{
  "study_id": "NCT04312269",
  "agent": "framework",
  "run": 2,
  "p_primary_hypothesis_supported": 0.17,
  "effect_direction": "favours intervention",
  "effect_size_90pct_interval": {"low": -0.35, "high": 0.85, "metric": "Cohen's d, between-group difference in WMFT change (time-based), positive = more improvement with TMR"},
  "moderator_prediction": "All-phase TMR responds more than SWS-only: the framework locates writes at sleep thresholds (N1/transitions), while deep sleep is the near-weightless drip where single cues book almost nothing.",
  "claim_ids_used": ["PM-0129", "PM-0130", "PM-0095", "PM-0096", "PM-0333", "PM-0364", "PM-0132", "PM-0226", "PM-0337", "PM-0851", "PM-0104"],
  "rationale": "Framework: reactivation re-books what attentive MyoCI training stored (PM-0095, PM-0096), and cues paired with attentive learning carry the entry (PM-0333, PM-0364), so direction favours TMR. But cues during deep sleep resemble the drip, nearly weightless (PM-0129, PM-0130); the installer's quadrant is the threshold, not SWS (PM-0132, PM-0226). The motor gain must also be verified in daytime use (PM-0337, PM-0851); both arms get identical daytime training, which dominates. With ~17 per arm, four arms, chronic stroke and a noisy time-based WMFT, a modest add-on effect (d~0.25) is unlikely to reach p<0.05. Expected: positive trend, not significant."
}