Admission Scoring Engine
Supply-Side Logit Model — Calibrated to SFFA v. Harvard Data
Logit(admit) = (AI − 46) / 20 + Feeder_Bonus + Hook_Tips − Thresholdi + Noise
Academic Profile
Academic Index: 0 / 240
Hook Tips (ALDC)
+2.01 logits
+1.74 logits
+1.50 logits
+1.10 logits
Other Factors
+0.30 logits
Holistic noise ~ N(0, σ) represents officer subjectivity. Set to 0 for deterministic view.
Logit Score Waterfall — Component Contributions
Admission Probability by School (σ(logit))
Logit → Probability (Sigmoid)
Calibration Sources:
• Hook odds ratios from SFFA v. Harvard trial exhibits (2018)
• ALDC = Athletes, Legacies, Dean's interest, Children of faculty/staff
• Athlete OR = 7.5×, Legacy OR = 5.7×, Donor OR = 4.5×, Dean's = 3.0×
• Logit tips = ln(OR): Athlete +2.01, Legacy +1.74, Donor +1.50, Dean's +1.10
• Feeder bonus estimated from Naviance data patterns
• AI formula is a simplified composite; real AI formulas are proprietary
• Thresholds set so base rates roughly match published admit rates
• Hook odds ratios from SFFA v. Harvard trial exhibits (2018)
• ALDC = Athletes, Legacies, Dean's interest, Children of faculty/staff
• Athlete OR = 7.5×, Legacy OR = 5.7×, Donor OR = 4.5×, Dean's = 3.0×
• Logit tips = ln(OR): Athlete +2.01, Legacy +1.74, Donor +1.50, Dean's +1.10
• Feeder bonus estimated from Naviance data patterns
• AI formula is a simplified composite; real AI formulas are proprietary
• Thresholds set so base rates roughly match published admit rates