jianzhichun/emerge/emerge-reverse-synthesis

Emerge Reverse Synthesis

Purpose

Turn repeated raw operator events into a deterministic pipeline candidate. The daemon only packages the job and validates your result.

Inputs

Use the synthesis_job_ready job payload:

• normalized_intent: detected operator behavior
• events: raw operator events
• connector_notes and synthesis_hints
• context_hint, machine_ids, detector_signals

Rules

1. Infer the smallest reusable operation from raw operator events.
2. Parameterize operator-specific values through __args[...] only when they are likely inputs.
3. Keep stable connector constants literal.
4. Assign __result for read mode or __action for write mode.
5. Remove debug code and narration from final code.
6. Include a clear rationale describing event evidence and parameter choices.
7. Validate the candidate through icc_exec with no_replay=true; write only a pending artifact unless the operator explicitly approves activation.

Output JSON

Use this shape in your working notes and pending artifact rationale:

{
  "connector": "desktop_drafting_app",
  "mode": "write",
  "pipeline_name": "create_room_labels",
  "code": "__action = {'ok': True, 'created': []}",
  "confidence": 0.82,
  "rationale": "raw operator events repeatedly added room labels on the same layer; label text is parameterized via __args.",
  "verify_strategy": {
    "required_fields": []
  }
}

Quality Bar

The generated code is compiled once, then runs without LLM. Treat this as compile-time distillation, not runtime reasoning.