# cad-ui ignored-test inventory (UI-00)

**Date:** 2026-09-14
**Owner:** UI-00
**CI budget:** `CAD_IGNORED_BUDGET: 20` in `.forgejo/workflows/cad.yml`
**Scope:** `crates/apps/cad/cad-core/src` + `crates/apps/cad/cad-ui/src`,
counted by `grep -rn '#\[ignore' <both trees> | wc -l`.

## Counting rule (read this before "fixing" the budget)

The gate counts **grep hits for the literal string `#[ignore`**, not
`#[ignore]` attributes:

- 18 × `#[ignore = "benchmark: ..."]` attributes in
  `cad-ui/src/profile_benchmarks.rs`
- 1 × `#[ignore = "slow: ..."]` attribute in
  `cad-ui/src/script_bindings.rs`
- 1 × doc-comment mention of the literal `#[ignore]` in
  `cad-ui/src/script_bindings.rs:881`
  (`/// Marked \`#[ignore]\` because it deliberately burns the whole budget;`)

That is **20 grep hits = 19 true attributes + 1 doc mention**.
The budget 20 is correct for the grep-defined gate. Do not "correct"
it to 19 without also changing the gate predicate and this file.
`cad-core/src` currently contributes zero ignores.

## Policy

- Every true `#[ignore]` below needs owner, issue, reason, and expiry,
  one row per test. Owner is the inventorying tranche (UI-00 for all
  rows at baseline); Issue is the tranche owning final disposition —
  UI-15 for timing/infra-bound tests bound for nightly/device jobs,
  UI-00 for the baseline slow-suite entry.
- CI fails if the grep count drifts in either direction (silently adding
  ignores hides coverage; silently dropping the count means the budget
  is stale) and fails if any expiry date has passed (`No ignore expiry
  has passed` step in `cad.yml` scans this file for `YYYY-MM-DD`).
- Bumping `CAD_IGNORED_BUDGET` requires a tranche note with owner,
  reason, and expiry. Expiry extensions are reviewable edits here,
  not silent CI edits.
- Genuine infrastructure blockers belong in required nightly/device
  jobs with expiry (UI-15), not as permanent ignores.

## True ignores (19)

| # | File:line | Test fn | Owner | Issue | Reason | Expiry |
|---|-----------|---------|-------|-------|--------|--------|
| 1 | `profile_benchmarks.rs:68` | `bench_parallel_threshold_warm_vs_cold_cache` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 2 | `profile_benchmarks.rs:148` | `bench_geometry_buffers_shared_by_shape` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 3 | `profile_benchmarks.rs:250` | `bench_param_hash_cost_per_frame` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 4 | `profile_benchmarks.rs:327` | `bench_parts_script_regeneration_per_drag_frame` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 5 | `profile_benchmarks.rs:401` | `bench_pick_broadphase_world_aabb_recompute_vs_cache` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 6 | `profile_benchmarks.rs:507` | `bench_pick_broadphase_mesh_bounds_vs_size` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 7 | `profile_benchmarks.rs:586` | `bench_mesh_cache_hit_cost` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 8 | `profile_benchmarks.rs:620` | `bench_size_parametric_vs_mesh_derived` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 9 | `profile_benchmarks.rs:681` | `bench_scene_cache_hit_vs_rebuild` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 10 | `profile_benchmarks.rs:729` | `bench_glb_export_with_and_without_cache` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 11 | `profile_benchmarks.rs:795` | `bench_frame_submission_budget` | UI-00 | UI-15 | benchmark (counts, not wall-clock): run explicitly with `--ignored` | 2027-06-14 |
| 12 | `profile_benchmarks.rs:995` | `bench_shared_cache_export_reuse` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 13 | `profile_benchmarks.rs:1035` | `bench_scene_cache_scaling` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 14 | `profile_benchmarks.rs:1077` | `bench_command_execute_overhead` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 15 | `profile_benchmarks.rs:1145` | `bench_parallel_vs_sequential_export` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 16 | `profile_benchmarks.rs:1186` | `bench_gpu_upload_mesh_source` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 17 | `profile_benchmarks.rs:1296` | `bench_delete_invalidation_clear_vs_evict` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 18 | `profile_benchmarks.rs:1356` | `bench_viewport_snapshot_sync_per_frame` | UI-00 | UI-15 | benchmark: timing-dependent, run explicitly with `--ignored` | 2027-06-14 |
| 19 | `script_bindings.rs:884` | `a_runaway_script_is_terminated_by_the_budget` | UI-00 | UI-00 | slow: intentionally runs until the 5 s `CAD_SCRIPT_TIME_BUDGET` trips; would burn minutes in the default suite | 2027-06-14 |

Line numbers are advisory (they drift with edits); the CI gate matches
test-fn names, and the budget gate matches the grep count.

## Non-attribute hit (1, not a test)

- `script_bindings.rs:881` — doc comment `/// Marked \`#[ignore]\`
  because ...` explaining ignore #19 above. Counts toward the
  grep budget by construction. If this comment is reworded to avoid
  the literal, the budget must drop to 19 in the same tranche.

## Demo triage (fixed legitimately — regression guard stays)

`cad-ui/src/demo.rs::demo_has_slab_and_wall` **failed** at the UI-00
baseline and was kept red by policy until a legitimate fix landed
(UI-02 demo fix): `SceneBuilder::domain_box()` (cad-core
`cad_scene.rs`) set layer/material/name but never set `kind_hint`;
`CadNode::part_kind()` prefers `kind_hint` and falls back to
deriving from the solid — both wall and slab are `CadSolid::Box`,
so both classified as `PartKind::Cube` and `demo_counts()` returned
`(0, 0)`.

- Fix (not a weakening): `domain_box()` takes the domain `PartKind`
  and records it on the pending node, `commit_pending` carries it into
  `CadNode.kind_hint`, and `column()` does the same for `Column`;
  covered by `domain_builders_record_kind_hint` in `cad-core`.
  `demo.rs` itself is untouched — same test, same assertions.
- CI guard: `cad.yml` still asserts `demo.rs` contains the
  `demo_has_slab_and_wall` test and its `walls >= 1` / `slabs >= 1`
  assertions, so the regression cannot be hidden by deletion,
  inversion, or `#[ignore]`.