brand-gen-intent-preservation-trace

For Sage brand work in Pi, use the paste-ready prompt at docs/prompts/pi-sage-brand-gen-full-pipeline.md. Keep this link instead of copying the full prompt into skill bodies.

Brand-Gen Intent Preservation Trace

Risk addressed: MaterialPlan (pipeline_types.py:72) holds typed fields, but create_material_plan returns a dict, and assemble_generation_scratchpad reads via plan.get("aesthetic_capsule"), plan.get("selected_inspiration_sources"), plan.get("system_mechanic") etc. Many high-information fields are flattened to strings before generation: selected_mechanic_labels becomes "borrow {a} and {b}; avoid {x}" (prompt_assembly.py:670); aesthetic_capsule becomes free-form prose via render_capsule_prompt; surface_strategy_candidates is computed as a list but only the chosen selected_surface_strategy_prompt_directive string survives. Scoring (scoring/program.py) gets brand_dna: str and story_objective: str — the structured plan is gone.

Once intent is a string, downstream stages cannot reason about why something was chosen. Critique can only re-parse prose. Iteration memory captures style_anchor strings without the structured "this anchor came from variant X with score Y because Z." This caps the ceiling of the auto-feedback loop.

What this skill produces

{
  "summary": {
    "material_plan_fields_total": 28,
    "preserved_to_scratchpad": 14,
    "preserved_to_scorer": 5,
    "preserved_to_iteration_memory": 4,
    "lossy_boundaries": [
      "plan_builder → prompt_assembly (string render)",
      "scratchpad → scoring (only brand_dna/story_objective survive)",
      "any → iteration_memory (only style_anchor + score)"
    ]
  },
  "field_matrix": [
    {
      "plan_field": "aesthetic_capsule",
      "type": "dict",
      "scratchpad": "preserved_dict",
      "scorer": "lost — flattened to brand_dna prose",
      "iteration_memory": "lost",
      "string_render_site": "prompt_assembly.render_capsule_prompt:NNN",
      "consequence": "scorer cannot fault palette mismatch except via brand_dna prose; cannot suggest 'try alt capsule X'"
    },
    {
      "plan_field": "selected_mechanic_labels",
      "type": "list[str]",
      "scratchpad": "preserved_list",
      "scorer": "lost — joined as 'borrow X; avoid Y' string",
      "iteration_memory": "lost",
      "string_render_site": "prompt_assembly.py:670",
      "consequence": "iteration cannot promote a single mechanic; entire string treated atomically"
    },
    {
      "plan_field": "surface_strategy_candidates",
      "type": "list[dict]",
      "scratchpad": "lost — only selected_surface_strategy_prompt_directive (str) survives",
      "scorer": "lost",
      "iteration_memory": "lost",
      "string_render_site": "plan_builder.py:??",
      "consequence": "cannot A/B alternative strategies; rejection of selected has no structure to reuse"
    }
    /* ... per-field ... */
  ],
  "rubric_input_surface": {
    "scorer_signature_inputs": ["brand_dna", "story_objective", "rubric_axes"],
    "structured_plan_addressable": false,
    "minimum_extension": [
      "add aesthetic_capsule_id",
      "add selected_mechanic_ids: list[str]",
      "add inspiration_role_ids: list[str]",
      "add experiment_id (per brand-gen-experiment-modeling)"
    ]
  },
  "iteration_memory_surface": {
    "fields_persisted": ["version_id", "score", "style_anchor", "notes"],
    "structured_intent_persisted": false,
    "minimum_extension": ["plan_id", "selected_mechanic_ids", "aesthetic_capsule_id"]
  }
}