V3 QA — Scenario 12: Integration (Task 13)
=============================================

Test file: experiments/v3/tests/test_integration.py
Result: 1/1 PASSED

Full Pipeline:
  - In-memory DB with 3 markets (less, greater, between)
  - Mock forecasts, prices, accuracy, orderbook data
  - Control treatment + 1 mock experimental treatment
  - Mock LLM returning deterministic buy_yes/buy_no decisions
  - Harness runs: 3 markets × 2 treatments × 2 replicates × 5 timesteps
  - 60 expected decisions stored in treatment_decisions

Verification Points:
  1. Within-subjects: each market has decisions for ALL treatments
  2. No future peeking: settlement data not leaked into context
  3. Decisions stored in DB correctly (all 60)
  4. Scoring produces: P&L per treatment, Brier score, skip rate
  5. Statistics produces: delta profit, t-test, Cohen's d, CI
  6. Full pipeline runs end-to-end without errors

Coverage: Exercise harness → LLM → store → score → statistics path.

VERDICT: PASS
