The Pre-Print Cohort Read
Heading into the print, the live cohort_analyze pull on GILD returned n=250 historical analogs with a tightness score of 0.60. The 5-day forward median was +0.24% with IQR [-2.13%, +2.58%] and hit-rate 52%. The 10-day median was +0.51% with IQR [-2.69%, +3.56%]. The dominant 5-day feature was sector_etf=XLF (positive direction, importance 1.34).
What Actually Printed
Print-day close: $134.06, prior close $136.30. Day-over-day change: -1.64% (-2.24). Intraday range: $133.76 – $136.50.
Cohort vs. Reality — The Verdict
The -1.64% reaction is sitting inside the cohort's IQR (-2.13% to +2.58%). The cohort's central tendency held — neither the bull nor bear feature bucket fired hard enough to push the reaction into the tails.
- n=250, 5d median +0.24%, IQR [-2.13%, +2.58%]
- Conformal bucket boundaries: p10 -4.48% / p90 +4.97%
- Top 5d feature: sector_etf=XLF (positive direction, importance 1.34)
- Actual reaction: -1.64%; verdict: central tendency hold
The Calibration Lesson
For GILD, the calibration verdict is: central tendency hold. The cohort's structured features (sector_etf=XLF dominant) provided the directional read; the realized magnitude lands relative to that read at the bucket described above. Agent systems should retrieve the post-print cohort 24-48 hours after the close — the chart shape will have stabilized, the feature attribution sharpens, and the forward 5-10 day continuation distribution becomes actionable.