{
  "study_id": "NCT04312269",
  "agent": "baseline",
  "run": 5,
  "p_primary_hypothesis_supported": 0.12,
  "effect_direction": "null",
  "effect_size_90pct_interval": {"low": -0.45, "high": 0.8, "metric": "Cohen's d, 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 TMR, since lab TMR effects come from NREM2/SWS and REM cueing adds nothing",
  "claim_ids_used": [],
  "rationale": "About 70 participants over four arms gives ~17 per arm, so power to detect even d = 0.5 is low. The TMR meta-analytic effect is g = 0.29 for lab memory tasks. Here the outcome is a generalized functional test (WMFT), not the cued task itself, measured over weeks in chronic stroke at home or in the lab, so the expected transfer effect is smaller still. Both arms get the same MyoCI training, so within-group gains are shared and the between-group contrast is small. The design is double-blind with an objective timed outcome, which limits inflation. Chance of a significant primary contrast is low; a positive trend or a secondary or subgroup claim is more likely than a clean hit."
}