spine-rigc 2.55.0 → 2.56.0

package/README.md

@@ -601,7 +601,7 @@ commands take it and what its default is.
 | `repack <dir> --out <dir> --keep-pages` | **`skeleton.model.json` alone, for the pages already in place** — for a player that reads the document beside an atlas and pages that already shipped. The same read and stage rules, then `build --atlas-in` over the input atlas in place of the lift and the pack; it writes the document, and nothing else, only after its `pages` are shown to state the atlas as it stands, the rebuilt skeleton byte-identical to the one on disk (so `spine.sha256` names it — no acceptance flag here) and the gate green. `--out` is usually the build's own directory and must not already hold a `skeleton.model.json`; a page or region line the document has no field for is refused by name ([AUTHORING §0.4](docs/AUTHORING.md)) |
 | `ingest <skeleton.json> --out <dir>` | `build` run backwards: reads a Spine 4.3 skeleton and writes the rig spec and motion spec that **rebuild it**, plus a findings report naming everything it could not carry. a skeleton that declares no stage is carried as declaring none, `--stage x,y,w,h` adds a box to one — and is refused, rather than ignored, beside one that declares a box — and `--images <dir>` writes the spec's own images directory — the opposite direction from `build --images`, which overrides it — so the rebuild carries no flag at all |
 | `explain --rig … --motion …` | the compiled rig as a table — every bone with its resolved parent, the slots in draw order, every timeline key by key. Writes nothing. What to reach for when a rig compiles and still looks wrong |
-| `render --candidate <dir>` | PNG frames plus a contact sheet, in `render/`. `--hide <slot,…>` or `--slot <slot,…>` draws part of the rig on the **same grid** as the whole, so the two frames overlay and the difference is the part; `frames.json` records the subset and `check` refuses such a set as a reference. `--geometry` adds a `geometry.json` per set: every frame's bone world transforms and skinned attachment vertices, with the rest geometry, on the frames' own grid. `--folds` adds a `folds.json` per set and prints one line per mesh with a triangle turned over against the setup pose: the frames, the worst frame's count and which triangles, their world box and their setup-pose area. `--pass-through <cloth>:<leg>` adds a `pass-through.json` per set and prints the frames where a cloth drawn over a leg has its side inside the leg's drawn outline, with the depth and the world box — a reading, not a gate. `--environment <schedule.json>` draws the rig under the wind and gravity its physics answers, in time — the same frames on the core and on spine-core; an inspection of the response the rig states, not a scene |
+| `render --candidate <dir>` | PNG frames plus a contact sheet, in `render/`. `--hide <slot,…>` or `--slot <slot,…>` draws part of the rig on the **same grid** as the whole, so the two frames overlay and the difference is the part; `frames.json` records the subset and `check` refuses such a set as a reference. `--geometry` adds a `geometry.json` per set: every frame's bone world transforms and skinned attachment vertices, with the rest geometry, on the frames' own grid. `--folds` adds a `folds.json` per set and prints one line per mesh with a triangle turned over against the setup pose: the frames, the worst frame's count and which triangles, their world box and their setup-pose area. `--pass-through <cloth>:<leg>` adds a `pass-through.json` per set and prints the frames where a cloth drawn over a leg has its side inside the leg's drawn outline, with the depth and the world box — a reading, not a gate. `--angle-range <bone>:<relative-to>` prints how far one bone turns against another over each set: the lowest and highest angle and the frames they fall on — a reading, not a gate. `--viewport <x,y,w,h>` or `--viewport-from <frames.json>` draws over a given box instead of a fitted one, so two builds land on one pixel grid; a slot that reaches past it is named by set and frame, and `frames.json` records it. `--environment <schedule.json>` draws the rig under the wind and gravity its physics answers, in time — the same frames on the core and on spine-core; an inspection of the response the rig states, not a scene |
 | `preview --candidate <dir>` | one self-contained `.html` that plays it, headed by the line `validate <dir>` prints for it and the rigc version — a refused candidate is still previewed, and its header says so in the gate's words. Repeat `--candidate` for one page with a pane per candidate, in the order given; a green `build` ends by naming this command for its own `--out` |
 | `vote --candidate a --candidate b` | one `.html` that asks a human which; `vote --record <file>` checks the answer into `votes.jsonl` |
 | `pose --images <dir> --frame <png>` | reads part placements **out of** a picture |

package/docs/AUTHORING.md

@@ -247,8 +247,11 @@ What the flags mean:
 | `--geometry` | `render` only: also write a `geometry.json` into each frame directory — per frame, every bone's world transform and every slot's region or mesh vertices in world units after skinning, plus each attachment's rest geometry and topology, on the frames' own grid and viewport. Every other file is byte for byte what the render writes without it. Refused with `--slot`/`--hide`, because the geometry is the whole pose whatever is drawn — **§8.2** |
 | `--folds` | `render` only: also read which mesh triangles **fold** — turn over against the setup pose — and write a `folds.json` into each frame directory: per mesh, the frames with a folded triangle, the worst frame's count and which triangles, by index, with their world box and setup-pose area. One printed line per mesh that folds; a triangle degenerate or inverted at the setup pose is said once and never counted. A reading, not a gate. Every other file is byte for byte what the render writes without it, `geometry.json` included. Refused with `--slot`/`--hide`, for `--geometry`'s reason — **§8.3** |
 | `--pass-through` | `render` only: `<cloth>:<leg>[,<cloth>:<leg>…]` — also read where each cloth slot **passes through** its leg slot, and write a `pass-through.json` into each frame directory: the frames where the cloth draws over the leg and its drawn side lies inside the leg's drawn outline by 2 steps or more, each with the depth and the world box. One printed line per pair, and one per frame that passes through. The first name must draw over the second for a frame to be read — the declared order, never both. A reading, not a gate. Every other file is byte for byte what the render writes without it. An unknown slot, a slot that draws on no frame, a malformed list and a pair named twice are refused by name, and so is `--slot`/`--hide` beside it — **§8.5** |
+| `--angle-range` | `render` only: `<bone>:<relative-to>[,<bone>:<relative-to>…]` — also read, for each pair, the range of the first bone's world direction against the second's over each set's frames, and print one line per pair per set: the lowest and the highest angle in degrees, the span, and the frame each was first reached on. A reading, not a gate: nothing is written, and every file is the bytes the render writes without it. An unknown bone (with the bones there are), a bone against itself, a malformed list and a pair named twice are refused by name, and so is a bone with no direction on a frame — **§8.7** |
+| `--viewport`, `--viewport-from` | `render`: give the box every frame is drawn over instead of fitting one — `--viewport <x,y,w,h>`, a world box, y up, `x,y` its bottom-left corner, at the pixel size `--max` gives; or `--viewport-from <frames.json \| dir>`, an earlier render's grid exactly (its box, scale and pixel size), so two candidates land on one pixel grid. A slot that reaches past the box is named by set and frame, and `frames.json` records the box was given under `pinned`. The two together are refused, and so is `--max` beside `--viewport-from` — **§8.6**. On `check` and `inspect`, `--viewport` pins the box those commands draw over instead |
 | `--environment` | `render` only: the wind and gravity the rig's physics constraints answer, in time — a schedule file `{ "spec": "rigc-environment/1", "steps": [{ "t": <seconds>, "wind"?: [x, y], "gravity"?: [x, y] }, …] }`, x right and **y down**, each step holding from its `t` until the next. Both posers set each change before the step that poses the first frame at or after its `t`, and draw the same frames, `geometry.json` and `frames.json` (`RE01`). `frames.json` and `geometry.json` record it, and the framing reads it. A malformed file, a half no physics constraint answers, two steps on one frame and a step after every set's last frame are refused by name, exit 2, nothing written; `check` refuses a frame set drawn under one as a reference. Without it every file is the bytes it always was. An inspection of the response the rig states, not a scene — **§8.4** |
 | `--poser` | `render`, `inspect`, `check` and `explain`: `core` or `spine` — on `render`, `inspect` and `check`, which implementation poses the frames (on `check`, the candidate's; the reference side is pixels either way). By default a **rigc build** (the `skeleton.model.json` `build` writes; its atlas, when one is named or present, the one beside it) is posed by rigc's own core and a **Spine export** by spine-core; the render prints a `poser` line saying which and why, and so does the check report (`check.json`'s `poser`, and the pictures' `frames.json` under `--out`; `bench --frames` prints the same line). A concave or an inverse clip is posed by the core, and so is a build rendered with no `--skin` whose default skin names a skin-required bone or constraint, or whose skins name no `default` one — posed as a fresh skeleton is, no skin's bones or constraints applied (the default skin's included) and every slot resolved through the default skin alone, so the second draws nothing and stops at `there is nothing to draw` (exit 2) under either poser (issue #1051). Where the input cannot carry the core, or the core refuses it, the render poses it through spine-core and the `poser` line says why, in one of these words: `no skeleton.model.json beside` (a Spine export) · `is not beside` (`--atlas` names an atlas outside the build's directory) · `does not describe` (a model document whose bones, slots or animations are not the skeleton's) · `is not the skeleton.json` (a `skeleton.json` whose SHA-256 is not the `spine.sha256` its model document records — edited or replaced after the build — both digests named) · `the polygon is not simple` (a clip polygon two of whose edges cross or touch, the slot named) · `depends on the runtime's previous pass` (a constraint moves a posed bone from a bone that a constraint writes into while that bone is inactive, all four named) · `--poser spine` (asked for). Any other refusal of the core's — an unreadable model document among them — reads `the core refused` and then the core's own words. `--poser core` on an input that cannot carry it is refused in the same words (exit 2, nothing written); `--poser spine` forces the runtime. A fallback whose pair the runtime cannot load — another build's `skeleton.json` or atlas beside this build's document, or an export beside an atlas it was not exported with — is not drawn: it is refused `… does not load against <the atlas>: spine-core draws this pair (<why>) and could not resolve it — "<the runtime's own words>"` — or, where the atlas names a page that is not there, `nothing at <the page>: …` — exit 2, nothing written, on `render`, `render --geometry`, `check` and `bench --frames` alike; under `--poser core` the same pair is that flag's refusal, in the words above (`RC18`; the `--candidate` row says what to change). The repository's selftest renders one probe per reason above and holds this list to the lines they print (`RC09`). Both posers draw the same frames, geometry and framing box to the bit on every corpus (issue #968), and `check` through either writes the same report, `check.json` and pictures but for the poser it names on every tree row. A rigc build the core poses reads nothing through spine-core on `render`, `render --geometry` and `check`, `--texture-from` included (and the check `bench --frames` folds in; `bench`'s own `validate`, like `build`'s, is the runtime's round trip by design): its bone tree, slot list and slot roster come off its model document; so do the order its animations and skins are listed in, its stage and its pages' `scale:` lines, on a `rigc-compiled/3` document (`editorOrder`, `stage`, `pages`) — a `/2` or `/1` document states none of those, so they come off its own `skeleton.json` and atlas, and the `poser` line then says `a rigc-compiled/2 document, which does not state the order its skins and animations are listed in or its stage, read from` and names both files; where each region sits and the page images it draws off the document's `pages` (a `rigc-compiled/1` document's off the atlas beside it, and the `poser` line then says `a rigc-compiled/1 document, which does not state where each region sits on its page: that is read from` and names the atlas); and a `--texture-from` atlas off rigc's own atlas reader. The runtime is loaded for a Spine export, `--poser spine` and a fallback the `poser` line names; a run that cannot load it there is refused naming the input and why it needs the runtime (exit 1; `RC11`, `RC13`, `RC34`). On the entry that links none of spine-core — the installed `rigc` without the runtime beside it, `bun cli_core.ts` in a clone — that refusal ends on the route to the entry that does (``the entry that links it runs them — installed, the same `rigc` once @esotericsoftware/spine-core is installed beside the package; from a source checkout, `bun cli.ts` ``); a runtime that is installed and fails to load is refused without it, since installing it is not what that reader has to do. On `explain` the flag chooses the `DEFORM` block's survey (§4.11.2): `core`, the default, reads the skeleton's structure — its deform timelines and their keys, curves and slots, the skin each is keyed under, the sliders — off the model document the compile writes and poses it through the core, and spine-core is not touched (`RC15`); `spine` parses and poses the Spine skeleton through spine-core. The two are the same survey to the byte on every corpus row (`DM12`). Where the core refuses the document, the survey falls back to spine-core and the block says so on a line of its own (`read and posed through spine-core: …`); `--poser core` there is refused instead (exit 2), and a run that cannot load the runtime where it needs it is refused naming the survey and why (exit 1), ending on the same route where the entry links none of it |
+| `--lever-fold` | `explain` only: a fraction above 0 and at most 1. Also read, on the turn lines' walk, the steps where a physics constraint that drives `scaleX` has its inertia lever divide by less than this fraction of the divisor it takes at the setup pose, and the steps where the scale factor it applies moves by more than the bone's setup scale. **No default**: without it the reading is not taken and one line says so; a value that is not such a fraction is refused by name (exit 2). A reading, not a gate — **§4.6.2** |
 | `--record` | `vote` only: a saved vote to check against its ballot and append to the ledger, instead of writing a ballot. This is the command's second mode; it takes no `--candidate` |
 | `--ballot` | `vote --record` only: the ballot the vote answers (default `ballot.html`). Its embedded manifest is what the vote is checked against, so the ballot file is the record of the question |
 | `--ledger` | `vote --record` only: the append-only JSONL the vote lands in (default `votes.jsonl`), one vote per line |
@@ -268,7 +271,8 @@ What the flags mean:
 
 `render` also takes `--fps <n>` (the rate it samples at, default 12 — the same
 protocol rate the reference frames use) and `--max <px>` (the long side of a
-frame, default 256). A clipping attachment is applied as the runtime applies it —
+frame, default 256). A set drawn at any rate but the default is written to
+`<anim>@<fps>fps/` rather than `<anim>/`, so two rates sit side by side: `--fps 24` writes `<anim>@24fps/`, and the default 12 writes `<anim>/`. A clipping attachment is applied as the runtime applies it —
 spine-core's own `SkeletonClipping`, in spine-webgl's call order, on a Spine export,
 and the core's clipper, drawing the same pixels, on a rigc build (the `--poser`
 row) — so a slot from
@@ -1117,6 +1121,8 @@ bun cli.ts render   --candidate path/to/spine --slot <name,…> | --hide <name,
 bun cli.ts render   --candidate path/to/spine --geometry   # + geometry.json per set: the numbers the frames were drawn from
 bun cli.ts render   --candidate path/to/spine --folds      # + folds.json per set: which mesh triangles turn over, and when
 bun cli.ts render   --candidate path/to/spine --pass-through skirt:leg   # + pass-through.json: where a cloth goes through a leg
+bun cli.ts render   --candidate path/to/spine --angle-range tail:hip   # how far a bone turns against another, per animation
+bun cli.ts render   --candidate path/to/spine --viewport-from render/   # an earlier render's grid: two builds, one pixel grid
 bun cli.ts preview  --candidate path/to/spine [--candidate path/to/another …] [--animation …] [--out preview.html]
 bun cli.ts vote     --candidate path/to/a --candidate path/to/b [--out ballot.html]
 bun cli.ts vote     --record vote-<id>.json [--ballot ballot.html] [--ledger votes.jsonl]
@@ -1129,8 +1135,13 @@ bun cli.ts pose     --images path/to/parts --frame poseA.png [--out pose.json]
   key with the curve kind. A rig with a `deform` timeline also gets the **`DEFORM`
   block** — per key, what that key did to the geometry: the area and stretch
   extremes, how far each vertex moved and whether the winding survived
-  (**§4.11.2**). It takes no `--profile`, it never gates, and it does not write
-  anything — so the figures are readable on a build the gate is refusing.
+  (**§4.11.2**). A rig with a physics constraint also gets the **turn lines** —
+  per constraint and animation, the largest turn its bone makes in one 1/60 s step,
+  walked from the setup pose as a player starts it (**§4.6.1**) — and, given
+  `--lever-fold <fraction>`, the **lever lines**: the steps a `scaleX` drive's
+  inertia lever divides by a collapsed length (**§4.6.2**). It takes no
+  `--profile`, it never gates, and it does not write anything — so the figures are
+  readable on a build the gate is refusing.
 
   ⭐ **It reads what `ingest` writes, given the pack** — `--atlas-in <pack.atlas>`,
   with `build`'s meaning (§0.2). That matters more here than it looks: printing the
@@ -4313,6 +4324,97 @@ physics, no skin set. `A51_BONES_LIE_ALONG_THEIR_DECLARED_AXIS` holds it (§5.2)
   declaration the profile did not read would be one nothing reads. The
   declaration is a statement to the gate and changes no emitted Spine byte.
 
+### 3.8 `stops` — a swinging chain held off a leg
+
+Spine has no collider and no angle limit — a physics constraint's `limit` is a
+velocity cap — so a skirt chain that physics swings carries its hem straight
+through the thigh it hangs in front of, on a build that passes every assertion.
+`render --pass-through` (§8.5) names those frames. A **stop** is the stand-in:
+a point on the leg, a frame that aims at it from the chain's root, the swinging
+link's rotation copied onto that frame, and the **drawn** link — the bone the
+mesh is weighted to — held by a transform clamp, so it may swing away from the
+leg and not past its rest angle toward it. One entry here is that network,
+written once:
+
+```json
+"stops": [
+  {
+    "chain": "hem",
+    "links": ["hem0", "hem1"],
+    "against": "thigh",
+    "at": [-10, -84],
+    "inward": 0,
+    "outward": 60,
+    "lower": { "inward": 0, "outward": 30 }
+  }
+]
+```
+
+| Field | What it is |
+| --- | --- |
+| `chain` | the name the point, the frame and the aim are called after |
+| `links` | the chain's swinging links, root first — each written by a constraint in the rig spec's `constraints` (its physics, or whatever swings it), each one below the first hanging from the drawn link of the one above |
+| `against` | the bone the point rides on — the leg |
+| `at` | the point, `[x, y]` in `against`'s local space, as a bone's `x`/`y` are in its parent's. Put it on the leg's outer edge at the hem's height |
+| `inward`, `outward` | the first drawn link's range, as its two ends, in degrees from its rest angle, counted **positive away from the point**. `inward: 0, outward: 60` is "never past rest toward the leg, up to 60° away"; `inward: -5` lets it turn 5° past rest toward the leg |
+| `lower` | `{ "inward", "outward" }`: each lower drawn link's range off its own rest bend, the same way round. Required when `links` has more than one, refused when it has one |
+
+Every field is required and none has a default. **Inward** is the turn from the
+first link toward the aim line — toward the point, which is on the leg — read off
+the setup pose, so a stop on the left side of a skirt and one on the right need no
+sign of their own.
+
+**What it expands to.** Before anything else reads the rig, `parseRigSpec`
+replaces the entry with bones and constraints — nothing downstream knows a stop
+was there, and the build is byte for byte the build of the same network written
+by hand (`LG01`):
+
+| Generated | Kind | What it is |
+| --- | --- | --- |
+| `<chain>_touch` | bone | the point: parent `against`, `x`/`y` from `at`. Listed after every other bone |
+| `<chain>_stop` | bone | the frame: the first link's parent, `x`, `y` and `length`, rotated at the point. Listed just before the first link |
+| `<link>_rel` | bone | a child of the frame, `length` the link's, rotated by the link's rest angle off the aim line. Just after the first link |
+| `<link>_d` | bone | the first link's **drawn** link — the same as `_rel`, just after it. For each lower link, its drawn link: a sibling with the link's own `length`, `x`, `y`, `rotation`, scale, shear and `inherit`, just after that link |
+| `<chain>_stop_aim` | ik | aims the frame at the point (`mix` 1). Just before the last constraint that writes the first link |
+| `<link>_rel_copy` | transform | gives `_rel` the first link's world rotation (`mixRotate` 1). Just after that constraint |
+| `<link>_clamp` | transform | local to local, `clamp: true`: gives the drawn link its source's turn, held to the range — the first link's from `_rel`, each lower link's from the link. Just after `_rel_copy` for the first, just after the last constraint that writes the link for a lower one |
+
+So a chain with a stop is authored with each link's parent below the drawn link
+above it — `hem1`'s control bone under `hem0_d` — and its mesh weighted to the
+`_d` bones; those names are the expansion's, and the build refuses a name it
+generates that the rig already declares. The order is the point of the
+placement: the frame is aimed before the link swings, and each swing is copied
+and clamped before anything below reads the drawn link.
+
+The frame's angles are read off the setup pose the rig spec states — its bones as
+written, no constraint applied, in the exact frame `src/transform.ts` measures in — so at rest the drawn
+link lies on the link and the clamp's bounds are its rest angle moved by
+`inward` and `outward`. A bone the angles are read through that takes its place
+`from` a manifest is refused: its numbers are not known until the compile.
+
+⚠️ **The link's swing has to be declared in the rig spec.** The motion spec's
+`physics` table (§4.6) is emitted after every rig constraint, past the copy and
+the clamp, which would then read the link before it swings; a link only that
+table drives is refused as one nothing swings, and a table entry on a stopped link
+is refused by name.
+
+Refused, each naming the stop, the field and the value (§5.1): a missing
+`chain`, `links`, `against`, `at`, `inward`, `outward` or (on a chain of two or
+more) `lower`; a `lower` on one link; a link or an `against` the rig does not
+declare; a link stopped twice; the root as a link; `at` that is not two numbers;
+`inward` above `outward`, at either range; a bound past ±180° once the rest angle
+is added (the clamp reads a local rotation in (−180, 180]); a generated bone or
+constraint name already declared, or generated by two stops; a link no rig
+constraint writes; a lower link not below the drawn link above it; an `against`
+that hangs below the chain; a point on the link's root or on its own line at
+rest; a frame parent scaled unevenly, sheared or reflected, or such a bone
+between two links; a link below a bone a later stop generates; a `from` bone in
+the angles' path.
+
+The stop decides nothing about *when* — it is what the object can do. Read it
+with `render --pass-through <cloth>:<leg>` (§8.5) on the motion that swung
+through before.
+
 ---
 
 ## 4. The motion spec, field by field
@@ -4613,7 +4715,8 @@ a delta from the constraint's own setting.
   of `mix`.** The runtime documents no range for the first three, and
   `PhysicsConstraintPose` documents `mix` as "a percentage (0+)" — so a negative
   wind is the other direction and a `mix` of `1.5` is an over-mix. Both are real
-  and both compile.
+  and both compile. An `inertia` above 1 is kept and posed by the editor too, and
+  what it does to a bone is read rather than refused — §4.6.1.
 - ⚠️ **A `mix` or `strength` key of exactly `0` is legal where a setup value of
   `0` is not**, and the difference is not an inconsistency: a setup pose says what
   the constraint IS and a key says what it is doing for a stretch. For `mix`,
@@ -5023,6 +5126,21 @@ file can be read; it is a **key** outside its bound that `build` refuses (§4.4)
 number: a non-number would reach the file as `null`, which the runtime reads as
 **zero**, so `"mass": "heavy"` would ship a constraint that never settles.
 
+🌀 **`inertia` has no upper bound, and above 1 a link can whirl** (issue #1423).
+Values above 1 are kept by the editor and posed: the Spine 4.3.26 editor imports
+2, 3 and 10 and exports them unchanged, the export posing as the build does, and
+the user guide describes the field as 0–100% with no cap stated. So `build`
+refuses none of them. What one does is a turn: a bone whose parent moves is kicked
+by `inertia` times that motion, and a 1-link rotate drive under a parent keyed
+±15° turns its bone at most 2.29–4.10° (by its `fps`), 18.78°, 35.61° and 153.38°
+in a single 1/60 s step at `inertia` 1, 2, 3 and 10 — the last half a turn between
+two frames. **`explain`
+prints that turn for every physics constraint and every animation** (§4.6.1), so a
+whirl is a figure you read rather than a frame you happen to watch. On a `scaleX`
+drive, a high `inertia` down a chain can also make a link's lever divide by a
+length near zero and throw its scale offset by hundreds in one step — `explain
+--lever-fold <fraction>` names those steps (§4.6.2).
+
 🧮 **Each value inside its bound is not enough: the step has to settle**
 (issue #1416). With `step = 1 / fps`, a constraint that drives anything must
 satisfy
@@ -5059,6 +5177,16 @@ keeps more velocity per step, yet it is what widens the bound: from 2 at `0`
 toward 4 as it nears `1`, which is itself out); the refusal prints both sides so
 you can see how far each lever has to go.
 
+📏 **What a value does to a link is read off a render, not estimated.** The bound
+above says whether a spring settles; it does not say how far it swings, which
+depends on the link's length, its parent's motion and every value together. Render
+the animation with `--angle-range <link>:<parent>` (§8.7) and the line prints how
+far the link turns against the bone it hangs from — the lowest and the highest
+angle and the frames they fall on — so a change of `inertia` or `strength` is read
+as the change of that range, before and after. On the selftest's angle probe a
+link's span widens when its `inertia` is raised from 0.2 to 0.8, and `AR02` prints
+both spans.
+
 ⚠️ **What `rigc render` draws a spring with is not a player's clock to the bit.** A
 constraint integrates whole steps of `1 / fps` and shows the rest as a lag weight, so
 a frame lasting a whole number of steps (12, 20, 30 or 60 fps against the default
@@ -5126,6 +5254,134 @@ skeleton JSON, so the parser is the only check it gets: it is
 `default` that is not a number would pass the gate and the round trip before
 becoming a `NaN` mix duration in the player.
 
+#### 4.6.1 How far a physics bone turns in one step — `explain`'s turn lines
+
+Under its `physics constraints` block, `explain` prints one line per physics
+constraint per animation: the largest turn the constraint's bone makes in one
+step of the animation's walk, the step it happens at and its time.
+
+```
+turn per step  (a reading, never a gate: each animation walked by rigc's core as a player starts it — …)
+  ..    turn     "swing": physics "p_i10_f60" (bone "l_i10_f60") turns at most 153.38° in one step — step 1 of 240 (t=0.0167s)
+  ..    turn     "sweep": physics "whip" (bone "ahoge_whip") turns at most 0.29° in one step — step 90 of 193 (t=1.5000s)
+```
+
+**The walk.** Each animation is walked as a player that starts it plays it: a
+skeleton with no skin set, at the setup pose with every physics state reset there,
+then the animation in steps of 1/60 s until its duration (one step at least) — the
+physics stepped by the step itself, as spine-core's `update(1/60)` steps it. It is
+posed by rigc's core; the repository's selftest walks the same probe through
+spine-core and holds the two to the same figure and step, to the bit (`WH01`).
+
+⚠️ **Not `render`'s walk, on purpose.** `render` resets the physics at the
+animation's first pose, so a bone the first key moves away from the setup pose is
+already there when the physics starts. A player starts from the setup pose, and the
+first step moves every keyed bone at once — which is where a high `inertia` answers
+hardest: on the probe above, inertia 10 turns 153.38° in step 1 from the setup pose,
+and walked from the animation's first pose at most 4.10°, 3.82° and 13.68° at `fps`
+15, 30 and 60. The whirl is a property of
+starting the animation, so it is read where a player starts it.
+
+**The turn** is the signed angle between the bone's world x axis at two consecutive
+steps, in (−180°, 180°], printed as its magnitude to two decimals; the largest over
+the walk, the earliest step on a tie. It is the change of the direction the bone is
+drawn along, whatever moved it — its physics, its parent, a key, a mirrored scale —
+so every physics constraint has a line, whichever components it drives. A turn near
+180° in one step is a direction a frame cannot tell from its opposite. A bone that
+never turns reads `does not turn in any of N step(s)`.
+
+**Not read, and said so:** a skin-required constraint (`not read — skin-required,
+and a skeleton with no skin set does not step it`); an animation a slider applies —
+a table posed at a value the consumer holds, not a span of time (`A39`'s reading) —
+whose line names the slider, while the animation that moves the slider's value is
+walked; a walk the core refuses to pose (`not read — the core refused to pose the
+walk — …`); a bone whose world direction stops being a number (`not read — its
+bone's world direction is not a number from step k (t=…s)`), rather than the finite
+figure it reached before (`WH04`); and a skeleton with no animation.
+
+**A reading, not a gate.** Nothing in it is refused, and `explain`'s exit is the
+same whatever it prints — `src/whirl.ts`, which computes it, is imported by
+`explain` alone (`WH02`). Whether half a turn per frame is the motion you meant is
+yours: a link that should trail its parent wants an `inertia` at or under 1, and
+the figure is how you see the difference without watching the right frame. A walk
+that ignores the parent's motion — the turn read in the parent's frame, or
+`render`'s walk — misses the figures above, and the selftest plants both to show it
+(`WH03`).
+
+#### 4.6.2 Where a `scaleX` lever divides by a collapsed length — `explain --lever-fold`
+
+A physics constraint that drives `scaleX` kicks its scale offset by its bone's
+motion **divided by a length**: `scaleOffset += (dx·cos + dy·sin) · inertia / rr`.
+`limit` caps `dx` and `dy`; nothing caps the division, and the kick is skipped
+only when `rr` is not above 0. `rr` is the bone's `length` times its world x-axis
+length (`√(a² + c²)` of the world matrix the step is handed) — and, on a drive
+that does not also rotate (`rotate` and `shearX` both 0), **less `scaleLag ·
+shown`**, the last integration's move of the offset weighted by how much of it was
+shown. When that divisor collapses, the next kick is enormous, children inherit
+the scale, and the lengths multiply down a chain: [measured] on a 5-link chain
+driven on `scaleX` alone at `inertia` 10, `fps` 15 and 0.95 of the step bound,
+link 3's offset jumps 24.3 → 349 in one step and the tip passes 10⁶ units with
+every assertion green (issue #1422). Whether a chain gets there depends on how far
+an offset swings, which no number in the spec states — so `build` refuses nothing
+for it, and `explain` reads it when you ask:
+
+```
+lever under 0.5× setup  (a reading, never a gate: the turn lines' walk; …)
+  ..    lever    "quiet": physics "p_v_1" (bone "v_1") reads nothing in 240 step(s): no lever divides by less than 0.5× its setup divisor 50.00, and no step moves its scale factor by its setup scale
+  ..    lever    "quiet": physics "p_v_3" (bone "v_3") lever divides by less than 0.5× its setup divisor 50.00 on 8 of 240 step(s) [8, 17, 28, 56, 83-84, 92, 99] — least 2.940 (0.05880× setup; world x axis 0.5636× setup) at step 17 (t=0.2833s)
+  ..    lever    "quiet": physics "p_v_3" (bone "v_3") scale factor moves by more than its setup scale in one step on 73 of 240 step(s) [5-13, …] — largest 1268, scale offset -616.8 → 651.4 at step 25 (t=0.4167s)
+```
+
+**The walk** is the turn lines' (§4.6.1): from the setup pose with every physics
+state reset, in steps of 1/60 s to the duration, posed by rigc's core. The
+repository's selftest walks the same probe through spine-core and holds every
+step and figure to the core's, to the bit (`LV01`).
+
+**What it names**, per animation, for every constraint that drives `scaleX`:
+
+- **a collapsed division** — a step where the lever divides (`rr` above 0) by less
+  than `--lever-fold` × the divisor it takes at the setup pose (`length` × the
+  world x-axis length there). A lever divides on a step where the constraint is
+  not pending (the first update after a reset integrates nothing), its `mix` is
+  not 0 and its `inertia` is not 0. The line gives the steps as runs and the
+  least divisor, as that fraction, with the world x-axis length the lever read on
+  that step beside it.
+- **a jump** — a step where the scale factor the constraint applies, `1 +
+  scaleOffset · mix · scaleX`, moves by more than 1: by more than the bone's own
+  setup scale, since the factor is 1 at the setup pose and multiplies whatever
+  scale the bone already has. The line gives the steps and the largest move,
+  with the offset on either side of it.
+
+⚠️ **The divisor, not the world length alone.** On the step above, link 3's lever
+reads its world x axis at 0.5636 of its setup length — not near zero — and divides
+by 0.0588 of its setup divisor, because `scaleLag · shown` (25.24) takes almost all
+of `length · |world x axis|` (28.18) away. A reading of the world length alone misses
+that step at every fraction below 0.5636, and names steps where the kick is skipped
+altogether: the same link reads 0.481 of its setup length three steps earlier while
+its divisor is −8.47, so nothing is divided (`LV04`). On a drive that rotates too
+the two are one number. A reading of the parent's **local** scale sees none of it —
+physics moves the world matrix, never a local transform (`LV03`).
+
+**The fraction is yours, stated.** `--lever-fold` has no default: what counts as
+collapsed is not a number the spec carries. Without the flag `explain` prints one
+line saying the reading was not taken (`` not read — `--lever-fold <fraction>` … ``);
+a value that is not a number above 0 and at most 1 is refused by name, exit 2
+(`LV05`).
+
+**Not read, and said so:** constraints that drive no `scaleX`, on one line —
+no lever of theirs divides by a length, which is what the gallery's and the
+public fixtures' physics read (`LV02`); and, per constraint, the cases §4.6.1
+lists — skin-required, an animation a slider applies, a walk the core refuses — plus
+a divisor, world length or offset that stops being a number (`not read — … is not
+a number from step k`), rather than the steps it named before.
+
+**A reading, not a gate.** Nothing in it is refused, `explain`'s exit is the same
+whatever it prints, and the module that computes it is the turn lines' own
+(`src/whirl.ts`, imported by `explain` alone — `WH02`). A chain that collapses is
+yours to judge: an `inertia` at or under 1, a stiffness further from the bound, or
+`rotate` driven alongside (which drops the lag term from the divisor) are the
+levers you hold.
+
 ### 4.7 `drawOrder` — reordering the slots over time
 
 The one timeline that names no target, so it sits on the animation rather than in
@@ -5627,7 +5883,9 @@ convention:
 an edit of part of it, so it starts at deform index 0 and runs to the end — which
 is why `fromVertex` and `offset` are refused beside it. A model applied to part
 of a run leaves a **step at the run's edge**. If you want a
-partial run, write it.
+partial run, write it — or, to move part of the attachment by the model, give the
+key an `envelope` (below), which weighs every vertex rather than cutting the run,
+so there is no edge to step at.
 
 **There is no `parallax` kind, and the reason is worth stating.** A pure depth
 slide — `d = z · offset`, no angle — is this form with a term dropped: subtract
@@ -5691,7 +5949,10 @@ about where they came from. A `yaw` past its fold angle folds the mesh and A39
 says so — the construct removes the transcription, not the arithmetic's
 consequences. One kind is the exception and it is an exception with a proof:
 `affine` refuses a determinant at or below zero, and above zero a positive
-determinant means no triangle **can** reverse.
+determinant means no triangle **can** reverse. ⚠️ The proof is about the
+un-enveloped key: under an `envelope` each vertex takes its own share of the map,
+so the keyed map is no longer affine, the `derived` line says so, and A39
+measures it like any other.
 
 **A model that evaluates to nothing is refused too.** Every parameter can be
 individually legal and the model still come out as a **run of zeros** — a `wave`
@@ -5758,6 +6019,83 @@ the same turn, where the missing number is a depth rather than a vertex, take a
 per-key `DEFORM` report block quotes the model above rather than re-evaluating it
 — **§4.11.2**.
 
+##### `envelope` — which vertices the model moves, and how much
+
+A transform moves **every** vertex of the attachment, so a `wave` on a skirt
+ripples the waistband too and "only the hem ripples" cannot be written as the
+model alone. An `envelope` on the transform gives each vertex a **weight in
+[0, 1]**, and each vertex's offset is multiplied by its weight before it is
+written. Every kind takes one:
+
+```json
+{ "t": 1, "transform": {
+    "kind": "wave", "amplitude": 6, "wavelength": 64, "phase": 90, "along": "x", "axis": "y",
+    "envelope": { "axis": [0, 1], "from": 224, "to": 120, "shape": "smooth" } } }
+```
+
+⚠️ **These numbers are the public skirt probe's** (`fixtures/public.ts`,
+`skirtFixture`): its waist row sits at world `y` 224 and its hem rows at 120
+and below. On your part, `from` and `to` are positions you read off the part —
+`explain` prints where its vertices sit — and nothing here is a default.
+
+It comes in two forms, and a key states exactly one:
+
+| Field | Meaning |
+| --- | --- |
+| `axis` | **ramp, required.** The unit direction a vertex's position is read along, `[x, y]` — `[0, 1]` reads `y`, `[1, 0]` reads `x`, a diagonal reads across both. Its length is 1 on the model's 1e-6 grid |
+| `from` | **ramp, required.** The position along `axis` where the weight is **0** — and before which it stays 0 |
+| `to` | **ramp, required.** The position along `axis` where the weight is **1** — and past which it stays 1. Differs from `from`; either may be the larger |
+| `shape` | **ramp, required.** How the weight rises between them: `"linear"` is `u`, `"smooth"` is `u²(3 − 2u)`, flat at both ends — the two curves `influence.shape` names (§3.4.7), read rising rather than falling |
+| `mask` | **mask, required, and alone.** The file name of this attachment's own `soft` mask (§3.4.6). The weight is the level that mask was sampled to at the vertex when the mesh was built, black 0 and white 1 — the same number the soft region's binding carries |
+
+For a ramp, `u = clamp((axis · position − from) / (to − from), 0, 1)` per
+vertex, and the weight is the shape at `u`. Weights are quantised onto the
+model's 1e-6 grid and each emitted offset is `round(w × the un-enveloped
+offset)` on the same grid — so a weight of 0 writes exactly 0, a weight of 1
+writes exactly the offset the key writes with no envelope, and a reader holding
+the weights `explain` prints reproduces every number.
+
+⭐ **The ramp is read in the space the model is.** The weight multiplies the
+offset the model produced at a vertex, so the position that picks it is the
+position the model read: the attachment's own space while every vertex is one
+pair of one bone, and setup **world** past that (above). `from` and `to`
+therefore change units exactly when `radius`, `about` and a bend's `from` do,
+and the `derived` line that says so applies to them too. A skirt weighted to
+its chains is on the world path, which is why the example's positions are world
+`y`.
+
+🔒 **It is baked, not played.** The weighted offsets are the deform table the
+file carries, so the runtime is asked to do nothing it did not already do:
+rigc's core and spine-core pose an enveloped key identically, and the gate
+judges it like any table — A35 the run, A39 the winding, `render --folds` the
+posed triangles. An envelope that rises over too short a distance can fold a
+triangle the model alone could not (a wave that reads `x` and displaces `y` is
+a shear and folds nothing), and A39 names it.
+
+What is refused, each by the field and the value:
+
+| You wrote | You get |
+| --- | --- |
+| `from` equal to `to` | `transform fields "envelope.from" and "envelope.to" are both 224; they are where the weight is 0 and where it is 1, so they differ` |
+| an `axis` that is not a unit direction | `transform field "envelope.axis" is [0, 2], whose length is 2; it is a unit direction, length 1 on the model's 1e-6 grid` |
+| a `shape` outside the two | `transform field "envelope.shape" is "cubic"; it is one of "linear", "smooth"` |
+| a ramp field missing | `transform field "envelope.from" is undefined; it is a finite number the spec has to state` |
+| a `mask` on an attachment with no `soft` block | `transform field "envelope.mask" is "skirt_soft.png", and this attachment has no "soft" block, so the mask names nothing` |
+| a `mask` naming another file than the attachment's soft mask | `… and this attachment's soft mask is "blob_soft.png", so the mask names nothing it was built with` |
+| a `mask` beside a ramp field | `transform field "envelope" states "mask" … beside the ramp's "axis", "from", "to", "shape". They are two answers to one question` |
+| an empty `envelope` | `transform field "envelope" is {}, which states no weight at all` |
+| a weight outside [0, 1] | `transform field "envelope.shape" weighs vertex 2 at 1.2; an envelope's weight is in [0, 1]` — unreachable from a spec (the ramp clamps, both shapes map [0, 1] onto itself, a mask is a sampled level), and kept because a weight past 1 would push a vertex past the model in silence |
+| a ramp that weighs every moving vertex 0 | `… and envelope axis=[0, 1] from=-16 to=-152 shape=smooth weighs every one of them 0 — so the key states a deformation and emits the identity`, with where the vertices sit along the axis |
+| a ramp the part never enters, so nothing changes | `… it changes no emitted number: every vertex the model moves takes weight 1` — a field a reader would credit for an output it did not shape |
+
+The last two are the "evaluates to nothing" rule above, one step on, and a key
+whose model states the identity (`amplitude: 0`, …) is exempt from both — the
+envelope multiplies the zeros it already said.
+
+`explain` prints the envelope under the model — its fields, its formula, how
+many vertices took 0 and 1, and every vertex's weight — and the `DEFORM` block's
+model line carries it as `envelope(…)` after the model's parameters.
+
 ---
 
 #### 4.11.2 The `DEFORM` block — what a key does to the geometry
@@ -6509,6 +6847,22 @@ or the key's position in its own track. These are the frequent ones, verbatim:
 | `ik constraint "X" names 3 bones ("a", "b", "c"); the solver applies one or two, so a constraint over 3 moves nothing — name one bone, or two where the second is the first's child` | §3.5 — the constraint's `bones` is what changes: one bone, or a parent and its child. The runtime's ik update has a case for one bone and one for two, so a longer list loads, sits in the update cache and applies nothing — [measured] every bone posed exactly where the same rig with no constraint has it. A longer chain is several constraints, each over one bone or a pair |
 | `ik constraint "X": "B" is not a child of "A" ("C" stands between), so the two-bone solve is not of the chain drawn — name "B"'s own parent as the first bone, or make "B" a child of "A"` (also `("A" is not above it; its parent is "P")`, `("B" has no parent)`) | §3.5 — the constraint's `bones` is what changes, or `B`'s `parent`. The two-bone solve places `B` through `A`'s matrix from `B`'s own local offset, so what stands between them is left out of the triangle it solves — [measured] a bone between them at `B`'s origin left the tip up to 2.60 off the target, and one at `A`'s origin landed only while it held its identity |
 | `physics constraint "X" is declared in both the rig spec and the motion spec's physics table` | §4.6 — the rig spec declares a physics constraint's structure and the motion spec's `physics` table declares one outright; pick the file it belongs in. Per kind, like every other constraint name: an `ik` "X" in the rig spec beside a `physics` "X" here is two constraints and is not this error |
+| `stops[i] needs a "chain" …` / `stop "C" states "links" …; it is the chain's swinging links, root first …` / `stop "C" needs "against" …` / `stop "C" states no "at" …` / `stop "C" states no "inward"` (or `"outward"`, `"lower.inward"`, `"lower.outward"`) / `stop "C" states no "lower" …` | §3.8 — state the field; a stop has no defaults. `at` is `[x, y]` in `against`'s local space, and `lower` is required on a chain of two or more links |
+| `stop "C" states "at" [V]; it is [x, y], two numbers in "A"'s local space` | §3.8 — two numbers |
+| `stop "C" states "lower" and its chain has one link …` | §3.8 — remove `lower`, or list the links below the first |
+| `stop "C": links[k] names bone "X", which this rig does not declare` / `stop "C": "against" names bone "X", which this rig does not declare` | §3.8 — fix the name; both resolve against the rig spec's own `bones` |
+| `stop "C": links[k] "X" is already stopped by stop "D" …` | §3.8 — a link has one drawn link and one clamp; stop it once |
+| `stop "C": links[k] "X" is the skeleton's root …` / `… takes its place from the manifest ("from") …` | §3.8 — a link hangs from a parent, and its drawn link copies its stated `x`, `y` and `rotation`; state them. The same `from` refusal names a bone the stop's angles are read through |
+| `stop "C": "inward" N is greater than "outward" M …` (also `"lower.inward"`) | §3.8 — the range is `[inward, outward]`, counted positive away from the point; `inward` is the lower end |
+| `stop "C": "outward" N puts "L_d"'s bound at B°, past ±180° …` (also `"inward"`, `"lower.…"`) | §3.8 — the clamp reads a local rotation in (−180, 180]; the bound is the drawn link's rest angle moved by the range, so narrow the range |
+| `stop "C" generates bone "X", which this rig already declares …` / `… generates ik constraint "X" …` / `… generates transform constraint "X" …` / `… which stop "D" generates too …` | §3.8 — the names are the expansion's (`<chain>_touch`, `<chain>_stop`, `<chain>_stop_aim`, `<link>_rel`, `<link>_rel_copy`, `<link>_d`, `<link>_clamp`); rename the bone or constraint, or the `chain` |
+| `stop "C": links[k] "X" is written by no constraint of this rig spec …` | §3.8 — declare what swings the link (a physics constraint, usually) in the rig spec's `constraints`; the motion spec's `physics` table is emitted after the clamp |
+| `stop "C": links[k] "X" does not hang below "P_d", the drawn link of links[k-1] …` | §3.8 — parent the link (or its control bone) below the drawn link above it, so the clamp that holds that one carries it |
+| `stop "C": "against" "A" hangs below "B", which the stop moves …` | §3.8 — name the leg; a point on the chain would swing with it |
+| `stop "C": "at" [x, y] on "A" sits on the root of "L" …` / `… lies on "L"'s own line at the setup pose …` | §3.8 — move the point onto the leg's edge beside the hem; on the link's line it has no side to call inward |
+| `stop "C": "P", where the frame hangs, is scaled unevenly, sheared or reflected …` / `stop "C": "B", between links[k] "L" and the drawn link above it, is scaled unevenly, sheared, reflected or inherits other than normally …` | §3.8 — the stop's angles add only through a rotation and an even scale; take the scale off that bone |
+| `stop "C": "L" hangs below "B", which a later stop generates …` | §3.8 — list that stop first |
+| `physics constraint "X" drives "L", a link stop "C" holds, from the physics table …` | §3.8, §4.6 — declare the constraint in the rig spec's `constraints`, where the stop's copy and clamp are placed after it |
 | `skin "S" lists "X" under "transform", but the rig declares it as a "ik" constraint — a skin looks its constraints up by name AND type, so this one is a miss and the loader throws` | §3.4.8 — move the name to the list named after the constraint's own kind. A name the rig declares under **no** kind is the other miss and says so (`skin "S" activates ik constraint "X", which this rig does not declare`). ⚠️ A name listed by **several skins** is not an error: the lists are per-skin sets and a bone or constraint in two of them is active under either |
 | `slot "X" names bone "Y", which this rig does not declare` | add the bone, or fix the slot's `bone` |
 | `no setup pose for slot "X": give the motion spec a \`setup\` entry or the rig slot an \`attachment\`` | R3 — pick one file and declare it there. A slot **nothing** fills is exempt: its setup pose can only be "show nothing" and is not asked for |
@@ -6604,6 +6958,8 @@ or the key's position in its own track. These are the frequent ones, verbatim:
 | `deform … (t=…): transform yaw has radius R, and vertex v sits at x=… past it` | §4.11.1 — the cylinder has no surface there; raise the radius to where the part sits |
 | `deform … (t=…): transform affine has scale […], whose determinant is …` | §4.11.1 — at or below zero the map reverses every triangle |
 | `deform … (t=…): transform <kind> states …, and every one of this attachment's N vertices evaluates to an offset of 0` | §4.11.1 — the parameters state a deformation and the geometry sampled it to nothing; the message names the measured cause. A key that means the setup pose states the identity in its parameters, or carries no run |
+| `deform … (t=…): transform field "envelope.<field>" is …` | §4.11.1 `envelope` — the field the message names: `from` and `to` differ, `axis` is a unit direction, `shape` is `"linear"` or `"smooth"`, a `mask` is this attachment's own soft mask, and a weight is in [0, 1] |
+| `deform … (t=…): transform <kind> states … and envelope … weighs every one of them 0` / `… it changes no emitted number` | §4.11.1 `envelope` — the ramp does not cross the part; the message says where its vertices sit along the axis |
 | `vertexCount is N, which is not a multiple of 3` | §3.4 — a path's vertices are knots and handles read in groups of three: `3(K + 1)` open, `3K` closed |
 | `vertexCount is N and an open path needs at least 6` | §3.4 — an open path drops its first and last point, so it needs six for one curve |
 | `"lengths" has N entry(ies) where the parser sizes M — vertexCount / 3 on …` | §3.4 — `vertexCount / 3` entries on either shape; give the array the source states, or leave the field out and rigc measures it |
@@ -7459,6 +7815,8 @@ exports names the frame that moved. Two exports of one skeleton are byte-identic
 | `setup` | the setup pose, sampled the way a skeleton with no animation is rendered: `{ bones, attachments }` |
 | `frames` | one per PNG, in index order: `{ index, time, bones, attachments }`. `index` is the number in `f####.png`, `time` is `index / fps` in seconds |
 
+**The keys, as the file writes them.** Top level: `spec`, `coordinates`, `animation`, `skin`?, `environment`?, `fps`, `viewport`, `bones`, `slots`, `rest`, `setup`, `frames` — a `?` marks a key written only when it is set. Each frame: `index`, `time`, `bones`, `attachments`; and `setup`: `bones`, `attachments`. What a frame draws is under `attachments`: the top-level `slots` is the table of the bone each slot hangs from, not a frame's pieces.
+
 Inside `setup` and each frame:
 
 - `bones` — every bone in declaration order: `{ name, a, b, c, d, worldX, worldY }`,
@@ -7804,6 +8162,88 @@ passes through is a defect, and what to do about it — a leg stop, a wider hem,
 different swing — is yours. To look at one, draw it:
 `inspect --show art --animation <name> --at <time> --zoom slot:<cloth>`.
 
+### 8.6 One box for two renders — `render --viewport`, `render --viewport-from`
+
+`render` fits its box to the rig it draws, so two builds of one rig — before and
+after a change — usually land on two grids of different sizes, and their pixels
+cannot be compared without a transform written by hand. Give the box instead:
+
+```bash
+bun cli.ts render --candidate before/ --out render-before
+bun cli.ts render --candidate after/ --viewport-from render-before --out render-after
+bun cli.ts render --candidate after/ --viewport -40,70,90,140 --max 256 --out render-box
+```
+
+- `--viewport-from <frames.json | dir>` reads an earlier render's `frames.json` (the
+  file, or the directory `render --out` wrote it into) and draws on **its grid
+  exactly**: the same world box, the same scale and the same pixel size, so a world
+  point lands on the same pixel in both renders and the two frame sets overlay.
+  `--max` is refused beside it, because the file already fixes the pixel size.
+- `--viewport <x,y,w,h>` draws over a world box you state, y up, `x,y` its
+  bottom-left corner — the numbers `frames.json`'s `viewport` records, and the
+  spelling `check --viewport` and `inspect --viewport` take. Its pixel size follows
+  from `--max` as a fitted box's does.
+
+The render prints a `viewport given by …` line with the box and the box it would
+have fitted, so you can see what the given one leaves out.
+
+⚠️ **Nothing is cropped in silence.** Each frame is the box: what lies past it is
+not drawn. A fitted box is measured over every pose the rig strikes at 60 fps and padded; a given one is not,
+so every slot whose drawn geometry reaches past it — a piece's posed vertex outside
+the box — is named, once per set: the frames, the farthest it reaches in world
+units, and the sides (`left`, `right`, `bottom`, `top`, y up). `frames.json` records
+the box was given and those slots under `pinned`: `{ "by": "--viewport" |
+"--viewport-from", "outside": [{ "set", "slot", "frames" }] }` — written only when a
+box was given, so a fitted render is the bytes it always was.
+
+**Refused, nothing written:** the two flags together; `--max` beside
+`--viewport-from`; a `--viewport` that is not four numbers or whose width or height
+is not above 0; a `--viewport-from` path with no `frames.json`, a file whose `spec`
+is not `rigc-frames/1`, and a `viewport` field that is not a finite number (`x`,
+`y`), a number above 0 (`width`, `height`, `scale`) or a whole number of pixels above
+0 (`pixelWidth`, `pixelHeight`) — each named with the value found. The framing still
+runs under a given box, so a pose with nothing to draw, one that is not finite and
+one with no extent are refused as they are without one.
+
+### 8.7 How far a bone turns against another — `render --angle-range`
+
+A physics value has no scale you can see from the spec: what `inertia` 0.5 does to
+a link depends on its length, on how its parent moves and on every other value
+together. The answer is a number the pose already holds — how far the link turned
+against the bone it hangs from — and `--angle-range` reads it off the frames
+`render` draws:
+
+```bash
+bun cli.ts render --candidate path/to/spine --angle-range tail:hip,sleeve:arm [--animation …] [--fps 12]
+```
+
+One line per pair per set, beside the set's line: the lowest and the highest angle
+in degrees, the span, the frames read and the rate, and the frame each extreme was
+first reached on. Nothing is written, and every file is the bytes the render writes
+without the flag.
+
+**What the angle is.** At each frame, the first bone's world direction less the
+second's — each the direction its x axis points, degrees counter-clockwise, y up,
+as the poser that drew the frame poses it (`rotationX`). The first frame's
+difference is taken into (−180°, 180°], and each later frame's is moved by whole
+turns to within half a turn of the frame before, so a link that sweeps 200° reads
+200°, not −160°. ⚠️ It is the angle **drawn**, not the bone's local `rotation`: the
+two agree for a child of the other bone when the parent carries no non-uniform scale
+or shear.
+
+**Where it is read.** On the frames' own grid — `--fps`, from t = 0 to each set's
+last frame — so a peak between two frames is not read; raise `--fps` to read finer.
+It reads bones, which `--slot`/`--hide` do not change, so it takes them.
+
+**Refused, nothing written:** a bone the skeleton does not have (with the bones it
+does), an entry without exactly one `:`, an empty entry or name, a bone against
+itself (0° on every frame), a pair named twice, and a bone with no direction on a
+frame — a world matrix of zero, which a bone the posed skin leaves unposed has, or
+one scaled to nothing — named with the frame.
+
+**A reading, not a gate.** Whether a range is too wide is yours; §4.6 says which
+values move it.
+
 ---
 
 ## 9. Checking against the frames — `rigc check`

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "spine-rigc",
-  "version": "2.55.0",
+  "version": "2.56.0",
   "description": "AI-authored 2D rigging and animation, verified before it is written. rigc compiles a rig spec into skeleton data in the Spine 4.3 format, gates it with named assertions held to a spine-core round trip's verdicts, and emits nothing that fails. Ships as an agent skill.",
   "type": "module",
   "bin": {