{
  "study_id": "NCT04312269",
  "agent": "framework",
  "run": 4,
  "p_primary_hypothesis_supported": 0.2,
  "effect_direction": "favours intervention",
  "effect_size_90pct_interval": {"low": -0.35, "high": 0.8, "metric": "Between-group difference in WMFT change (time-based score), Cohen's d (positive = more improvement with TMR)"},
  "moderator_prediction": "All-phase TMR at least equal to or larger than SWS-only: the framework holds that cues in deep sleep resemble the drip and book almost nothing, so extra cueing in lighter stages and near thresholds should add weight; differences between TMR arms will be small and non-significant.",
  "claim_ids_used": ["PM-0095", "PM-0096", "PM-0129", "PM-0130", "PM-0132", "PM-0333", "PM-0364", "PM-0851", "PM-1353", "PM-0104"],
  "rationale": "Reactivation re-books what attentive MyoCI training stored rather than installing new content (PM-0095, PM-0096); cues paired with attentive learning carry the entry (PM-0333, PM-0364), so a small positive direction is plausible. But sleep cues outside the thresholds are drip-like and book little (PM-0129, PM-0130, PM-0132), and the motor gain is mainly driven by daytime enactment shared by all arms (PM-1353, PM-0851), diluting the between-group contrast. With ~17 per arm, a variable time-based WMFT in moderate-severe chronic stroke, and a sham that still gets full training, significance at alpha 0.05 is unlikely; expected d around 0.2-0.3."
}