Records (VALID rows with 5y forward returns): 260

================================================================
 MATCHED-PAIR ANALYSIS
 Within fiscal year, control for fundamentals, isolate price
================================================================

  L1 LOOSE  — ROIC ±10pp, g_f ±10pp, any sector
    Pairs tested: 736
    Gap-differential predicts: hits=368/736 = 50.0% CI [46%, 54%] p=0.5147 
    MoS-differential predicts: hits=390/736 = 53.0% CI [49%, 57%] p=0.0565 *

  L2 MEDIUM — ROIC ±5pp, g_f ±5pp, any sector
    Pairs tested: 248
    Gap-differential predicts: hits=128/248 = 51.6% CI [45%, 58%] p=0.3284 
    MoS-differential predicts: hits=129/248 = 52.0% CI [46%, 58%] p=0.2839 

  L3 STRICT — ROIC ±3pp, g_f ±3pp, SAME sector
    Pairs tested: 9
    Gap-differential predicts: hits=2/9 = 22.2% CI [0%, 56%] p=0.9805 
    MoS-differential predicts: hits=4/9 = 44.4% CI [11%, 78%] p=0.7461 

================================================================
 + GAP-DIFFERENTIAL FLOOR (require |Gap_A − Gap_B| > 5pp)
 Tests whether the framework predicts when the price signal
 is loud, not just whenever fundamentals are matched.
================================================================

  L1 LOOSE  + |ΔGap| > 5pp
    Pairs tested: 438
    Gap-differential predicts: hits=208/438 = 47.5% CI [43%, 52%] p=0.8641 
    MoS-differential predicts: hits=223/438 = 50.9% CI [46%, 55%] p=0.3690 

  L2 MEDIUM + |ΔGap| > 5pp
    Pairs tested: 140
    Gap-differential predicts: hits=69/140 = 49.3% CI [41%, 58%] p=0.6000 
    MoS-differential predicts: hits=74/140 = 52.9% CI [45%, 61%] p=0.2771 

  L3 STRICT + |ΔGap| > 5pp
    Pairs tested: 2
    Gap-differential predicts: hits=0/2 = 0.0% CI [nan%, nan%] p=1.0000 
    MoS-differential predicts: hits=2/2 = 100.0% CI [nan%, nan%] p=0.2500 

  L1 LOOSE  + |ΔGap| > 10pp
    Pairs tested: 213
    Gap-differential predicts: hits=106/213 = 49.8% CI [43%, 56%] p=0.5545 
    MoS-differential predicts: hits=106/213 = 49.8% CI [43%, 56%] p=0.5545 

================================================================
 INTERPRETATION
================================================================
 - If Gap-differential accuracy > 50% with significance after
   matching, the Brina Gap adds price-discovery value beyond
   what fundamentals already predict.
 - If accuracy ≈ 50%, the Gap signal is noise once fundamentals
   are controlled for; the framework's univariate accuracy
   reflects fundamental differences, not the price signal.
 - Compare Gap accuracy vs MoS accuracy at each level — they
   should diverge if our earlier finding (MoS is at chance) is
   robust within matched pairs.
