Harden Posecode motion for launch

.changeset/heroic-owls-land.md

@@ -0,0 +1,7 @@
+---
+"posecode-parser": patch
+"posecode-render": patch
+"posecode-mcp": patch
+---
+
+Strengthen Posecode motion authoring and playback with strict contact validation, grounded multi-contact solving, continuous sparse transitions, and more lifelike canonical movement guidance.

README.md

@@ -426,7 +426,7 @@ It explains biomechanical problems in natural language and proposes specific rev
 | Package | Purpose |
 | --- | --- |
 | [`posecode-language`](packages/posecode-language) | Language definitions and editor support |
-| [`posecode-parser`](packages/posecode-parser) | Converts `.posecode` text into validated, ROM-constrained intermediate representation |
+| [`posecode-parser`](packages/posecode-parser) | Converts `.posecode` text into a validated, range-constrained intermediate representation |
 | [`posecode-render`](packages/posecode-render) | Renders animated figures with Three.js, forward kinematics, and IK |
 | [`posecode-share`](packages/posecode-share) | Encodes Posecode documents into URL-safe share tokens |
 | [`posecode-mcp`](packages/posecode-mcp) | Exposes Posecode capabilities to AI agents through MCP |
@@ -482,7 +482,7 @@ Posecode currently focuses on:
 The following are outside the current scope:
 
 - two-person or partner motion,
-- collision detection,
+- comprehensive collision detection and rigid-body dynamics,
 - detailed object physics,
 - advanced equipment simulation,
 - multi-joint finger animation,

ROADMAP.md

@@ -1,18 +1,19 @@
 # Posecode: domains & roadmap
 
-Posecode's vision is broad: **describe any single-person movement as text and render it
-safely in 3D.** The *protocol* is general; the *current renderer* is deliberately
-scoped (one figure, forward kinematics, ground-locked hands/feet, no props). This
+Posecode's vision is broad: **describe single-person movement as text and render it
+as inspectable 3D motion.** The *protocol* is general; the *current renderer* is deliberately
+scoped to one simplified figure with forward kinematics, constrained contacts,
+starter props, and no rigid-body dynamics. This
 doc maps the domains Posecode can serve, what already works, and what each remaining
 domain needs, so contributions land where they unlock the most.
 
 ## Where Posecode fits
 
 | Domain | Example uses | Status |
 | --- | --- | --- |
-| **Physiotherapy / rehab** | Range-of-motion demos, home-exercise programs, post-op protocols, cervical/shoulder mobility |  Strong fit today: the ROM safety clamp is a clinical feature. Many moves render now; equipment (bands, balls) is future. |
-| **Ergonomics / desk & posture** | Posture resets, seated/standing stretch breaks, "do this every hour" prompts |  Works today for standing variants; true *seated* needs a chair prop (below). |
-| **Yoga & mobility** | Standing poses (chair, side bend, twist), flows, mobility drills |  Standing poses work; floor/inversion/lying poses need lying base poses + a mat. |
+| **Physiotherapy / rehab** | Range-of-motion demos, home-exercise programs, post-op protocols, cervical/shoulder mobility |  Useful for reviewed visualization: ROM constraints and diagnostics are explicit. Clinical use still requires qualified review; bands and balls are future. |
+| **Ergonomics / desk & posture** | Posture resets, seated/standing stretch breaks, "do this every hour" prompts |  Standing and floor-seated variants work; chair-seated motion needs a dedicated base pose. |
+| **Yoga & mobility** | Standing poses (chair, side bend, twist), flows, mobility drills |  Standing and basic supine/prone/long-seated poses work; inversions and detailed mat contacts remain limited. |
 | **Movement education / anatomy** | Demonstrate joint actions ("what is shoulder abduction?"), biomechanics teaching |  Excellent fit: single-joint demos are exactly what the rig does. |
 | **Fitness / strength** | Body-weight and free-form movement coaching |  Core today (squat, curl, raise). Barbell/dumbbell/machine work needs props + grip. |
 | **Functional / elderly care** | Sit-to-stand, balance, gentle ROM, fall-prevention drills |  Partial: sit-to-stand works; reaching/balance need reach-IK + props. |

docs/market-research.md

@@ -8,18 +8,18 @@
 ## 1. The thesis
 
 Posecode is **"Mermaid for human movement."** A person describes a movement in
-words; an LLM writes a short `.posecode` document; the browser parses it, clamps it
-to a safe range of motion, and renders an animated 3D mannequin, producing a
+words; an LLM writes a short `.posecode` document; the browser validates it,
+constrains joint targets to configured ranges, and renders an animated 3D mannequin, producing a
 **shareable URL**. The loop is:
 
 > **ask an LLM for a movement → it renders → share the link.**
 
 This is structurally different from video. A `.posecode` doc is **editable text**:
 an LLM can generate it, a human can tweak one angle, a clinician can fork it, and
-it diffs in version control. Diffusion/video models hallucinate anatomy and can't
-be safely constrained; Posecode's ROM clamp is a *correctness* feature, not a filter.
-Text is also how LLMs natively "think" about structure, so authoring is reliable
-and improves as models improve.
+it diffs in version control. Posecode exposes the authored anatomy and constraints
+for inspection instead of hiding them in a generated trajectory. Range constraints
+are one correctness layer, not a certification of the complete movement. Text also
+gives LLMs a structured target that can be parsed, evaluated, and revised.
 
 ### Why it can spread
 
@@ -30,8 +30,9 @@ and improves as models improve.
 - **Agent-native.** The [`posecode-mcp`](../packages/posecode-mcp) server lets Claude/
   ChatGPT author, validate, and return a render link *inside the chat*: the
   movement appears where the user already is.
-- **Safe to trust.** Clinical ROM limits mean shared links can't depict unsafe
-  joint angles: important for the health-adjacent buyers below.
+- **Inspectable constraints.** Configured ROM limits prevent impossible authored
+  joint rotations, while contact residuals and movement checks expose failures.
+  Health-adjacent uses still require qualified review.
 
 ## 2. How we scored the domains
 
@@ -67,14 +68,14 @@ on the roadmap.
 
 - **Customer:** physios, chiros, athletic trainers, and their patients running
   home-exercise programs; post-op ROM protocols.
-- **Aha use case:** a clinician types a prescription and hands the patient a link
-  that *shows* the exercise, with a built-in safe-range guarantee.
+- **Aha use case:** a clinician authors or reviews a movement and hands the patient
+  a link that *shows* it, with explicit joint targets and validation diagnostics.
 - **Viral loop:** clinician → patient link sharing is high-frequency and trusted;
   patients forward to family; clinics build reusable libraries.
 - **Ships:** `heel-raises`, `standing-hamstring-curl`, `hip-abduction`,
   `good-morning` (back-health hinge), plus existing `neck-rotation`,
   `shoulder-stretch`.
-- **Engine-fit:** ★★★★☆, the ROM clamp is a clinical feature; bands/balls and
+- **Engine-fit:** ★★★★☆, range constraints support review; bands/balls and
   lying poses are future (see roadmap).
 
 ### 3c. Desk & workplace wellness

packages/posecode-embed/test/compat.test.ts

@@ -10,7 +10,7 @@ const SPORTS_MOVE = `posecode exercise "Crossover"
     knees: flex 68
     ground-lock: feet
   step "Cross" 0.22s drive:
-    pelvis: rotate-out 14
+    turn: -14
     travel: -0.18 0
     ground-lock: feet
   repeat 2