brand-gen-experiment-modeling

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 Experiment Modeling

Risk addressed: brand-gen computes 2-3 ranked aesthetic variants in build_aesthetic_direction_brief (brand_gen/aesthetic_curation.py:231) — and discards them. Only the _resolved_capsule (one variant) flows into the prompt. branch_id and parent_branch_id exist in pipeline_types.py:261 but are populated to workflow_id everywhere; nothing forks. Three parallel selection systems (aesthetic_archetypes rotation, aesthetic_curation capsule scoring, surface_strategy strategy scoring, plus a design_variance 1-10 dial) merge in plan_builder.py:506-528 with no shared "Variant" type.

This skill answers: does the system support trying N aesthetic directions and converging on one, today? If not, what's the smallest abstraction that would?

What this skill produces

{
  "summary": {
    "first_class_variant_type_exists": false,
    "variants_materialized_per_run": 3,
    "variants_used_per_run": 1,
    "branch_id_populated_distinct_from_workflow": false,
    "parallel_selection_systems": 4,
    "shared_selection_signature": null
  },
  "variant_lifecycle": [
    { "stage": "create", "site": "aesthetic_curation.py:231 build_aesthetic_direction_brief", "shape": "{primary, alternates[2], difference_axes[], selection_rule}" },
    { "stage": "plan_attach", "site": "plan_builder.py:523", "shape": "embedded as plan['aesthetic_direction_brief'] dict" },
    { "stage": "prompt_consume", "site": "prompt_assembly.py:??", "shape": "prose render of primary only — alternates not referenced" },
    { "stage": "scratchpad_persist", "site": "generation_flow.py:assemble_generation_scratchpad", "shape": "?" },
    { "stage": "score_address", "site": "scoring/program.py BrandScorer", "shape": "no variant addressing — single brand_dna string" },
    { "stage": "iteration_memory_address", "site": "iteration_memory.py", "shape": "no variant addressing — flat per-version entries" }
  ],
  "selection_systems": [
    { "name": "aesthetic_archetype", "module": "aesthetic_archetypes.py", "score_fn": "pick_rotating_archetype", "output_type": "dict", "merges_at": "plan_builder.py:506-528" },
    { "name": "aesthetic_capsule", "module": "aesthetic_curation.py", "score_fn": "_capsule_score", "output_type": "dict", "merges_at": "plan_builder.py:506-528" },
    { "name": "surface_strategy", "module": "surface_strategy.py", "score_fn": "_score_strategy", "output_type": "dict", "merges_at": "plan_builder.py:506-528" },
    { "name": "design_variance_dial", "module": "card_engine.py", "score_fn": "n/a (int 1-10 from caller)", "output_type": "int", "merges_at": "card_engine.py taste_design_directives" }
  ],
  "branch_field_usage": [
    { "site": "pipeline_types.py:261 declaration", "populated_to": "n/a" },
    { "site": "pipeline_runner.py", "populated_to": "workflow_id (no actual branching)" }
  ],
  "recommendation": {
    "abstraction": "AestheticExperiment",
    "fields": ["experiment_id", "parent_experiment_id", "variants: List[Variant]", "selected_variant_id", "selection_rationale", "scoring_dimension", "decision_state"],
    "variant_fields": ["variant_id", "archetype", "capsule", "surface_strategy", "design_variance", "selection_score", "selection_reasons", "differentiating_axes"],
    "seam": "between MaterialPlan (plan-build phase output) and GenerationScratchpad (execute phase input)",
    "minimum_changes": ["promote variants[] from plan dict to typed list", "carry through generation_flow scratchpad payload", "extend BrandScorer.forward to accept variant context", "extend iteration_memory entries to reference variant_id"]
  }
}