Overview

Generation pipeline

Five steps from input to pure.json, and which layer each step reads.

Steps

StepQuestionReads
1 Meaning recognitionWhich behavior is this movement?Component.recognize
2 Signature measurementHow did it move (P1–P6)?Style axis definitions
3 RetargetWhich part of this robot does it?Template role mapping + source profile
4 StyleUser style × per-behavior ρStyle + Component.ρ
5 Guardrail checkFix MUST violations, joint and speed limitsFoundation MUST + Template limits

Input: a dog video, an image or link, or a conversation. Output: pure.json + preview + correction notices.

Formula

final = clamp( source + w × ρ × (style − source), guardrail range )

  source  signature measured from the source video (P1–P6 → operator parameters)
  style   the style coordinates the user picked (e.g. Animation)
  w       style strength (0–1, set by the user)
  ρ       the behavior's style acceptance (a Component property)
  clamp   MUST guardrails + the robot class's joint and speed limits

Example: Joy (high ρ) takes on the animation style strongly; Shake (low ρ) keeps almost its original shape. A dog shaking its body irregularly hits "no irregular tremor", is corrected to an even rhythm (pulse), and the user is told.