{
  "study_id": "NCT04312269",
  "agent": "baseline",
  "run": 1,
  "p_primary_hypothesis_supported": 0.12,
  "effect_direction": "null",
  "effect_size_90pct_interval": {"low": -0.45, "high": 0.75, "metric": "Cohen's d for between-group difference in WMFT change (time-based), baseline to 6 weeks, positive = more improvement with TMR"},
  "moderator_prediction": "none; if any difference, SWS-only TMR roughly equal to all-phase, since lab TMR benefits come from NREM2/SWS and REM cueing adds nothing, but arms of ~17 cannot resolve it",
  "claim_ids_used": [],
  "rationale": "Four arms from ~70 enrolled gives ~17 per arm; detecting a between-arm difference needs d near 1.0. Lab TMR effects are g = 0.29 on the cued memory itself; here the outcome is generalisation to an untrained functional test (WMFT) in chronic stroke, which should be smaller still. Both arms receive identical MyoCI training, so the comparator is very active and within-group gains will be shared. Timed WMFT scores are skewed and noisy. Expected true effect ~0.1-0.2 d; power well under 20%. Small chance of a chance or selective-analysis positive, plus pooled-arm contrast option, puts probability around 0.12."
}