# Live results: the modelling ops on real SOLIDWORKS

Measured 2026-10-14 with bridge **0.10.3** on **SOLIDWORKS 2026** (major version 34, 3DEXPERIENCE edition, Windows 11).
These are the features requested in wiki issue #1 and the bugs reported in #2.

## Summary

- `tools/dsl-live-check.py`: **30 of 30 checks pass.** Each builds a part on SOLIDWORKS through `swx_run_script`, reads it back with `swx_inspect_bodies` / `swx_mass_properties`, and compares it with geometry computed independently in the script (closed-form volumes, bounding boxes from the documented conventions). Tolerances: volume 0.2 %, bounding box 0.01 mm.
- `swx_selftest {"level":"write"}`: **43 passed, 0 failed, 1 skipped** (the active-document reads, skipped because the test leaves no document open). 538 s.
- Selections after a geometric pick used the first rung of the ladder, `MultiSelect2` with a typed object array, every time (11 of 11).
- Every scratch part was closed afterwards: SOLIDWORKS was left with no open documents.

## The LEGO 2x4 brick (`examples/lego-2x4.json`)

Built in 4 features, measured live:

| Measured | Value | Issue #1/#2 reference (SOLIDWORKS 2021, 22-feature build) |
|---|---|---|
| Volume | 2073.362 mm^3 | 2073.362 mm^3 |
| Bounding box | [-15.9, 0.0, -7.9] to [15.9, 11.3, 7.9] mm | 31.8 x 11.3 x 15.8 mm |
| Faces | 39 | 39 |
| Colour | [200, 0, 0] (set before the material, kept after it) | red requested |
| Density / mass | 1020.0 kg/m^3 / 0.0021148 kg | ABS 1020 / 0.0021148 |

## CSWA benchmark, measured 2026-10-15 (bridge 0.11.2, SW 2026)

The first live run of the CSWA benchmark: `tools/cswa-check.py` on Drew's laptop (target `adom_drew2`),
2026-10-15 23:04:22Z, bridge **0.11.2**, **SOLIDWORKS 2026** (34.3.0, 3DEXPERIENCE "Connected"). The eight
runnable parts were built in ONE bridge-made part (`reuseDocument`, `clear_part` between them), weighed with
`swx_mass_properties`, and compared with their published keys. **8 of 8 within tolerance**, every one within
0.0015 % of its key; Step 4 was skipped (it needed `offsetEntities`, added in 0.11.4). Each part took 7-23 s.
SOLIDWORKS held 2.2 GB afterwards and quit cleanly. The full evidence (every reply) is
[`tools/cswa/reports/cswa-2026-10-15-sw2026-0.11.2.json`](../tools/cswa/reports/cswa-2026-10-15-sw2026-0.11.2.json).

| Spec | Key (published) | Measured live | Off | Time |
|---|---|---|---|---|
| `cswa-00-lego-2x4` (spec 0, ABS) | 2073.3615208 mm³ | 2073.3615208 mm³ (2.1148288 g) | 1e-9 % | 14.7 s |
| `cswa-01-3dengr-q5-block` (6061) | 324.00 g, CoM (50, 10, 30) | 324.000 g, CoM (50.000, 10.000, 30.000) | 0.000 % | 9.4 s |
| `cswa-02-3dengr-q6-hole` (6061) | 307.04 g, CoM (51.38, 10.00, 30.00) | 307.035 g, CoM (51.381, 10.000, 30.000) | 0.0015 % | 12.0 s |
| `cswa-03-3dengr-q8-hole30` (6061) | 285.83 g | 285.830 g | 0.0001 % | 11.8 s |
| `cswa-04-3dengr-q9-l-bracket` (AISI 1020) | 410.80 g, CoM (26.54, 16.54, 20.00) | 410.800 g, CoM (26.538, 16.538, 20.000) | 0.000 % | 7.3 s |
| `cswa-05-tool-block-step1` (official Q3) | 939.54 g | 939.540 g | 0.00005 % | 7.7 s |
| `cswa-06-tool-block-step2` (official Q4) | 1032.32 g | 1032.321 g | 0.00007 % | 7.5 s |
| `cswa-07-tool-block-step3` (official Q5) | 628.18 g | 628.183 g | 0.0005 % | 23.0 s |
| `cswa-08-tool-block-step4` (official Q6) | 432.58 g | SKIPPED: needed `offsetEntities` | | |

The densities read back were the ones asked for (1020, 2700 and 7900 kg/m³), and each part carried its library
material (ABS, 6061 Alloy, AISI 1020).

**What this measured.** Every variant those eight scripts used is now measured live on SOLIDWORKS 2026
(`compat::OP_FACTS`, measurement `LIVE_2026_10_15_CSWA`). That includes six variants no earlier run had built:

- `sketch.line` and `sketch.arc`: the Tool Block profile's lines, its tangent R19 and its three-point R29 and R5;
- `sketch.faceOffset`: the R41 slot's plane, 12 mm inside the back face;
- `hole.simple`: Ø20 and Ø30 through all, and the Tool Block's Ø11;
- `set_material.density`;
- `sketch.face.front.positive`: sketches on a +Z face.

It also re-measured `clear_part`, `cut.one`, `cut.one.reversed`, `extrude.one`, `select.multiSelectTyped`,
`set_material`, `sketch.circle`, `sketch.multiProfile`, `sketch.offset.top.positive`, `sketch.plane.front`,
`sketch.plane.top` and `sketch.rect` on a newer bridge.

**Still unmeasured** (no spec in this run used them): `revolve` and `revolve_cut`, `mirror`, `slot`, `polygon`,
the counterbored and countersunk `hole`, the `upToNext` / `upToSurface` / `offsetFromSurface` end conditions,
`draft`, and construction geometry. 0.11.4 adds community specs that build each of them
(`cswa-09` to `cswa-12`), and the Step 4 pocket (`offsetEntities`, sketch regions), so the next live run measures them.

## HideFileExt, measured 2026-10-15 (bridge 0.11.2, SOLIDWORKS 2026)

On Drew's laptop, with Windows' "Hide extensions for known file types" turned
ON (HideFileExt=1, SOLIDWORKS restarted). This is Michael's 2021 condition,
reproduced on our machine to prove the 0.10.10 fix:

| Check | Result |
|---|---|
| `swx_new_document` title read back | `Part1-1792103965000`, with no extension. `swx_docs` lists it the same way |
| `swx_run_script` by path | ok, `bridgeCreatedDocument:true` |
| `swx_design_info`, `swx_get_parameters` | PASS by the reported title, by the stem, and by the stem + `.SLDPRT` |
| `swx_set_parameter`, `swx_export_step` (15,944 B), `swx_save`, `swx_close_doc` | PASS by the stem |
| Left open | nothing |

`compat::OP_FACTS` carries it as `titles.hiddenExtensions`.

### The restart that followed (the 0.11.3 finding)

The second phase restarted SOLIDWORKS with extensions shown. From the ledger:

| Time (Z) | Event |
|---|---|
| 22:40:01 | quit (processGone) |
| 22:40:02 | launch |
| 22:40:07 | `sldworksUp` |
| 22:40:39 | attached |
| 22:40:43 | `solidworks.ready` TRUE, 4 s after attach (the first phase's start took 22 s to ready) |
| 22:40:50 | `swx_new_document` OK, title `Part1-1792104049001.SLDPRT`, ledger `document_created` |
| 22:40:51 | `swx_docs`: `[]`; SOLIDWORKS pid 26316 had DIED |
| 22:40:58 | `swx_stop`: `host_app_not_running` |

This was the 2026-10-01 crash again: a document command sent while an add-in
was still initialising.

0.11.3 makes these changes:
- A performance-log marker counts only if it lies beyond the state the file was
  in when no SOLIDWORKS was writing it (launch, quit, or a first sight of a
  stale log), unless the file has been rewritten since.
- A fresh attach is ready only with StartupProcessCompleted TRUE, this session's
  own marker, and 20 s after attach.
- `swx_new_document`, `swx_open` and `swx_import` verify the effect on the live
  session: same process, still running, the document in the list. If the effect
  is missing they return `host_app_died` / `document_not_created`, never ok.
- A start within 60 s of a quit waits for that session's process tree to be gone.

## SOLIDWORKS 2021 (community)

Run 2026-10-15 by Michael (issue #1) with bridge **0.10.9** on **SOLIDWORKS 2021**
(COM RevisionNumber 29.5.1, classic desktop, Windows 11, ab 2.1.168). His reply,
not our machine: it counts as a measurement `by:"community"` in `compat::OP_FACTS`.

- `swx_selftest {"level":"write"}`: **44 passed, 6 failed** (380 s). Every
  modelling check passed: T01-T22 and T24-T32 (T23, the ownership check, was an
  unnumbered row until 0.10.10).
- The 6 failures were ONE bug: `swx_design_info`, `swx_get_parameters`,
  `swx_set_parameter`, `swx_export_step`, `swx_save` and `swx_close_doc` aimed
  at the scratch part as `Part14-1792088642002.SLDPRT` while SOLIDWORKS titled it
  `Part14-1792088642002` — Windows' "Hide extensions for known file types"
  (HideFileExt=1, the Windows default) is on there and off on our laptop. The
  close missed too, so the part was left open; closing it by path worked.
  Fixed in 0.10.10: title matching ignores `.sldprt/.sldasm/.slddrw`, the
  self-test addresses its own documents by path, and `swx_new_document` reports
  the title SOLIDWORKS shows.
- Issue #2's fixes confirmed: `densitySource:"bodyMassProperty"`,
  `bridgeCreatedDocument:true`.

The LEGO 2x4 brick on 2021, identical to 2026:

| Measured | SOLIDWORKS 2021 (0.10.9) | SOLIDWORKS 2026 (0.10.3) |
|---|---|---|
| Features | 4 | 4 |
| Volume | 2073.362 mm^3 | 2073.362 mm^3 |
| Bounding box | [-15.9, 0.0, -7.9] to [15.9, 11.3, 7.9] mm | the same |
| Faces | 39 | 39 |
| Colour | [200, 0, 0] | [200, 0, 0] |
| Density / mass | 1020 kg/m^3 / 0.0021148 kg | 1020 kg/m^3 / 0.0021148 kg |

## Every check

| Id | Check | Issue | Expected volume (mm^3) | Measured (mm^3) | Result |
|---|---|---|---|---|---|
| T01 | rectangle anchor: xMm/yMm is the CORNER by default | #1.5 | 1000.0 | 1000.0 | PASS |
| T02 | Top Plane: sketch +Y is model -Z | #1.5 | 1000.0 | 1000.0 | PASS |
| T03 | anchor center | #1.5 | 1000.0 | 1000.0 | PASS |
| T04 | sketch on a plane offset +10 mm from Top Plane | #1.1 | 4392.699 | 4392.699 | PASS |
| T05 | sketch on a NEGATIVE offset plane (-5 mm) | #1.1 | 6000.0 | 6000.0 | PASS |
| T06 | sketch on a FACE picked by a point (the block's top face) | #1.1 | 4113.097 | 4113.097 | PASS |
| T07 | Michael's case: a default cut from Top Plane with the body on +Y must FAIL LOUDLY, never succeed silently | #1.2 |  |  | PASS |
| T08 | cut reverse:true removes material on +Y | #1.2 | 3500.0 | 3500.0 | PASS |
| T09 | cut throughAll + reverse | #1.2 | 3000.0 | 3000.0 | PASS |
| T10 | extrude direction both (5 mm one way, 3 mm the other) | #1.2 | 800.0 | 800.0 | PASS |
| T11 | extrude midPlane | #1.2 | 1000.0 | 1000.0 | PASS |
| T12 | cut midPlane through the block | #1.2 | 3904.0 | 3904.0 | PASS |
| T13 | two concentric circles in ONE sketch extrude as a RING | #1.3 | 502.655 | 502.655 | PASS |
| T14 | two disjoint rectangles in ONE sketch: one feature, two bodies | #1.3 | 250.0 | 250.0 | PASS |
| T15 | set_appearance on the part | #1.4 | 4000.0 | 4000.0 | PASS |
| T16 | set_appearance on one face picked by a point | #1.4 | 4000.0 | 4000.0 | PASS |
| T17 | fillet every edge of the block, r = 1 | #1.6 | 3958.419 | 3958.419 | PASS |
| T18 | chamfer ONE edge picked by a point, 1 mm | #1.6 | 3990.0 | 3990.0 | PASS |
| T19 | shell 1 mm, top face removed | #1.6 | 1084 | 1084.0 | PASS |
| T20 | linear pattern: one boss -> 3 along X, 6 mm apart | #1.6 | 4188.496 | 4188.496 | PASS |
| T21 | circular pattern: one boss -> 6 around Y | #1.6 | 4169.646 | 4169.646 | PASS |
| T22 | issue #2: body mass uses the material's density (ABS 1020 kg/m^3) | #2.a | 4000.0 | 4000.0 | PASS |
| T25 | audit #1/#5: sketch on the BOTTOM face (hidden in an iso view, normal -Y) extrudes OUTWARD | #1.1 | 4084.823 | 4084.823 | PASS |
| T26 | audit unverified #1: a NEGATIVE offset plane keeps the base plane's axes (asymmetric sketch, separate body) | #1.1 | 4048.0 | 4048.0 | PASS |
| T27 | audit #6: an ASYMMETRIC two-way cut: depthMm goes along the cut's own direction (-normal), depth2Mm the other way | #1.2 | 968 | 968.0 | PASS |
| T28 | audit unverified #2: linear pattern with NEGATIVE spacing goes -X (instances stay on the block) | #1.6 | 4188.496 | 4188.496 | PASS |
| T29 | audit #1: shell removing the BOTTOM face (hidden in an iso view) removes that face, not a side | #1.6 | 1084 | 1084.0 | PASS |
| T30 | audit #3: fillet ONE feature's edges by allOfFeature on a cylinder (2 circular edges) | #1.6 | 772.517 | 772.517 | PASS |
| T24 | the LEGO 2x4 acceptance brick (examples/lego-2x4.json) | #1 acceptance | 2073.362 | 2073.362 | PASS |
| T23 | issue #2: run_script on the bridge's own new part (addressed by path) says bridgeCreatedDocument:true | #2.b |  |  | PASS |

## Added 2026-10-14 (bridge 0.10.4)

Eight more checks were added to `tools/dsl-live-check.py` and run on the same
install (SOLIDWORKS 2026, major 34, 3DEXPERIENCE "Connected") with bridge
**0.10.4**. All eight pass. T36 and T37 first failed on the checks' own
expectations, not on the bridge. A face sketch keeps the default plane's axes
and projects the origin onto the face, so their circles were moved to sketch
(0, 5) and both checks re-run. Since 0.10.5 these variants report
`proven.live: true` with this measurement in `measuredOn`.

| Id | Check | Issue | Expected volume (mm^3) | Measured (mm^3) | Measured bounding box (mm) | Result |
|---|---|---|---|---|---|---|
| T33 | a REVERSED boss: reverse:true from Top Plane goes -Y | #1.2 | 4000.0 | 4000.0 | [-10, -10, -10] to [10, 0, 10] | PASS |
| T34 | a boss through all (one way) from a plane below the block | #1.2 | 4062.832 | 4062.832 | [-10, -5, -10] to [10, 10, 10] | PASS |
| T35 | Right Plane axes: sketch +X is model -Z, +Y is model +Y, a boss goes +X | #1.5 | 1000.0 | 1000.0 | [0, 0, -10] to [5, 20, 0] | PASS |
| T36 | a sketch on the +X side face extrudes outward along +X (Right Plane's axes, origin projected) | #1.1 | 4037.699 | 4037.699 | [-10, 0, -10] to [13, 10, 10] | PASS |
| T37 | a sketch on the -Z face extrudes outward along -Z (Front Plane's axes, origin projected) | #1.1 | 4037.699 | 4037.699 | [-10, 0, -13] to [10, 10, 10] | PASS |
| T38 | a plane offset +12 mm from Front Plane (a separate body at z 12..14) | #1.1 | 4032.0 | 4032.0 | two bodies, the second [-2, -2, 12] to [2, 2, 14] | PASS |
| T39 | a linear pattern in two directions (X +12, Z -12): 4 bosses on the block | #1.6 | 4084.823 | 4084.823 | [-10, 0, -10] to [10, 13, 10] | PASS |
| T42 | a shell outward 1 mm with the top face removed: the block becomes the cavity | #1.6 | 1324.0 | 1324.0 | [-11, -1, -11] to [11, 10, 11] | PASS |

## Memory, measured 2026-10-14 (bridges 0.10.5 and 0.10.6)

| What | Measured |
|---|---|
| A new part per build (12 builds, each closed) | SOLIDWORKS private bytes +83 to +85 MB per part (520 to 2035 MB; again 538 to 1697 MB over 8); handles +~125 per part; `GetDocumentCount` 0 after every close |
| ONE reused part (`reuseDocument`, 12 builds) | 3260 to 3246 MB, handles 4373 to 4398: flat |
| A part created hidden (`visible:false`), then modelled | refused by SOLIDWORKS: a hidden document cannot be activated, and modelling selects |
| `swx_com_respawn` | about 210 MB freed (1697 to 1487 MB), handles unchanged (4022 to 4056): the bridge does not pin documents |
| Creating the reuse part (once) | +1.95 GB while open (1487 to 3437 MB), +285 MB after its close — observed, unexplained |
| `swx_selftest {"level":"write"}`, bridge 0.10.7 (one reused part) | complete, 475 s: 40 passed, 3 failed, 1 skipped; 15 documents created (more than 32 before); SOLIDWORKS 507 to 2361 MB (peak 3922 MB with parts open), handles 2809 to 5122; commit stayed "ok" (61.7 % to 70.0 %); no throttling, nothing recycled; launch attributed by `ancestry` |
| The self-test's closing `swx_com_respawn` | freed 0 MB (2361 to 2361 MB) — with the 210 MB above: not a saving to rely on |
| The 3 failures of that run | one bug: a single `DeleteSelection2` in `clear_part` returned true and left a top-level sketch (Sketch13, Sketch4, Sketch9) that an earlier check left unconsumed. 0.10.8 repeats the delete until the part is empty, and falls back to a fresh part for that check if it still cannot |

`clear_part` (the reuse step) is marked measured on 2026 in the variants table.

## CSWA canary 0.11.4-rc1, measured 2026-10-16 (SW 2026): 12 of 14, two bugs caught

The first run of the live gate on a release candidate: `tools/release.sh --canary` published 0.11.4-rc1
(exe sha256 35f04d75…), `bridge_install` put it on Drew's laptop from `manifests/canary.json`, and
`tools/cswa-check.py` built all 14 parts in one reused document, 22:36 to 22:55Z
(`tools/cswa/reports/cswa-2026-10-16-sw2026-0.11.4-rc1-FAILED.json`). Three earlier attempts that day were
refused by the bridge's presence rule when Drew returned to the laptop; each quit SOLIDWORKS cleanly.

| Part | Result | Key (g) | Measured (g) | Off % | Time |
|---|---|---|---|---|---|
| `cswa-00-lego-2x4` | PASS | 2.114829 | 2.115 | +0.000 | 44 s |
| `cswa-01-3dengr-q5-block` | PASS | 324.0 | 324.0 | +0.000 | 28 s |
| `cswa-02-3dengr-q6-hole` | PASS | 307.04 | 307.035 | +0.001 | 154 s |
| `cswa-03-3dengr-q8-hole30` | PASS | 285.83 | 285.83 | +0.000 | 35 s |
| `cswa-04-3dengr-q9-l-bracket` | PASS | 410.8 | 410.8 | +0.000 | 22 s |
| `cswa-05-tool-block-step1` | PASS | 939.54 | 939.54 | +0.000 | 21 s |
| `cswa-06-tool-block-step2` | PASS | 1032.32 | 1032.321 | +0.000 | 23 s |
| `cswa-07-tool-block-step3` | PASS | 628.18 | 628.183 | +0.001 | 65 s |
| `cswa-08-tool-block-step4` | PASS | 432.58 | 434.715 | +0.494 | 200 s |
| `cswa-08b-tool-block-step4-explicit` | PASS | 432.58 | 434.715 | +0.494 | 87 s |
| `cswa-09-community-stepped-shaft` | PASS | 103.568885 | 103.569 | +0.000 | 47 s |
| `cswa-10-community-mirrored-block` | PASS | 722.912919 | 722.913 | +0.000 | 57 s |
| `cswa-11-community-slotted-plate` | FAIL | 276.35927 | 274.188 | +0.786 | 87 s |
| `cswa-12-community-up-to-plates` | FAIL | 369.687533 | 364.35 | +1.444 | 220 s |

**What the two failures measured.** The deficits locate the error, because every mass and centre of mass
is exact:

- `cswa-11`: 804.248 mm3 too little material, centred at x 45, z 25, y 9.33 - the countersunk hole. That is
  exactly the outward frustum of the 45 deg cone (r 8 to 12 over 4 mm, 1273.4 mm3) minus the inward one
  (r 8 to 4, 469.1 mm3): the countersink's drafted cut opened downward. The counterbore beside it, the
  slots and the hexagon were right.
- `cswa-12`: 1976.9 mm3 short, which no single feature explains; the four-moment fit (volume and three
  first moments) picks one combination to 1 %: the up-to-surface pillar stopped at the next face
  (3392.9 mm3 missing), the up-to-surface hole cut only the first plate (1005.3 mm3 present), the 5 deg boss
  widened (+792 mm3) and the 10 deg pocket widened (-380.9 mm3). The up-to-next pillar and the
  offset-from-surface pillar were right.

So two API facts, both now in the code: `Ddir1` TRUE drafts OUTWARD on SOLIDWORKS 2026 (the help says
inward), and "up to surface" is `swEndCondUpToSurface` 4, not `swEndCondUpToSelection` 10, which left the
direction box empty and behaved as up-to-next for a boss and a cut alike. 0.11.4-rc2 carries the fixes and a
read-back of the stored end condition, depth and draft side on every extrude and cut.

The ten variants the passing parts exercised are measured on this run (`LIVE_2026_10_16_RC1` in
`src/compat.rs`): `revolve.360`, `revolve_cut.360`, `revolve_cut.midPlane`, `mirror.features.plane`,
`mirror.features.offset`, `mirror.bodies.plane`, `sketch.construction`, `sketch.offsetEntities.outer`,
`cut.regions`, `cut.offsetFromSurface`. The five the failing parts exercised (`extrude.draft`, `cut.draft`,
`hole.countersink`, `extrude.upToSurface`, `cut.upToSurface`) stay unmeasured and say so.

## CSWA canary 0.11.4-rc2, measured 2026-10-20 (SW 2026): 14 of 14

The fixes from the rc1 run, proven on the bytes that ship: 0.11.4-rc2 (exe sha256 1f812642…) on Drew's
laptop, 13:13 to 13:20Z, every part in one reused document
(`tools/cswa/reports/cswa-2026-10-20-sw2026-0.11.4.json`). `cswa-11` and `cswa-12`, the two that located the
draft flag and the up-to-surface id four days earlier, now land on their keys; nothing else moved.

| Part | Result | Key (g) | Measured (g) | Off % | Time |
|---|---|---|---|---|---|
| `cswa-00-lego-2x4` | PASS | 2.114829 | 2.115 | +0.0000 | 20 s |
| `cswa-01-3dengr-q5-block` | PASS | 324.0 | 324.0 | +0.0000 | 12 s |
| `cswa-02-3dengr-q6-hole` | PASS | 307.04 | 307.035 | +0.0015 | 23 s |
| `cswa-03-3dengr-q8-hole30` | PASS | 285.83 | 285.83 | +0.0001 | 21 s |
| `cswa-04-3dengr-q9-l-bracket` | PASS | 410.8 | 410.8 | +0.0000 | 11 s |
| `cswa-05-tool-block-step1` | PASS | 939.54 | 939.54 | +0.0000 | 18 s |
| `cswa-06-tool-block-step2` | PASS | 1032.32 | 1032.321 | +0.0001 | 15 s |
| `cswa-07-tool-block-step3` | PASS | 628.18 | 628.183 | +0.0005 | 35 s |
| `cswa-08-tool-block-step4` | PASS | 432.58 | 434.715 | +0.4936 | 49 s |
| `cswa-08b-tool-block-step4-explicit` | PASS | 432.58 | 434.715 | +0.4936 | 37 s |
| `cswa-09-community-stepped-shaft` | PASS | 103.568885 | 103.569 | +0.0000 | 16 s |
| `cswa-10-community-mirrored-block` | PASS | 722.912919 | 722.913 | +0.0000 | 23 s |
| `cswa-11-community-slotted-plate` | PASS | 276.35927 | 276.359 | +0.0000 | 36 s |
| `cswa-12-community-up-to-plates` | PASS | 369.687533 | 369.688 | +0.0000 | 54 s |

With this run the five variants rc1 caught are measured as well (`extrude.draft`, `cut.draft`,
`hole.countersink`, `extrude.upToSurface`, `cut.upToSurface`), together with `extrude.upToNext`,
`extrude.offsetFromSurface`, `hole.counterbore` and `cut.one.throughAll`, which only the failing parts had
exercised on rc1. The table in `src/compat.rs` records them on the release after 0.11.4 (the gate forbids a
code change between the canary and its publish), so `swx_describe` on 0.11.4 still lists them as not measured.

## Not yet measured

- SOLIDWORKS versions other than 2026 and 2021 (2021: a community run, 2026-10-15): supported by the API audit in `docs/COMPATIBILITY.md`, not yet run. `swx_selftest {"level":"write"}` on that install answers it.
- `throughAll` with `direction:"both"` on a boss, the `IEntity::Select4`, `SelectByRay` and `SelectByID2` selection rungs (never needed live: the first rung worked), and the `CreateDefinition` fillet fallback (never needed: `FeatureFillet3` worked).
- Sketches on a plane offset from Right Plane, a negative offset from Front Plane, or an offset from a named plane; sketches on a named plane; sketches on faces facing -X. (+Z faces: measured by the CSWA run, 2026-10-15.)
- A chamfer on all of a feature's edges, a colour on one body of a multi-body part, and `new_assembly`, `save_as`, `rebuild` and `zoom_fit` as script steps.
- Of the 0.11.0 exam ops, after the canary of 2026-10-16, still unmeasured: the counterbored and countersunk `hole`, `upToNext`, `upToSurface` and draft (the last two fixed in 0.11.4-rc2), `offsetFromSurface` on a boss, revolve by angle, mirror about faces and face-offset planes, and a plain through-all cut; of 0.11.4's, `offsetEntities` of a whole face and a boss of sketch regions (`regionsAtMm`). The specs `cswa-08`, `cswa-08b`, `cswa-11` and `cswa-12` build every one of them; the next live CSWA run (0.11.4-rc2) measures them.
- `swx_describe` lists every variant and its measurement as `liveMeasured` (one table, `compat::OP_FACTS`).