Adaptive Pricing Discovery

Resumable AI pipeline: idea context → analyze strength → generate/approve hypotheses → compile schema → produce tiered pricing.

State Machine

Strong Route (context score = strong):
S0_NORMALIZE → S1_ANALYZE → S2A_SYNTHESIZE → S3_COMPILE → S4_IDEATION → COMPLETED

Weak/Medium Route (hypothesis-driven with user feedback):
S0 → S1 → S2B1_VALUE → U1 [user picks] → S2B2_BUYER → U2 [user picks] → S2B3_MARKET → U3 [user picks] → S3 → S4 → COMPLETED

Rejection fallback (at U1/U2):
→ ASK_MINIMUM [user provides text] → S3 → S4 → COMPLETED

Statuses: running | awaiting_user | completed | failed

Pause points: U1 (value hypotheses), U2 (buyer hypotheses), U3 (market anchors), ASK_MINIMUM

Key Files

Edge Function

Types

Components

Service Layer

Database

DB Table: ai_prompt.pm_pricing_discovery_runs

AI Features (8 steps)

Edge Function API

Start: { mode: "start", idea_id } → runs S0→S1→(route)→pause or complete

Resume: { mode: "resume", run_id, action, selections? } where action = approve|reject|skip|minimum_answers

Response: { success, run_id, status, current_step, options?, result?, pricing_schema?, usage }

PricingInputSchemaV1 (key fields)

value_definition: { core_value_proposition, value_type, value_levers, perceived_uniqueness }
tier_intents: [{ tier_name, target_segment, positioning, key_features }]
buyer_context: { buyer_type, payer_types, budget_sensitivity, purchase_type }
market_anchors: { reference_products[], psychological_alternatives[] }
stress_test_inputs: { churn_scenario, price_elasticity_guess, volume_assumption }

Common Patterns

Add new step to pipeline

1. Add step constant to edge function step enum
2. Add prompt template migration with ai_prompt.prompts + ai_prompt.prompt_versions
3. Register AI feature + quota in bible_schema.ai_features + ai_plan_quotas
4. Add step logic in edge function state machine (after which step, pause or auto-continue)
5. Update progress percentage mapping in PricingDiscoveryDialog
6. Add types for step output in types/pricing.ts

Modify prompt template

1. Find prompt by feature_key in ai_prompt.prompts
2. Create new prompt_versions row (keep old version for rollback)
3. Update ai_prompt.ai_feature_bindings if changing model/vendor
4. Test via edge function (start a new pricing run)

Modify dialog UI at a U-step

1. PricingDiscoveryDialog.tsx — handles step rendering logic
2. PricingOptionCard.tsx — individual choice card
3. Step data comes from edge function options field in response
4. User selection sent back via resumePricingDiscovery(runId, action, selections)

Debug a stuck/failed run

1. Query: SELECT * FROM ai_prompt.pm_pricing_discovery_runs WHERE id = '<run_id>'
2. Check status, current_step, error_message
3. Check step_results jsonb for partial outputs
4. Edge function logs: check pricing-discovery in Supabase logs

References

• Full schema details: See references/schema.md (pricing runs table, AI features, prompts)
• Parent skill: idea-machina for general Idea Machina development