d36508185aae2e0753c9d1694)
by [Unito](https://www.unito.io)
Summary
- Add hidden setting `Comfy.Queue.ShowRunProgressBar` (default `true`).
- Add `Show run progress bar` toggle to the shared `...` job history
menu (`JobHistoryActionsMenu`), placed next to `Docked Job History`.
- Use that setting to control both the inline run progress bar and the
inline summary text under it.
- Keep queue button right-click context menu focused on queue actions.
- Add/update tests for the new toggle behavior and summary visibility.
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9176-fix-add-run-progress-toggle-to-job-history-menu-3116d73d365081118202d8d67a857367)
by [Unito](https://www.unito.io)
## Summary
Cache `canvas.style.cursor` to avoid redundant DOM writes that dirty
Firefox's style tree.
## Changes
- **What**: Add `_lastCursor` field to
`LGraphCanvas._updateCursorStyle()` — only writes `canvas.style.cursor`
when the value changes. Eliminates ~347 redundant style mutations per
profiling session.
## Review Focus
- The fix is 2 lines (cache field + comparison). The unit test validates
the caching pattern without requiring full LGraphCanvas instantiation.
- This is one of several contributors to Firefox's cascading style
recalculation freeze. Each `canvas.style.cursor` write dirties the style
tree, which is flushed during the next paint in the canvas render loop.
## Stack
2 of 4 in Firefox perf fix stack. Depends on #9170.
<!-- Fixes #ISSUE_NUMBER -->
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9171-fix-cache-canvas-cursor-style-to-avoid-redundant-DOM-writes-3116d73d36508139827fe1d644fa1bd0)
by [Unito](https://www.unito.io)
## Summary
Replace `eval()` in `evaluateInput()` with a custom recursive descent
math parser, eliminating a security concern and enabling the `no-eval`
lint rule.
## Changes
- **New**: `mathParser.ts` — recursive descent parser for `+`, `-`, `*`,
`/`, `%`, `()`, decimals, unary operators. Zero new dependencies.
- **Modified**: `widget.ts` — replaced `eval()` call with
`evaluateMathExpression()`, use `isFinite()` instead of `isNaN()` to
reject `Infinity`
- **Modified**: `.oxlintrc.json` — `no-eval` rule changed from `"off"`
to `"error"`
- **Tests**: 59 parser tests + 23 integration tests covering complex
expressions, edge cases, and invalid input
## Review Feedback Addressed
- Renamed `unit()` → `primary()` for clarity
- Added modulo (`%`) operator support
- Normalized negative zero to positive zero
- Added depth limit (200) for nested parentheses
- Used `isFinite()` instead of `isNaN()` to reject
`Infinity`/`-Infinity`
- Added tests for edge-case number formats, unary-after-binary
operators, modulo, depth limits, scientific/hex notation, and `Infinity`
Fixes#8032Fixes#9272Fixes#9273Fixes#9274Fixes#9275
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9263-fix-Replace-eval-with-safe-math-expression-parser-3136d73d3650812f9f8dea21d1ea4f06)
by [Unito](https://www.unito.io)
## Summary
Narrow `CreateNodeOptions` from `Partial<Omit<LGraphNode, ...>>`
(exposing hundreds of properties/methods) to an explicit interface
listing only creation-time properties.
## Changes
- Replace `Partial<Omit<LGraphNode, 'constructor' | 'inputs' |
'outputs'>>` with explicit `CreateNodeOptions` interface containing
only: `pos`, `size`, `properties`, `flags`, `mode`, `color`, `bgcolor`,
`boxcolor`, `title`, `shape`, `inputs`, `outputs`
- Rename local `CreateNodeOptions` in `createModelNodeFromAsset.ts` to
`ModelNodeCreateOptions` to avoid collision
## Ecosystem verification
GitHub code search across ~50 repos confirms only `pos` and `outputs`
are used externally. All covered by the narrowed interface.
Fixes#9276Fixes#4740
## Summary
Sort execution error cards within each error group by their node
execution ID in ascending numeric order, ensuring consistent and
predictable display order.
## Changes
- **What**: Added `compareExecutionId` utility to
`src/types/nodeIdentification.ts` that splits node IDs on `:` and
compares segments numerically left-to-right; applied it as a sort
comparator when building `ErrorGroup.cards` in `useErrorGroups.ts`
## Review Focus
- The comparison treats missing segments as `0`, so `"1"` sorts before
`"1:20"` (subgraph nodes follow their parent); confirm this ordering
matches user expectations
- All comparisons are purely numeric — non-numeric segment values would
sort as `NaN` (treated as `0`)
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-9334-feat-error-groups-sort-execution-error-cards-by-node-execution-ID-3176d73d365081e1b3e4e4fa8831fe16)
by [Unito](https://www.unito.io)
