diff --git a/content/docs/core.md b/content/docs/core.md
index 31e7edf..8a528e9 100644
--- a/content/docs/core.md
+++ b/content/docs/core.md
@@ -133,7 +133,7 @@ bun add -d @rig-c/core
 bun rigc build --rig rig.json --motion motion.json --out build/
 ```
 
-The published 2.74.0 package declares Bun 1.4.0 or later. The canonical npm package is `@rig-c/core`; `rig-c` is its convenience install name, and the command is `rigc`. Licensing by playback path: [Licensing](/docs/licensing/).
+The published package declares Bun 1.4.0 or later. The canonical npm package is `@rig-c/core`; `rig-c` is its convenience install name, and the command is `rigc`. Licensing by playback path: [Licensing](/docs/licensing/).
 
 ## Inputs and outputs
 
@@ -164,6 +164,7 @@ The published 2.74.0 package declares Bun 1.4.0 or later. The canonical npm pack
 - Start here: [/docs/getting-started/](/docs/getting-started/)
 
 <!-- SOURCES (not published)
+- install requirements: npm view @rig-c/core@2.81.0 engines and package.json at 666b30bf27b1447e71b02271b48253a8b47e7520 (2026-10-30): bun >=1.4.0; this is the declared runtime floor, not an older-runtime test result.
 - declared runtime refreshed from npm rig-c@2.74.0 tarball package.json engines.bun (>=1.4.0); recorded cases below retain their original pins.
 - key-pose film: rig-c/core@539572d (539572d34bb034b9e905e30dc2d45bcd2650b34f, the commit that last changed it; identical at v2.64.1 51dba78) assets/rigc-keypose.gif — 234 frames, 600x405, 5 cs each = 11.70 s, opaque; encoded by scripts/encode-video.sh row core-keypose (20 fps, cropped to 600x404); caption from films/two-keypose/README.md (summary, shot table, honesty stance; shot 2 = spine-rigc@0.9.0 output); "spine-rigc" is drawn in the caption band of every frame
 - three cases: rig-c 2.61.0 on rig-c/examples@c239b09 family/roa/out/rig — verbatim fail.log / pass.log / guide.txt of each case (case run 2026-10-19)
diff --git a/content/docs/getting-started.md b/content/docs/getting-started.md
index 0408467..c63c3d8 100644
--- a/content/docs/getting-started.md
+++ b/content/docs/getting-started.md
@@ -22,7 +22,7 @@ Repository: <https://github.com/rig-c/getting-started>
 
 This builds a published example character whose painting and See-through layers are already made. Its joints are hand-placed, so this route needs **Bun 1.4.4** (the version tested below), `curl` and `tar`, with a macOS or Linux shell. The new reproduction below ran on Linux; Windows was not tested. Painting and layer generation happen before this recipe, on the [full guide's GPU or hosted routes](https://github.com/rig-c/getting-started#readme). A person still judges how the result looks.
 
-**Choose a versioned run.** The commands here pin four packages, their dependency overrides, the helper-script commit and the example input commit. They reproduce this page's historical package combination; they are not an install of the latest packages. For a rolling-latest walkthrough, use the [upstream guide](https://github.com/rig-c/getting-started#try-it-first) and its existing [scheduled CPU walkthrough CI](https://github.com/rig-c/getting-started/actions/workflows/adapters.yml). Its [tested-with record](https://github.com/rig-c/getting-started/blob/main/docs/REFERENCE.md#tested-with) has its own date and versions. That job polls npm daily and runs weekly regardless.
+**Choose a versioned run.** The commands here pin four packages, their dependency overrides, the helper-script commit and the example input commit. They reproduce this page's historical package combination; they are not an install of the latest packages. For a rolling-latest walkthrough, use the [upstream guide](https://github.com/rig-c/getting-started#try-it-first--no-gpu-about-3-minutes) and its existing [scheduled CPU walkthrough CI](https://github.com/rig-c/getting-started/actions/workflows/adapters.yml). Its [tested-with record](https://github.com/rig-c/getting-started/blob/main/docs/REFERENCE.md#tested-with) has its own date and versions. That job polls npm daily and runs weekly regardless.
 
 ### What the time records include
 
@@ -244,6 +244,7 @@ The full guide — making your own character, the GPU routes, troubleshooting an
 The example inputs in step 2 come from [rig-c/exam](https://github.com/rig-c/exam) and are **CC BY-NC 4.0**, not MIT: free to use with credit, not for commercial use, and what you build from them carries that licence. Which terms each route brings: [Licensing](/docs/licensing/).
 
 <!-- SOURCES (not published)
+- rolling-latest upstream guide fragment: getting-started README at 78f44c247fbd7d8c416f597149bad83bbea65b99 (2026-10-30), Contents and heading: #try-it-first--no-gpu-about-3-minutes; historical recipe below remains separately pinned.
 - The idle clip (drawn by @rig-c/play; the step-10 wave beside it came from the same run until 2026-10-21): one run of steps 1-10 as written, 2026-10-20, Apple M4, Bun 1.4.2, from an empty folder. `bun add` resolved @rig-c/parts 1.42.0, rig-c 2.64.1, @rig-c/pose 0.36.0, @rig-c/play 1.12.0 (bun.lock; `npm view` the same day gave the same four); rig-c/getting-started cloned at 2928714. Every command exited 0. Hand-placed keypoints (step 6's default), so step 7 printed 0 LINT lines; the 1.41.1 LINT line in step 6 is for the script-detected joints and was not re-run here.
 - Expected lines matched: st_input_full.png 2432x2432; 29 layer(s): 29 with opaque pixels; 0 WARN line(s); head_box [616,32,1132,548]; st_input_head.png 516x516; 22 part(s) on a 832x1216 rig; recomposite mean |d|=1.13, within 8: 97.8%; coverage: 42 bone(s) - 29 checked and clean; check: PASS; 11 of 12 bar(s) measured, 1 skipped (STRESS_OFF_CHAIN); build: PASS; wave.ts: 73 frame(s), 3 s at 24 fps; GATE rigc build --profile spine-html: green; wrote wave.motion.json - animations: idle, wave.
 - getting-started-idle.webp + getting-started-idle-poster.png: scripts/play-capture-webp.ts (written as scripts/getting-started-idle-webp.ts, generalised 2026-10-20; re-run through it with --root <run> --model out/check/build/skeleton.model.json --anim idle --fps 12 --frames 40 the WebP and poster came out md5-identical to these; Bun 1.4.2, Playwright 1.63.0 / Chromium 153 headless with --use-angle=metal, libwebp 1.6.0 img2webp) on the run's out/check/build/skeleton.model.json, drawn by the run's own @rig-c/play 1.12.0 through rig-c/getting-started@2928714 scripts/play.html + serve.ts (`bun serve.ts --port 0`, page /?model=out/check/build/skeleton.model.json&anim=idle), html/body/#stage made transparent and #bar hidden. Page on Playwright's fake clock (paused, so its rAF loop does not step); frame i = RigPlayer.seek(i/12) + draw(), screenshot of #stage with omitBackground -> RGBA PNG. idle duration from loadModel's assets.animations: 3.3333 s = 40 frames at 12 fps. Framing as rigc render --max 640: union of player.bounds() over the idle at 60 fps, padded 4% of the long side, long side 640 -> 317x640 (scale 0.50395), the same size as out/idle.png. Two runs gave identical frames (40 PNGs concatenated md5 53c314fdcbd3f61647bb065d413c16e3). WebP: lossy q85, -m 6, -exact, alpha lossless (libwebp default alpha_quality 100; decoded alpha equals the capture's, max error 0); frame durations round((i+1)*1000/12)-round(i*1000/12) = 27 x 83 ms + 13 x 84 ms = 3333 ms per loop; animation background colour set to 0,0,0,0 with webpmux. 713104 bytes (md5 6ac3e8c0b31c1949dc368cfa245e54a4); lossless was 2363528 bytes; q75 596356, q70 581414 - the lossless alpha plane alone is about 12 KB a frame, so no exact-alpha encode of this clip comes near 500 KB. Poster = captured frame 0, 121305 bytes, md5 a124eb937745fc4a36c0a32bd01285ed. Alpha: corners 0, interior 255, 3430 edge px at alpha 30-225 on frame 0; no grey fringe - composited over 232 the edge band's mean luminance differs from rigc render's by +0.02 (a fringe would bias it). Loop seam (scripts/loop-seam.py over --bg #070b18): captured frames 0.451 vs max step 0.552; WebP 0.865 vs 0.877 PASS (over #e8e8e8: 0.823 vs 0.868 PASS). Against rigc render (out/idle.png, md5 b958dbf8574d7939667d61a47d46b26a, composited capture over 232,232,232): MAD frame 0 1.655, mean over the 40 frames 1.674, max 1.771; interior 3.52, edge band 18.2, outside 0.01; silhouettes differ by 1923 of ~57k px. Same pose and framing; the difference is resampling - the browser scales the part images (CSS transform, 0.504) and rigc rasterizes them itself, so line art and edges land up to a pixel apart and play's is a little softer.
diff --git a/content/docs/licensing.md b/content/docs/licensing.md
index 459dcf7..5b600d1 100644
--- a/content/docs/licensing.md
+++ b/content/docs/licensing.md
@@ -12,7 +12,7 @@ Three separate questions, answered separately: the licence of rig-c's own code,
 
 ## rig-c's own code: MIT
 
-`rig-c` (core), `@rig-c/parts`, `@rig-c/pose`, `@rig-c/review`, `@rig-c/play`, `@rig-c/pixi`, `@rig-c/phaser` and `@rig-c/three` are MIT — each repository's `LICENSE`. That grant covers rig-c's code. It says nothing about Spine, and it does not cover the sample assets below.
+`@rig-c/core` (`rigc`), `@rig-c/parts`, `@rig-c/pose`, `@rig-c/review`, `@rig-c/play`, `@rig-c/pixi`, `@rig-c/phaser` and `@rig-c/three` are MIT — each repository's `LICENSE`. That grant covers rig-c's code. It says nothing about Spine, and it does not cover the sample assets below.
 
 ## Spine terms: they follow what plays the output
 
@@ -38,6 +38,7 @@ Esoteric Software's licence pages read *Last updated April 5, 2025* when this pa
 The image models that generated those paintings, and See-through itself, have their own terms; each repository's `NOTICE.md` records them, and [rig-c/parts `NOTICE.md`](https://github.com/rig-c/parts/blob/main/NOTICE.md) collects See-through's. No rig-c package contains See-through code or weights.
 
 <!-- SOURCES (not published)
+- canonical core package identity: core README at 666b30bf27b1447e71b02271b48253a8b47e7520, canonical npm package / convenience install name; npm view @rig-c/core@2.81.0 name license (2026-10-30): @rig-c/core, MIT.
 - MIT per repo: LICENSE files of rig-c/core, parts, pose, review, play, adapters (gh api .../LICENSE)
 - "what the terms turn on is the integration", "neither adds those terms nor removes them": rig-c/core NOTICE.md "What that means for rigc" + README "Licensing, stated plainly" (origin/main)
 - play/adapters link nothing from Esoteric, own core: rig-c/getting-started README "On a web page" licence line; rig-c/adapters README:99-104
diff --git a/content/docs/parts.md b/content/docs/parts.md
index f423223..655800c 100644
--- a/content/docs/parts.md
+++ b/content/docs/parts.md
@@ -99,7 +99,7 @@ npm i -D @rig-c/parts
 rig-parts --help
 ```
 
-Declares [Bun](https://bun.sh) 1.4.0 or later in the published 1.68.0 package; rig-c comes with it as a dependency. Which Spine terms apply depends on what plays the output: see [Licensing](/docs/licensing/).
+The published package declares [Bun](https://bun.sh) 1.4.0 or later; `@rig-c/core` comes with it as a dependency. Which Spine terms apply depends on what plays the output: see [Licensing](/docs/licensing/).
 
 ## Inputs and outputs
 
@@ -133,6 +133,7 @@ From the [1.68.0 README](https://github.com/rig-c/parts/blob/89a7671229d155976fc
 - Character 1's own files: [its folder in rig-c/examples](https://github.com/rig-c/examples/tree/c239b09328ca7364361b9761fac817502c09ede0/family/roa)
 
 <!-- SOURCES (not published)
+- install requirements: npm view @rig-c/parts@1.78.0 engines dependencies and package.json at 97a123395b83f5bee9ebce6407735f2a408855a6 (2026-10-30): bun >=1.4.0, @rig-c/core ^2.72.0; feature/config references below retain their historical pins.
 - hero clip parts-explode.mp4 + poster: scripts/parts-explode.py (reads rig-c/examples@c239b09 family/roa/out/parts/*.png + parts.json via git show; assembled frames = parts.json x/y in list order, placement vs out/recomposite_rig.png mean abs diff 0.0134; grid layout and dissolves are illustration)
 - Character 1 assets: rig-c/examples@c239b09 family/roa — docs/roa_layers_sheet.png, out/parts/*.png, out/parts.json (17 parts, recomposite: mean_abs 0.873, within_share 0.9831, uncovered_error_px 6041, holes[0] 834 px bordering bottomwear), out/recomposite_rig.png, out/recomposite_error_rig.png, docs/idle.gif (41 frames, 80 ms), out/check/contact.png
 - red = uncovered error, blue = covered error (>40 off): rig-c/parts docs/AUTHORING.md lines 2200, 2633
diff --git a/content/docs/play.md b/content/docs/play.md
index d717d96..616e756 100644
--- a/content/docs/play.md
+++ b/content/docs/play.md
@@ -79,13 +79,13 @@ Runs in the browser as an ES module; rig-c's posing core is bundled, so there is
 | (optional) your own `PosedFrame` records, via `DomRenderer.render` | |
 
 ## Limits
-From the README's *Status* and *No counterpart in 1.0.0* (it has no limits section):
-- **Status: "Production."** — 60 fps on every engine measured for the practical scenario (README §"Status").
-- **Reads rigc's model only**: "a Spine export is refused by name"; `loadModel` reads a compiled model document and nothing else (README §"Why", §"No counterpart in 1.0.0").
-- **Which render mode is faster where is "not measured on a device yet"**, so the default stays `dom` (README §"Status").
-- **Dark/two-colour tint stays out of scope in DOM mode**; RGB tint needs reference-filter support, Safari 15+ (README §"Status").
-- **Clipping caveats**: an inverse polygon matches Spine's clipper only when convex, and the cost of many `clip-path`s on real-device Safari is not measured (README §"Status").
-- **No per-slot changes from code** (attachment, colour) and no `scale` on load; scaling the root element is the caller's (README §"No counterpart in 1.0.0").
+From [REFERENCE.md §Status](https://github.com/rig-c/play/blob/main/docs/REFERENCE.md#status) and [MIGRATION.md §No counterpart in 1.0.0](https://github.com/rig-c/play/blob/main/docs/MIGRATION.md#no-counterpart-in-100):
+- **Status: "Production."** — 60 fps on every engine measured for the practical scenario (REFERENCE.md §"Status").
+- **Reads rigc's model only**: "a Spine export is refused by name"; `loadModel` reads a compiled model document and nothing else ([README §Why](https://github.com/rig-c/play#why), MIGRATION.md §"No counterpart in 1.0.0").
+- **Which render mode is faster where is "not measured on a device yet"**, so the default stays `dom` (REFERENCE.md §"Status").
+- **Dark/two-colour tint stays out of scope in DOM mode**; RGB tint needs reference-filter support, Safari 15+ (REFERENCE.md §"Status").
+- **Clipping caveats**: an inverse polygon matches Spine's clipper only when convex, and the cost of many `clip-path`s on real-device Safari is not measured (REFERENCE.md §"Status").
+- **No per-slot changes from code** (attachment, colour) and no `scale` on load; scaling the root element is the caller's (MIGRATION.md §"No counterpart in 1.0.0").
 
 ## Go deeper
 - [Measured runtime costs and unresolved results](/docs/performance/) — historical package comparisons, with CPU, GPU and presentation boundaries.
@@ -97,6 +97,7 @@ From the README's *Status* and *No counterpart in 1.0.0* (it has no limits secti
 - [How it works](https://github.com/rig-c/play#how-it-works)
 - [npm](https://www.npmjs.com/package/@rig-c/play) · [CHANGELOG](https://github.com/rig-c/play/blob/main/CHANGELOG.md)
 <!-- SOURCES (not published)
+- six Limits bullets rechecked against @rig-c/play 1.21.1 (npm gitHead = main 7f84c22f1fb8dacbf8db4aa1333efa3fbac6120a, 2026-10-30): docs/REFERENCE.md Status (production, practical scenario, mode cost, tint and clipping caveats); docs/MIGRATION.md No counterpart in 1.0.0 (model input, no load scale, no per-slot API); README Why (Spine export refused). Older line references below document the original 1.12.0 page/recording.
 - live player picks: the 20 Family rigs = rig-c/examples@c239b09 family/ (same pin as /examples/), each with one animation `idle` (skeleton.model.json animations); sizes measured on jsDelivr: skeleton.png 1.19–2.00 MB + model ~0.09 MB compressed
 - live player: @rig-c/play 1.12.0 from jsDelivr; Character 1 office model assets/rigs/character-1-swing = the Pose case's swing-build (Character 1 rig @examples c239b09 + retargeted cup swing; rig-c 2.61.0, @rig-c/pose 0.35.0)
 - minimal page: rig-c/play README "A minimal page" (lines 301-326), style attribute only in the sample code
diff --git a/content/docs/pose.md b/content/docs/pose.md
index febb22b..c3dfe41 100644
--- a/content/docs/pose.md
+++ b/content/docs/pose.md
@@ -118,7 +118,7 @@ bun run rig-pose retarget --rig out/rig --map map.json --pose swing.pose.clean.j
 
 Installing `@rig-c/pose` provides the `rig-pose` executable used above. The unscoped convenience name `rig-pose` installs the same package. See [package names and aliases](/#package-names-and-aliases).
 
-The published 1.2.1 package declares [Bun](https://bun.sh) 1.4.0 or later on `PATH`; rig-c comes as a dependency.
+The published package declares [Bun](https://bun.sh) 1.4.0 or later on `PATH`; `@rig-c/core` comes as a dependency.
 
 ## Inputs and outputs
 | In | Out |
@@ -145,7 +145,6 @@ See [derived depth](https://github.com/rig-c/pose/blob/35d4d140fa45410051117251c
 - **The input is joint positions**, not video. Pose extraction and camera fitting from pixels happen outside the shipped CLI; without a declared camera, the picture is read as a fixed camera.
 - **It remains a 2D retarget.** Depth signs and draw-order keys are constrained inferences. Large turns and intense motion on a designed character can need replacement drawings declared in the map; a green structural gate does not supply missing art.
 - **No eyelid, mouth or finger keys.** Body-18 has no hand points; head yaw/pitch inference does not provide expressions or lip-sync.
-- **Gait quality still has open work** (checked 2026-10-27): [#444](https://github.com/rig-c/pose/issues/444) tracks contralateral arm depth from step phase; [#445](https://github.com/rig-c/pose/issues/445) tracks keeping shoe soles level across the cycle; [#446](https://github.com/rig-c/pose/issues/446) tracks continuous depth-derived receding-limb shading. The released depth, plant and scale options do not establish those fixes.
 
 ## Go deeper
 - [Try it](https://github.com/rig-c/pose#try-it) — a self-contained wave example in the repository
@@ -157,8 +156,9 @@ See [derived depth](https://github.com/rig-c/pose/blob/35d4d140fa45410051117251c
 - [What it does not do](https://github.com/rig-c/pose#what-it-does-not-do)
 - [npm](https://www.npmjs.com/package/@rig-c/pose) · [CHANGELOG](https://github.com/rig-c/pose/blob/main/CHANGELOG.md)
 <!-- SOURCES (not published)
-- current capability/limits: npm @rig-c/pose@1.2.1 README and package.json; docs/REFERENCE.md read at npm gitHead 35d4d140fa45410051117251c9e55a9fb36f4920 (not shipped in npm tarball): Derived depth, Draw order incl forearm layout, The legs incl depth-scale, What it does not do. The reference still carries old blanket wording about front/back; scoped flag contracts establish the available paths.
-- live issue checks: gh issue view 444/445/446 --repo rig-c/pose --comments; all OPEN with no comments, 2026-10-27. Public page describes the general issues only.
+- depth/projection/feet and limits baseline: npm @rig-c/pose@1.2.1 README and package.json; docs/REFERENCE.md read at npm gitHead 35d4d140fa45410051117251c9e55a9fb36f4920 (not shipped in npm tarball): Derived depth, Draw order incl forearm layout, The legs incl depth-scale, What it does not do. The reference still carries old blanket wording about front/back; scoped flag contracts establish the available paths.
+- install requirements: npm view @rig-c/pose@1.5.3 engines dependencies and package.json at a6a665c121db0a2d3de8ddd459d111eb179fc893 (2026-10-30): bun >=1.4.0, @rig-c/core ^2.60.0.
+- gait issue recheck (2026-10-30): pose#444 and #446 CLOSED completed; #445 CLOSED not_planned. CHANGELOG at a6a665c121db0a2d3de8ddd459d111eb179fc893, 1.3.0: authored arm swing #448 and --depth-shade #450. Obsolete open-work bullet removed; no claim that declined work shipped.
 - reproducible swing inputs (#49): assets/rigs/character-1-swing/keypoints.json (sha256 76b3152592362524921dc44843b04b525bc98ef6d873a579fca7d5a83318e3fd) and map.json (sha256 cf27eef4c4fa173249807f68e3531e670eaf4a6f0caeb663c0eda039d1e1cbfb) are the original hand-placed inputs; scripts/pendulum.py with no arguments runs the shipped preset, clean, retarget and a byte-identical model comparison in a new isolated workspace, never overwriting model/art. It fetches only rig.json/motion.json from https://github.com/rig-c/examples/tree/c239b09328ca7364361b9761fac817502c09ede0/family/roa/out/rig; uses the committed unchanged PNGs; pins Bun 1.4.2, @rig-c/pose 0.35.0, rig-c 2.61.0 and @esotericsoftware/spine-core 4.3.13 with an embedded frozen lock and fresh cache. It preserves the recorded out/rig and swing-build sibling layout: --copy-images emits skeleton.images as ../swing-build/, and spine.sha256 hashes the exact serialized skeleton.json bytes, including that output-relative metadata. Positional keypoints/out generation retains its old defaults.
 - pendulum swing (2026-10-20, issue #42; owner: the arm movement read as awkward): scripts/pendulum.py keypoints.json swing.pose.json --fore 4 --centre 2.5 --upper 0 --lag 0.15 (period 1.5 s, 2 periods, 73 frames 0..3 s inclusive, frame 72 = frame 0; rest alignment -1.19 / +0.37 / 2.553 as before) -> rig-pose clean --loop --hold off --rest 1 (@rig-c/pose 0.35.0; loop true, no holds; --rest 1 because the input is an exact sine and the default window, wrapped over the repeated last frame, moved frame 72 3.3 px off frame 0) -> rig-pose retarget --rig out/rig --map map.json --pose … --name swing --profile spine-html --build … (rig-c 2.61.0; GATE green, 4 LAND, no folds, 26 IDLE-DROPPED); model differs from the previous one only in the swing animation and the skeleton digest; part PNGs md5-identical. Tracks: sleeve_r1 -1.489..6.491 deg (key 0 = key 72 = 4.852), sleeve_r0 0.001, chest -0.001. Thresholds stated before measuring, from this page: holes <= 1500 px worst frame, covered <= idle + 400 = 1326 px; measured on the fixed viewport (rigc render --max 1400 --viewport -440.24,-65.03,875.10,2295.92; scripts holes.py / core_hidden.py of the earlier cases, which first reproduced idle 926, previous 1242/1124, ±6° 5173 exactly): this swing holes 1476, covered 1295. Symmetric sweep about the rest (lag 0.15 s), holes/covered: ±1.5 1482/866, ±2 2019/914, ±3 3117/964, ±4 4184/1055, ±5 5045/1136, ±6 5632/1243, ±8 6430/1392, ±10 6642/1451; with the upper arm at a third of the forearm, all fail holes (±3: 4601/1317); lagged chains (upper 1-3 deg, forearm 1-3 deg, lag 0.25-0.35 s) fail both (holes 3655-5769, covered 1390-2736). ±3.5 centred 2.5: 1122/1240. The idle alone: holes 1138. Most counted hole px are the arm-leg gap widening; the tear is the notch below the cuff (pose-swing-tear.jpg: rest f0, symmetric ±4 frame 13, this swing frame 13; crops of those renders at 2x, rig-c 2.64.1). The hand bone past the wrist (sleeve_r2) is not mapped: body-18 has no joint beyond the wrist
 - hero clip (2026-10-20, issue #42): scripts/pose-swing.py --window 0:72 (the default): motion frames 0..71 (72 frames, 3.000 s; frame 72 repeats frame 0), rotated to start at frame 18 (sleeve_r1 0.149 deg, the rest pose; the poster); rigc render rig-c 2.61.0; 806x792, 213979 bytes. Seam on the encoded file (MAD 0-255): rig panel 0.880 vs max step 1.127, whole frame 0.936 vs 1.171, both PASS. Starting at the swing's centre (frame 4, 2.222 deg) failed encoded (rig panel 1.021 vs 0.972, whole frame 1.111 vs 0.925; lossless 0.804 vs 0.864). pose-swing-ends.jpg: scripts compose.py of the earlier cases over rigc render --max 720 --poser core of the build, frames 13 and 31 (outermost -1.489, innermost 6.491); its rig panels match the hero's frames (MAD 0.083, the inset corners)
diff --git a/content/exam.md b/content/exam.md
index 27797ee..bdbda0b 100644
--- a/content/exam.md
+++ b/content/exam.md
@@ -4,23 +4,25 @@ title: rig-c/exam — Open-Book Rigging Benchmark & Leaderboard for AI Agents
 description: An open-book exam for AI coding agents: build a 2D rig from one painting, make it move, correct a broken draft, add a soft chest and a dance, using only the published rig-c packages. 2026-10-20 — eight sittings from three model families, with a blind score for the dance.
 keywords: rig-c exam, AI agent benchmark, LLM coding benchmark, 2D rigging benchmark, Claude Fable, Claude Opus, Claude Sonnet, Claude Haiku, Gemini Flash, Gemini Argon, gpt-6-astra, gpt-6-luna, Sol 6.1
 repoUrl: https://github.com/rig-c/exam
-summaryForLlms: rig-c/exam benchmark rules and results: Basic Exam (13 independent sittings — 5 of the three-task round on 2026-10-12/13, 8 of the five-task round on 2026-10-20 from three model families, with the dance judged blind; 2 runs kept and not counted) and Advanced Exam (R1 walk, R2 run, R3 dance — 18 sittings of 6 models).
+summaryForLlms: rig-c/exam benchmark rules and results, historical snapshot at 30cae36: Basic Exam (13 independent sittings — 5 of the three-task round on 2026-10-12/13, 8 of the five-task round on 2026-10-20 from three model families, with the dance judged blind; 2 runs kept and not counted) and Advanced Exam (R1 walk, R2 run, R3 dance — 18 sittings of 6 models).
 ---
 # rig-c/exam — Does Your AI Model Pass?
 
 **An open-book exam for AI agents: build a rig from one painting with the published rig-c tools, and write down what you expected before each step.**
 
+The rules and leaderboards below are a historical snapshot at [`30cae36`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1), covering the Basic sittings through 2026-10-20 and the Advanced results at that commit. For the current sitting count and results, use the [exam repository](https://github.com/rig-c/exam#results).
+
 - **Repository**: <https://github.com/rig-c/exam>
-- **Prompt (Basic)**: <https://github.com/rig-c/exam/blob/main/PROMPT.md>
-- **Advanced Edition**: <https://github.com/rig-c/exam/tree/main/advanced>
-- **Results**: <https://github.com/rig-c/exam#results> · **The proctor's report**: <https://github.com/rig-c/exam/blob/main/results/REPORT.md>
+- **Current prompt (Basic)**: <https://github.com/rig-c/exam/blob/main/PROMPT.md>
+- **Current Advanced Edition**: <https://github.com/rig-c/exam/tree/main/advanced>
+- **Current results**: <https://github.com/rig-c/exam#results> · **Current proctor's report**: <https://github.com/rig-c/exam/blob/main/results/REPORT.md>
 - **License**: CC BY-NC 4.0
 
 The rig-c tools are written for agents that cannot see what they build. `rig-c/exam` measures whether that holds for a fresh agent session that has never seen the tools before: the only help is what ships inside the installed packages (`README.md`, `docs/`, `--help`). Before every command that changes a file, the examinee must record what it read and what it expects in `JOURNAL.md`. Where expectation and reality differ — where the agent says *WTF?* — a tool message or guide was missing something.
 
 ## 2026-10-20 sitting — the five-task round
 
-On 2026-10-20 eight models sat all five Basic tasks on `@rig-c/parts` 1.15.0 and `rig-c` 2.47.1. In the proctor's words: *"Eight models from three families sat all five tasks …, each in a fresh session, each regraded by the proctor from the files it handed in. Every regrade agrees with the run's own grader lines."* Their rows are the last eight of the [Basic leaderboard](#basic-exam-leaderboard-tasks-01-build-to-05-dance) below; the write-up is [the proctor's report](https://github.com/rig-c/exam/blob/30cae36/results/REPORT.md#2026-10-20--eight-sittings-of-the-five-task-round).
+On 2026-10-20 eight models sat all five Basic tasks on `@rig-c/parts` 1.15.0 and `rig-c` 2.47.1. In the proctor's words: *"Eight models from three families sat all five tasks …, each in a fresh session, each regraded by the proctor from the files it handed in. Every regrade agrees with the run's own grader lines."* Their rows are the last eight of the [Basic leaderboard](#basic-exam-leaderboard-tasks-01-build-to-05-dance) below; the write-up is [the proctor's report](https://github.com/rig-c/exam/blob/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/REPORT.md#2026-10-20--eight-sittings-of-the-five-task-round).
 
 ### What the machine lines cannot see
 
@@ -37,13 +39,13 @@ Task 05's fourteen lines read whether the dance holds together — the loop clos
 | `gemini-4-argon-high` | 14 / 14 | 3 |
 | `gpt-6-luna` | 5 / 14 (the build failed) | 0 |
 
-> "The six runs that passed every line were scored from 3 to 9. … Passing the lines is what a run needs to be judged at all; it says almost nothing about the score." — [the proctor's report](https://github.com/rig-c/exam/blob/30cae36/results/REPORT.md#what-the-machine-lines-cannot-see), rig-c/exam
+> "The six runs that passed every line were scored from 3 to 9. … Passing the lines is what a run needs to be judged at all; it says almost nothing about the score." — [the proctor's report](https://github.com/rig-c/exam/blob/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/REPORT.md#what-the-machine-lines-cannot-see), rig-c/exam
 
 The report adds two observations and frames them itself: *"each one sitting per model and one judge, so neither is a finding about the models"* — that the largest model of each family scored below the middle one, and that the highest score is the one run that eased its keys. The judge used one mark out of 10 for the whole clip; the clip and the score are in each run's `05-dance/`.
 
 ### What the journals found in the tools — now filed
 
-The day's `tools.md` items — handed in under task 05's directory and, in the report's words, *"reaching back over all five tasks"* — are grouped in the [report's tools table](https://github.com/rig-c/exam/blob/30cae36/results/REPORT.md#what-the-journals-found-in-the-tools-1), which now names the issue each row produced:
+The day's `tools.md` items — handed in under task 05's directory and, in the report's words, *"reaching back over all five tasks"* — are grouped in the [report's tools table](https://github.com/rig-c/exam/blob/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/REPORT.md#what-the-journals-found-in-the-tools-1), which now names the issue each row produced:
 
 | what the journals found | filed as |
 | --- | --- |
@@ -59,23 +61,23 @@ The rest of the table's rows name no issue, and one (`MOTION.md` is missing) is
 
 ## Basic Exam Leaderboard (Tasks 01-build to 05-dance)
 
-The exam's generated results table as of [`30cae36`](https://github.com/rig-c/exam/tree/30cae36) — **13 independent sittings so far**, each cell the lines passed of the lines graded, linking to the run. Tasks 04 and 05 and a sixth line on task 03 arrived with round 3 (`@rig-c/parts` 1.15.0, `rig-c` 2.47.1), so the 2026-10-12/13 rows have no cells there:
+The exam's generated results table as of [`30cae36`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1) — **13 independent sittings in this snapshot**, each cell the lines passed of the lines graded, linking to the run. Tasks 04 and 05 and a sixth line on task 03 arrived with round 3 (`@rig-c/parts` 1.15.0, `rig-c` 2.47.1), so the 2026-10-12/13 rows have no cells there:
 
 | model that did the work | run, as submitted | 01-build | 02-move | 03-correct | 04-soft | 05-dance | the dance, judged | minutes | tokens |
 | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
-| Claude Haiku (alias `haiku`; exact identifier not recorded) | [`haiku`](https://github.com/rig-c/exam/tree/30cae36/results/claude-haiku/2026-10-12-haiku), 2026-10-12 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36/results/claude-haiku/2026-10-12-haiku/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36/results/claude-haiku/2026-10-12-haiku/02-move) | [5 / 5](https://github.com/rig-c/exam/tree/30cae36/results/claude-haiku/2026-10-12-haiku/03-correct) | — | — | — | 21 | 531k in all |
-| Gemini 3.8 Flash | [`gemini-3.8-flash`](https://github.com/rig-c/exam/tree/30cae36/results/gemini-3.8-flash/2026-10-13-gemini-3.8-flash), 2026-10-13 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36/results/gemini-3.8-flash/2026-10-13-gemini-3.8-flash/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36/results/gemini-3.8-flash/2026-10-13-gemini-3.8-flash/02-move) | [5 / 5](https://github.com/rig-c/exam/tree/30cae36/results/gemini-3.8-flash/2026-10-13-gemini-3.8-flash/03-correct) | — | — | — | 20 | 1.4M |
-| `gpt-6-sol` — whether that is a name for Sol 6.1 or a model of its own is not established | [`gpt-6-sol`](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-sol/2026-10-13-gpt-6-sol), 2026-10-13 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-sol/2026-10-13-gpt-6-sol/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-sol/2026-10-13-gpt-6-sol/02-move) | [5 / 5](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-sol/2026-10-13-gpt-6-sol/03-correct) | — | — | — | 14.1 | 258k |
-| Sol 6.1 (`gpt-6.1-sol`) | [`gpt-6.1-sol`](https://github.com/rig-c/exam/tree/30cae36/results/sol-6.1/2026-10-13-gpt-6.1-sol), 2026-10-13 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36/results/sol-6.1/2026-10-13-gpt-6.1-sol/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36/results/sol-6.1/2026-10-13-gpt-6.1-sol/02-move) | [5 / 5](https://github.com/rig-c/exam/tree/30cae36/results/sol-6.1/2026-10-13-gpt-6.1-sol/03-correct) | — | — | — | 15.3 | 270k |
-| Sol 6.1 (`gpt-6.1-sol`) — the one sitting that is this model end to end | [`gpt-6.1-sol-20261013`](https://github.com/rig-c/exam/tree/30cae36/results/sol-6.1/2026-10-13-gpt-6.1-sol-20261013), 2026-10-13 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36/results/sol-6.1/2026-10-13-gpt-6.1-sol-20261013/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36/results/sol-6.1/2026-10-13-gpt-6.1-sol-20261013/02-move) | [5 / 5](https://github.com/rig-c/exam/tree/30cae36/results/sol-6.1/2026-10-13-gpt-6.1-sol-20261013/03-correct) | — | — | — | 14.2 | 195k |
-| claude-fable-5-1 (Fable 5.1; every assistant message in the session log names it) | [`claude-fable-5-1`](https://github.com/rig-c/exam/tree/30cae36/results/claude-fable-5-1/2026-10-20-claude-fable-5-1), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36/results/claude-fable-5-1/2026-10-20-claude-fable-5-1/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36/results/claude-fable-5-1/2026-10-20-claude-fable-5-1/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36/results/claude-fable-5-1/2026-10-20-claude-fable-5-1/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36/results/claude-fable-5-1/2026-10-20-claude-fable-5-1/04-soft) | [14 / 14](https://github.com/rig-c/exam/tree/30cae36/results/claude-fable-5-1/2026-10-20-claude-fable-5-1/05-dance) | 9 / 10 | 52.9 | 642k |
-| claude-haiku-5-5 (Haiku 5.5; every assistant message in the session log names it) | [`claude-haiku-5-5`](https://github.com/rig-c/exam/tree/30cae36/results/claude-haiku-5-5/2026-10-20-claude-haiku-5-5), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36/results/claude-haiku-5-5/2026-10-20-claude-haiku-5-5/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36/results/claude-haiku-5-5/2026-10-20-claude-haiku-5-5/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36/results/claude-haiku-5-5/2026-10-20-claude-haiku-5-5/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36/results/claude-haiku-5-5/2026-10-20-claude-haiku-5-5/04-soft) | [13 / 14](https://github.com/rig-c/exam/tree/30cae36/results/claude-haiku-5-5/2026-10-20-claude-haiku-5-5/05-dance) — not: on_her_feet | 5 / 10 | 70.3 | 591k |
-| claude-opus-5-5 (the identifier on every assistant message of the session log) | [`claude-opus-5-5`](https://github.com/rig-c/exam/tree/30cae36/results/claude-opus-5-5/2026-10-20-claude-opus-5-5), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36/results/claude-opus-5-5/2026-10-20-claude-opus-5-5/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36/results/claude-opus-5-5/2026-10-20-claude-opus-5-5/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36/results/claude-opus-5-5/2026-10-20-claude-opus-5-5/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36/results/claude-opus-5-5/2026-10-20-claude-opus-5-5/04-soft) | [14 / 14](https://github.com/rig-c/exam/tree/30cae36/results/claude-opus-5-5/2026-10-20-claude-opus-5-5/05-dance) | 3 / 10 | 64 | 584k |
-| claude-sonnet-5-5 (the identifier on every assistant message of the session log) | [`claude-sonnet-5-5`](https://github.com/rig-c/exam/tree/30cae36/results/claude-sonnet-5-5/2026-10-20-claude-sonnet-5-5), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36/results/claude-sonnet-5-5/2026-10-20-claude-sonnet-5-5/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36/results/claude-sonnet-5-5/2026-10-20-claude-sonnet-5-5/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36/results/claude-sonnet-5-5/2026-10-20-claude-sonnet-5-5/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36/results/claude-sonnet-5-5/2026-10-20-claude-sonnet-5-5/04-soft) | [14 / 14](https://github.com/rig-c/exam/tree/30cae36/results/claude-sonnet-5-5/2026-10-20-claude-sonnet-5-5/05-dance) | 6 / 10 | 53.6 | 330k |
-| Gemini 3.8 Flash (opened at the host's 'medium' setting; the log records no effort) | [`gemini-3.8-flash-medium`](https://github.com/rig-c/exam/tree/30cae36/results/gemini-3.8-flash/2026-10-20-gemini-3.8-flash-medium), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36/results/gemini-3.8-flash/2026-10-20-gemini-3.8-flash-medium/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36/results/gemini-3.8-flash/2026-10-20-gemini-3.8-flash-medium/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36/results/gemini-3.8-flash/2026-10-20-gemini-3.8-flash-medium/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36/results/gemini-3.8-flash/2026-10-20-gemini-3.8-flash-medium/04-soft) | [14 / 14](https://github.com/rig-c/exam/tree/30cae36/results/gemini-3.8-flash/2026-10-20-gemini-3.8-flash-medium/05-dance) | 7 / 10 | 31.1 | 1.7M |
-| gemini-4-argon-high | [`gemini-4-argon-high`](https://github.com/rig-c/exam/tree/30cae36/results/gemini-4-argon/2026-10-20-gemini-4-argon-high), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36/results/gemini-4-argon/2026-10-20-gemini-4-argon-high/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36/results/gemini-4-argon/2026-10-20-gemini-4-argon-high/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36/results/gemini-4-argon/2026-10-20-gemini-4-argon-high/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36/results/gemini-4-argon/2026-10-20-gemini-4-argon-high/04-soft) | [14 / 14](https://github.com/rig-c/exam/tree/30cae36/results/gemini-4-argon/2026-10-20-gemini-4-argon-high/05-dance) | 3 / 10 | 71.3 | 1.6M |
-| gpt-6-astra (the thread's default effort; the log records none) | [`gpt-6-astra-medium`](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-astra/2026-10-20-gpt-6-astra-medium), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-astra/2026-10-20-gpt-6-astra-medium/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-astra/2026-10-20-gpt-6-astra-medium/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-astra/2026-10-20-gpt-6-astra-medium/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-astra/2026-10-20-gpt-6-astra-medium/04-soft) | [14 / 14](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-astra/2026-10-20-gpt-6-astra-medium/05-dance) | 4 / 10 | 33.1 | 186k |
-| `gpt-6-luna` (effort: the thread's default; no reasoning effort was set in any turn of the session's log, so the `medium` in the run's name is the name it was given, not a setting the log shows) | [`gpt-6-luna-medium`](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-luna/2026-10-20-gpt-6-luna-medium), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-luna/2026-10-20-gpt-6-luna-medium/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-luna/2026-10-20-gpt-6-luna-medium/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-luna/2026-10-20-gpt-6-luna-medium/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-luna/2026-10-20-gpt-6-luna-medium/04-soft) | [5 / 14](https://github.com/rig-c/exam/tree/30cae36/results/gpt-6-luna/2026-10-20-gpt-6-luna-medium/05-dance) — not: built, checked, closes, both_hands, whole_body, arms_bend_at_the_joints, on_her_feet, keeps_time, soft_in_the_dance | 0 / 10 | 29.1 | 399k |
+| Claude Haiku (alias `haiku`; exact identifier not recorded) | [`haiku`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-haiku/2026-10-12-haiku), 2026-10-12 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-haiku/2026-10-12-haiku/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-haiku/2026-10-12-haiku/02-move) | [5 / 5](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-haiku/2026-10-12-haiku/03-correct) | — | — | — | 21 | 531k in all |
+| Gemini 3.8 Flash | [`gemini-3.8-flash`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-3.8-flash/2026-10-13-gemini-3.8-flash), 2026-10-13 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-3.8-flash/2026-10-13-gemini-3.8-flash/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-3.8-flash/2026-10-13-gemini-3.8-flash/02-move) | [5 / 5](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-3.8-flash/2026-10-13-gemini-3.8-flash/03-correct) | — | — | — | 20 | 1.4M |
+| `gpt-6-sol` — whether that is a name for Sol 6.1 or a model of its own is not established | [`gpt-6-sol`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-sol/2026-10-13-gpt-6-sol), 2026-10-13 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-sol/2026-10-13-gpt-6-sol/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-sol/2026-10-13-gpt-6-sol/02-move) | [5 / 5](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-sol/2026-10-13-gpt-6-sol/03-correct) | — | — | — | 14.1 | 258k |
+| Sol 6.1 (`gpt-6.1-sol`) | [`gpt-6.1-sol`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/sol-6.1/2026-10-13-gpt-6.1-sol), 2026-10-13 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/sol-6.1/2026-10-13-gpt-6.1-sol/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/sol-6.1/2026-10-13-gpt-6.1-sol/02-move) | [5 / 5](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/sol-6.1/2026-10-13-gpt-6.1-sol/03-correct) | — | — | — | 15.3 | 270k |
+| Sol 6.1 (`gpt-6.1-sol`) — the one sitting that is this model end to end | [`gpt-6.1-sol-20261013`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/sol-6.1/2026-10-13-gpt-6.1-sol-20261013), 2026-10-13 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/sol-6.1/2026-10-13-gpt-6.1-sol-20261013/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/sol-6.1/2026-10-13-gpt-6.1-sol-20261013/02-move) | [5 / 5](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/sol-6.1/2026-10-13-gpt-6.1-sol-20261013/03-correct) | — | — | — | 14.2 | 195k |
+| claude-fable-5-1 (Fable 5.1; every assistant message in the session log names it) | [`claude-fable-5-1`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-fable-5-1/2026-10-20-claude-fable-5-1), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-fable-5-1/2026-10-20-claude-fable-5-1/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-fable-5-1/2026-10-20-claude-fable-5-1/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-fable-5-1/2026-10-20-claude-fable-5-1/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-fable-5-1/2026-10-20-claude-fable-5-1/04-soft) | [14 / 14](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-fable-5-1/2026-10-20-claude-fable-5-1/05-dance) | 9 / 10 | 52.9 | 642k |
+| claude-haiku-5-5 (Haiku 5.5; every assistant message in the session log names it) | [`claude-haiku-5-5`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-haiku-5-5/2026-10-20-claude-haiku-5-5), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-haiku-5-5/2026-10-20-claude-haiku-5-5/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-haiku-5-5/2026-10-20-claude-haiku-5-5/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-haiku-5-5/2026-10-20-claude-haiku-5-5/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-haiku-5-5/2026-10-20-claude-haiku-5-5/04-soft) | [13 / 14](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-haiku-5-5/2026-10-20-claude-haiku-5-5/05-dance) — not: on_her_feet | 5 / 10 | 70.3 | 591k |
+| claude-opus-5-5 (the identifier on every assistant message of the session log) | [`claude-opus-5-5`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-opus-5-5/2026-10-20-claude-opus-5-5), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-opus-5-5/2026-10-20-claude-opus-5-5/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-opus-5-5/2026-10-20-claude-opus-5-5/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-opus-5-5/2026-10-20-claude-opus-5-5/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-opus-5-5/2026-10-20-claude-opus-5-5/04-soft) | [14 / 14](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-opus-5-5/2026-10-20-claude-opus-5-5/05-dance) | 3 / 10 | 64 | 584k |
+| claude-sonnet-5-5 (the identifier on every assistant message of the session log) | [`claude-sonnet-5-5`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-sonnet-5-5/2026-10-20-claude-sonnet-5-5), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-sonnet-5-5/2026-10-20-claude-sonnet-5-5/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-sonnet-5-5/2026-10-20-claude-sonnet-5-5/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-sonnet-5-5/2026-10-20-claude-sonnet-5-5/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-sonnet-5-5/2026-10-20-claude-sonnet-5-5/04-soft) | [14 / 14](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/claude-sonnet-5-5/2026-10-20-claude-sonnet-5-5/05-dance) | 6 / 10 | 53.6 | 330k |
+| Gemini 3.8 Flash (opened at the host's 'medium' setting; the log records no effort) | [`gemini-3.8-flash-medium`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-3.8-flash/2026-10-20-gemini-3.8-flash-medium), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-3.8-flash/2026-10-20-gemini-3.8-flash-medium/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-3.8-flash/2026-10-20-gemini-3.8-flash-medium/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-3.8-flash/2026-10-20-gemini-3.8-flash-medium/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-3.8-flash/2026-10-20-gemini-3.8-flash-medium/04-soft) | [14 / 14](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-3.8-flash/2026-10-20-gemini-3.8-flash-medium/05-dance) | 7 / 10 | 31.1 | 1.7M |
+| gemini-4-argon-high | [`gemini-4-argon-high`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-4-argon/2026-10-20-gemini-4-argon-high), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-4-argon/2026-10-20-gemini-4-argon-high/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-4-argon/2026-10-20-gemini-4-argon-high/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-4-argon/2026-10-20-gemini-4-argon-high/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-4-argon/2026-10-20-gemini-4-argon-high/04-soft) | [14 / 14](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gemini-4-argon/2026-10-20-gemini-4-argon-high/05-dance) | 3 / 10 | 71.3 | 1.6M |
+| gpt-6-astra (the thread's default effort; the log records none) | [`gpt-6-astra-medium`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-astra/2026-10-20-gpt-6-astra-medium), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-astra/2026-10-20-gpt-6-astra-medium/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-astra/2026-10-20-gpt-6-astra-medium/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-astra/2026-10-20-gpt-6-astra-medium/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-astra/2026-10-20-gpt-6-astra-medium/04-soft) | [14 / 14](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-astra/2026-10-20-gpt-6-astra-medium/05-dance) | 4 / 10 | 33.1 | 186k |
+| `gpt-6-luna` (effort: the thread's default; no reasoning effort was set in any turn of the session's log, so the `medium` in the run's name is the name it was given, not a setting the log shows) | [`gpt-6-luna-medium`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-luna/2026-10-20-gpt-6-luna-medium), 2026-10-20 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-luna/2026-10-20-gpt-6-luna-medium/01-build) | [7 / 7](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-luna/2026-10-20-gpt-6-luna-medium/02-move) | [6 / 6](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-luna/2026-10-20-gpt-6-luna-medium/03-correct) | [9 / 9](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-luna/2026-10-20-gpt-6-luna-medium/04-soft) | [5 / 14](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/gpt-6-luna/2026-10-20-gpt-6-luna-medium/05-dance) — not: built, checked, closes, both_hands, whole_body, arms_bend_at_the_joints, on_her_feet, keeps_time, soft_in_the_dance | 0 / 10 | 29.1 | 399k |
 
 Minutes are for the tasks sat, together; tokens are fresh input plus output, cached input left out; the dance is judged by a person who does not know whose clip it is.
 
@@ -83,12 +85,12 @@ Kept, and not counted — 2 runs:
 
 | model that did the work | run, as submitted | lines passed | why it does not count |
 | --- | --- | --- | --- |
-| `gpt-6-astra` | [`codex-direct`](https://github.com/rig-c/exam/tree/30cae36/results/void/2026-10-13-codex-direct), 2026-10-13 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36/results/void/2026-10-13-codex-direct/01-build) · [7 / 7](https://github.com/rig-c/exam/tree/30cae36/results/void/2026-10-13-codex-direct/02-move) · [5 / 5](https://github.com/rig-c/exam/tree/30cae36/results/void/2026-10-13-codex-direct/03-correct) · — · — | a repeat in the session that had already seen its sub-agent's answers, by its own statement |
-| `gpt-5.6-sol` (the run is named for a persona, not the model) | [`gpt-6-astra`](https://github.com/rig-c/exam/tree/30cae36/results/void/2026-10-13-gpt-6-astra), 2026-10-13 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36/results/void/2026-10-13-gpt-6-astra/01-build) · [7 / 7](https://github.com/rig-c/exam/tree/30cae36/results/void/2026-10-13-gpt-6-astra/02-move) · [5 / 5](https://github.com/rig-c/exam/tree/30cae36/results/void/2026-10-13-gpt-6-astra/03-correct) · — · — | a repeat in the session that had already seen its sub-agent's report, by its own statement; and the run is named for a persona, not its model |
+| `gpt-6-astra` | [`codex-direct`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/void/2026-10-13-codex-direct), 2026-10-13 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/void/2026-10-13-codex-direct/01-build) · [7 / 7](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/void/2026-10-13-codex-direct/02-move) · [5 / 5](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/void/2026-10-13-codex-direct/03-correct) · — · — | a repeat in the session that had already seen its sub-agent's answers, by its own statement |
+| `gpt-5.6-sol` (the run is named for a persona, not the model) | [`gpt-6-astra`](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/void/2026-10-13-gpt-6-astra), 2026-10-13 | [4 / 4](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/void/2026-10-13-gpt-6-astra/01-build) · [7 / 7](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/void/2026-10-13-gpt-6-astra/02-move) · [5 / 5](https://github.com/rig-c/exam/tree/30cae36b236c9b60c78d45619a5cd4230ad6fea1/results/void/2026-10-13-gpt-6-astra/03-correct) · — · — | a repeat in the session that had already seen its sub-agent's report, by its own statement; and the run is named for a persona, not its model |
 
 ## Advanced Exam Leaderboard (Walk · Run · Dance Pose Retargeting)
 
-Starts from a built rig: the agent retargets recorded walk (R1), run (R2), and dance (R3) pose sequences with [`@rig-c/pose`](/docs/pose/), one round per fresh session, fixing what the diagnostic numbers report without ever seeing the footage. 18 sittings of 6 models: 15 graded on edition 0.20 and, from 2026-10-20, 3 on edition 0.21, whose R1 · R2 · R3 tables are 10 · 15 · 14 rows (the shin-scale row is 0.21's one new row; for a 0.20 sitting it is a what-if):
+Starts from a built rig: the agent retargets recorded walk (R1), run (R2), and dance (R3) pose sequences with [`@rig-c/pose`](/docs/pose/), one round per fresh session, fixing what the diagnostic numbers report without ever seeing the footage. The [Advanced snapshot at `30cae36`](https://github.com/rig-c/exam/blob/30cae36b236c9b60c78d45619a5cd4230ad6fea1/advanced/README.md) records 18 sittings of 6 models: 15 graded on edition 0.20 and, from 2026-10-20, 3 on edition 0.21, whose R1 · R2 · R3 tables are 10 · 15 · 14 rows (the shin-scale row is 0.21's one new row; for a 0.20 sitting it is a what-if):
 
 ### A dance retarget example
 
@@ -123,6 +125,7 @@ This recording illustrates pose retargeting; it is not an exam sitting or a lead
 5. **The grader rebuilds from inputs.** `grade.py` trusts only the handed-in config/spec files and rebuilds from scratch.
 
 <!-- SOURCES (not published)
+- snapshot revalidated 2026-10-30 through gh API: 30cae36 resolves to 30cae36b236c9b60c78d45619a5cd4230ad6fea1; README has 13 Basic independent sittings and advanced/README has 18 sittings of 6 models. Current main 409d66021a861df5065fcbaf312d13a591ce7f1a README reports 26 Basic independent sittings. This page preserves the complete historical tables rather than mixing that current total with old rows.
 - rig-c/exam at 30cae36 (PR #13, "docs(report): the 2026-10-20 tools table names the issue each row produced", on top of PR #8 f72ca5f, "results: 2026-10-20 — eight Basic sittings, the first Codex Advanced R1–R3, the blind dance scores"); main = 30cae36 on 2026-10-21.
 - Basic leaderboard + void table: README.md lines 13-42 (generated by results.py), rows copied verbatim; relative links made absolute to tree/30cae36. "13 independent sitting(s)" is line 14; the footer sentence is line 39.
 - 2026-10-20 section: results/REPORT.md lines 93-110 (eight sittings, opening sentences quoted from 95-97); "What the machine lines cannot see" lines 160-194 (table 168-177 verbatim; the quote is lines 183-185 with one sentence elided; observations framing line 187; one mark out of 10, line 179).