Demand-Side College Agent Logic: Academic Index & Yield Management

Source: demand-side.md

Demand-Side College Agent Logic: Academic Index & Yield Management

1. The Academic Index (AI)

1.1 What Is It?

The Academic Index is a numeric score originally developed by the Ivy League to ensure recruited athletes meet minimum academic standards. It has since become the standardized measure Ivy League admissions offices use to compare applicants' academic credentials on a single scale.

The AI is scored on a 0-240 scale (test-inclusive) or a 0-221 scale (test-exclusive / test-optional applicants).

1.2 Current Formula (Post-2020)

The Ivies updated the formula after SAT Subject Tests were discontinued and class rank reporting declined. The old formula used three components (class rank, SAT I, SAT II averages) each on an 80-point scale. The modern formula is:

AI = CGS + (SAT / 20) * 2

Where:

CGS = Converted Grade Point Score (unweighted GPA mapped to 0-80 scale)
SAT = Combined SAT score (400-1600)

For ACT submissions:

AI = CGS + ACT_Conversion * 2

ACT superscoring is NOT allowed -- a single sitting composite is used.

1.3 GPA to CGS Conversion Table

Unweighted GPA	CGS (Converted Grade Score)
4.0	80
3.9	79
3.8	78
3.7	77
3.6	73
3.5	71
3.4	68
3.3	66
3.2	64
3.1	62
3.0	60
2.9	58
2.8	56
2.7	54
2.6	52
2.5	50

Note: The exact CGS table is proprietary. The values above are reconstructed from published examples (4.0 = 80, 3.6 = 73, 3.4 = 68) with linear interpolation for gaps. The curve is steeper at the top, reflecting diminishing returns for GPA differences below ~3.5.

1.4 SAT to AI Component

SAT Score	AI Contribution (SAT/20 * 2)
1600	160
1500	150
1400	140
1300	130
1200	120
1100	110
1000	100

1.5 ACT to AI Conversion

ACT Composite	Conversion Value	AI Contribution (Value * 2)
36	80	160
35	78	156
34	76	152
33	74	148
32	72	144
31	70	140
30	68	136
29	66	132
28	64	128

1.6 AI Thresholds for Simulation

AI Range	Interpretation	Simulation Action
230-240	Elite academic profile	Auto-read by top schools; strong candidate everywhere
215-229	Very strong	Competitive at all Ivies; strong at near-Ivy
200-214	Strong	Competitive at mid-Ivy; solid at near-Ivy/selective
185-199	Good	Borderline at Ivies; competitive at selective schools
170-184	Average for selective	Unlikely at Ivies; competitive at many selective schools
< 170	Below Ivy floor	Auto-reject at most Ivies (except recruited athletes)

Ivy League minimum for athletes: Historically ~171 (varies by sport and school). Schools can admit athletes below the school's mean AI, but the team average must stay within one standard deviation of the admitted class average.

1.7 Simulation-Ready JavaScript Pseudocode

```javascript proof:W3sidHlwZSI6InByb29mQXV0aG9yZWQiLCJmcm9tIjowLCJ0byI6MTAyMSwiYXR0cnMiOnsiYnkiOiJhaTpjbGF1ZGUifX1d function calculateAcademicIndex(gpa, sat, act) { // Convert GPA to CGS (0-80 scale) // Piecewise linear approximation of the proprietary table let cgs; if (gpa >= 3.7) { // Steep region: 3.7-4.0 maps to 77-80 cgs = 77 + (gpa - 3.7) * 10; // 10 points per 0.3 GPA } else if (gpa >= 3.0) { // Middle region: 3.0-3.7 maps to 60-77 cgs = 60 + (gpa - 3.0) * (17 / 0.7); // ~24.3 points per 1.0 GPA } else { // Lower region: 2.0-3.0 maps to 40-60 cgs = 40 + (gpa - 2.0) * 20; // 20 points per 1.0 GPA } cgs = Math.min(80, Math.max(0, cgs));

// Convert test score to AI component let testComponent; if (sat) { testComponent = (sat / 20) * 2; // SAT: divide by 20, multiply by 2 } else if (act) { // ACT conversion: 36->80, 30->68, linear between const actConversion = 80 - (36 - act) * 2; testComponent = actConversion * 2; } else { // Test-optional: max AI is 221 return Math.min(221, cgs * (221 / 80)); }

return Math.round(cgs + testComponent); }

***

## 2. Yield Rate Management

### 2.1 What Is Yield Rate?

**Yield rate** = (students who enroll) / (students admitted) * 100

It measures what percentage of admitted students actually choose to attend. It is the single most important metric for enrollment management because it determines how many offers a school must extend to fill its class.

### 2.2 Real Yield Rates (Class of 2029)

| School            | Yield Rate | Tier                                    |
| ------------------------------------------------------------------------ | ----------------------------------------------------------------- | ---------------------------------------------------------------------------------------------- |
| UChicago          | ~88%       | HYPSM-adjacent (inflated by binding ED) |
| MIT               | 86.58%     | HYPSM                                   |
| Harvard           | 83.62%     | HYPSM                                   |
| Stanford          | ~82%       | HYPSM                                   |
| Princeton         | 75.37%     | HYPSM                                   |
| Brown             | 73.12%     | Ivy+                                    |
| Dartmouth         | 70.92%     | Ivy+                                    |
| Cornell           | 63.55%     | Ivy+                                    |
| Columbia          | 61.30%     | Ivy+                                    |
| Duke              | ~57%       | Near-Ivy                                |
| Northwestern      | ~58%       | Near-Ivy                                |
| Bowdoin           | 53.81%     | Top LAC                                 |
| Johns Hopkins     | 51.37%     | Near-Ivy                                |
| Boston College    | 45.10%     | Selective                               |
| Carnegie Mellon   | 46.75%     | Near-Ivy                                |
| Rice              | 42.84%     | Near-Ivy                                |
| Amherst           | 39.69%     | Top LAC                                 |
| Average US 4-year | ~30%       | Baseline                                |

**Note on UChicago:** Their ~88% yield is heavily inflated by their aggressive ED program. Most of their class commits binding early. This is an important modeling consideration.

### 2.3 The Offer Math (Knapsack Problem)

The fundamental enrollment management equation:

Offers_needed = Target_class_size / Expected_yield_rate

Examples:

| School  | Target | Yield | Offers Needed | Acceptance Rate Impact         |
| -------------------------------------------------------------- | ------------------------------------------------------------- | ------------------------------------------------------------ | -------------------------------------------------------------------- | ------------------------------------------------------------------------------------- |
| Harvard | 1,650  | 84%   | ~1,964        | Low (1,964 / 54,000 = 3.6%)    |
| Yale    | 1,550  | 68%   | ~2,279        | Low (2,279 / 52,000 = 4.4%)    |
| Cornell | 3,500  | 64%   | ~5,469        | Higher (5,469 / 68,000 = 8.0%) |
| Duke    | 1,700  | 57%   | ~2,982        | Medium (2,982 / 50,000 = 6.0%) |
| Rice    | 1,000  | 43%   | ~2,326        | Higher (2,326 / 31,000 = 7.5%) |

### 2.4 Yield Modeling by Segment

Enrollment managers build yield models that predict yield differently for student segments:

* **ED/EDII applicants**: 100% yield (binding commitment)

* **Financial aid recipients**: Yield varies by package generosity. Full-ride students yield at 80-90%+; students with loans/gaps yield lower.

* **In-state vs. out-of-state** (public schools): In-state yield is often 2-3x higher (e.g., UVA in-state ~55%, out-of-state ~30%).

* **Geographic distance**: Closer students yield higher.

* **Demonstrated interest**: Students who visited campus, interviewed, opened emails yield 15-25% higher.

* **Academic overqualification**: Students whose stats are far above the school's median yield lower (they likely have better options).

### 2.5 Yield Prediction Simulation Pseudocode

```javascript proof:W3sidHlwZSI6InByb29mQXV0aG9yZWQiLCJmcm9tIjowLCJ0byI6Njc0LCJhdHRycyI6eyJieSI6ImFpOmNsYXVkZSJ9fV0=
function predictYield(college, student) {
  let baseYield = college.historicalYieldRate;

  // Adjust for round
  if (student.round === 'ED' || student.round === 'EDII') return 1.0;

  // Academic overqualification penalty
  const aiDelta = student.academicIndex - college.medianAI;
  if (aiDelta > 20) baseYield *= 0.7;  // Much stronger student -> less likely to enroll
  if (aiDelta > 35) baseYield *= 0.5;  // Way overqualified

  // Financial aid boost
  if (student.fullRideOffered) baseYield *= 1.3;

  // Demonstrated interest
  if (college.tracksDemonstratedInterest && student.demonstratedInterest) {
    baseYield *= 1.2;
  }

  return Math.min(1.0, baseYield);
}

2.6 Over-Enrollment Disasters

Over-enrollment happens when yield exceeds predictions and more students commit than there are beds/seats:

UC Santa Cruz (2021): Had to increase dorm occupancy and convert study rooms into bedrooms after accepting too many students.
Florida A&M (2022): Enrolled 1,119 freshmen vs. 524 the prior year. Over 500 freshmen and 800+ upperclassmen waitlisted for housing.
Texas State (2021): Ran out of housing when enrollment surged back post-pandemic.
Huston-Tillotson (2022): Had to house 14% of students in dorms at a school 4 miles away.

This is why enrollment managers are conservative and use waitlists as "insurance policies."

2.7 Waitlist Math

Waitlists serve as a buffer for enrollment uncertainty:

School	Waitlisted	Accepted WL Spot	Admitted Off WL	WL Admit Rate
Harvard	~2,000	~1,500	0-50 (varies)	0-3%
Yale	~1,000	~700	0-25	0-4%
Stanford	~800	~600	0-25	0-4%
Princeton	~1,200	~900	0-100	0-11%
Average selective	varies	varies	varies	~7%

Key insight: The number admitted off the waitlist varies enormously year-to-year (sometimes zero, sometimes hundreds) depending on how accurate the yield prediction was. Overall, only ~20% of students who accept a waitlist position get admitted, and at the most selective schools it drops to ~7%.

3. Yield Protection ("Tufts Syndrome")

3.1 What Is It?

Yield protection (also called "Tufts Syndrome") is the alleged practice of rejecting or waitlisting applicants whose credentials are significantly above a school's typical admits, under the assumption they will attend a more prestigious school if admitted, thus hurting the school's yield rate.

The term originated with Tufts University in the 1990s, though Tufts has never admitted to the practice.

3.2 Which Schools Are Most Associated With It?

Schools frequently cited in yield protection discussions:

Tufts (namesake of the phenomenon)
Tulane
University of Chicago (pre-ED dominance era)
Emory
Case Western Reserve
Northeastern
Boston University
Lehigh
WashU (Washington University in St. Louis)

Pattern: Yield protection is most associated with schools in the "near-Ivy" to "selective" range that compete for students who also apply to Ivies. True top schools (HYPSM, Ivies) do not need yield protection because virtually all their admits are competitive candidates and their yield is already high.

3.3 The Evidence

For:

Some Naviance scattergrams show admit rates increasing with GPA/SAT up to a point, then decreasing at the highest levels.
Students with 1550+ SATs sometimes report rejections from schools where 1450-range students are admitted.
Schools that do not require demonstrated interest are more likely to yield-protect, since they have no other way to gauge an applicant's interest level.

Against:

No school has ever publicly admitted to yield protection.
It cannot be proven statistically -- high-stat rejections could be due to poor essays, lack of fit, or holistic review factors.
Many admissions consultants and former officers say it does not exist as a formal policy.
The phenomenon may be explained by highly selective schools simply having enough qualified applicants that many strong students will inevitably be rejected.

Academic research: Limited. The SFFA v. Harvard trial data does not show systematic yield protection at Harvard. No peer-reviewed study has conclusively demonstrated the practice.

3.4 How to Implement in Simulation

Yield protection should be a tunable parameter applied primarily to non-top-tier schools:

```javascript proof:W3sidHlwZSI6InByb29mQXV0aG9yZWQiLCJmcm9tIjowLCJ0byI6ODcwLCJhdHRycyI6eyJieSI6ImFpOmNsYXVkZSJ9fV0= function applyYieldProtection(college, student, baseScore) { // Only schools below HYPSM tier use yield protection if (college.tier === 'HYPSM') return baseScore;

const aiDelta = student.academicIndex - college.medianAI;

// If student is dramatically overqualified and hasn't shown interest if (aiDelta > 30 && !student.demonstratedInterest(college)) { // Yield protection: reduce admission probability const protectionFactor = college.yieldProtectionStrength || 0.0; // protectionFactor: 0.0 (no protection) to 0.5 (strong protection)

if (aiDelta > 45) {
  // Extremely overqualified: likely reject/waitlist
  return baseScore * (1 - protectionFactor * 0.8);
} else if (aiDelta > 30) {
  // Significantly overqualified: moderate penalty
  return baseScore * (1 - protectionFactor * 0.4);
}

}

return baseScore; }

**Recommended parameter ranges:**

| Tier      | yieldProtectionStrength        |
| ---------------------------------------------------------------- | ------------------------------------------------------------------------------------- |
| HYPSM     | 0.0 (none)                     |
| Ivy+      | 0.0 - 0.1 (minimal)            |
| Near-Ivy  | 0.1 - 0.3 (light to moderate)  |
| Selective | 0.2 - 0.5 (moderate to strong) |
| Top LAC   | 0.1 - 0.4 (varies)             |

**Trigger thresholds:**

* AI delta > 30 above school median: mild yield protection begins

* AI delta > 45 above school median: strong yield protection

* Demonstrated interest (ED/campus visit) overrides yield protection entirely

***

## 4. Holistic Review Process

### 4.1 The Harvard Rating System (1-6 Scale)

Revealed during the SFFA v. Harvard trial, Harvard rates every applicant on a 1-6 scale (1 = best) across six categories:

#### Academic Rating

| Score | Label               | Description                                                                                             |
| ------------------------------------------------------------ | -------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| 1     | Summa Potential     | Perfect or near-perfect scores/grades, uncommon creativity, national/international academic recognition |
| 2     | Magna Potential     | Excellent grades, SAT 750+/ACT 33+, possible regional recognition                                       |
| 3     | Cum Laude Potential | Very good grades, mid-600s SAT / 29-32 ACT                                                              |
| 4     | Adequate            | Adequate preparation                                                                                    |
| 5     | Marginal            | Marginal preparation                                                                                    |
| 6     | Inadequate          | Below standards                                                                                         |

#### Extracurricular Rating

| Score | Description                                                                                        |
| ------------------------------------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------------------- |
| 1     | Uncommon strength in one or more areas; national-level achievement                                 |
| 2     | Strong contribution (e.g., class president, newspaper editor), possible local/regional recognition |
| 3     | Solid participation without special distinction                                                    |
| 4-6   | Minimal to no meaningful engagement                                                                |

#### Athletic Rating

| Score | Description                                                              |
| ------------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------- |
| 1     | Uncommonly strong varsity prospect; national/international/Olympic level |
| 2     | 3-4 years varsity with leadership; state/regional recognition            |
| 3     | Active participation with local/conference achievement                   |
| 4-6   | Limited or no athletic engagement                                        |

#### Personal Rating

| Score | Description                                       |
| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------- |
| 1     | Outstanding: courage, leadership, inspires others |
| 2     | Very strong character, maturity beyond years      |
| 3     | Generally positive                                |
| 4     | Bland or somewhat negative or immature            |
| 5     | Questionable personal qualities                   |
| 6     | Worrisome personal qualities                      |

#### Recommendation & Interview Ratings

Also scored 1-6 by guidance counselors, teachers, and alumni interviewers.

#### Overall Rating

A holistic composite score (1-6) determined by admissions officers after reading all materials:

* **1:** Virtually always admitted

* **2:** Usually admitted (strong candidate)

* **3-:** Borderline; discussed in committee

* **3 or worse:** Almost always rejected

**Key finding from trial data:** Receiving a 1 in ANY single category increases admission chances to over 60%, despite Harvard's overall 3.4% acceptance rate.

### 4.2 The Reader Workflow

At most selective schools, applications go through:

1. **First read:** A regional admissions officer reviews the application and assigns scores in each category. Writes a brief summary.
2. **Second read:** A senior officer independently reviews and scores.
3. **Committee discussion:** Applications with borderline scores (typically 2-/3+) go to committee. 15-25 officers discuss and vote.
4. **Final decision:** Dean of admissions resolves ties and makes final calls on borderline cases.

The entire process takes ~15-30 minutes per application on first read.

### 4.3 Likely Letters

**What:** An unofficial early notification (before official decisions) telling a student they are "likely" to be admitted.

**Who gets them:**

* Top ~1% of the applicant pool

* Recruited athletes who have passed admissions review

* Students from underrepresented backgrounds whom the school is actively courting

* Students with nationally recognized achievements

**Timing:** Mid-February to early March (Ivy League agreement restricts contact before October 1).

**Purpose:** Prevent top admits from committing elsewhere during the waiting period.

**Simulation relevance:** Likely letters are essentially "pre-admit" signals. In a simulation, they could be modeled as early-round offers for the top 1-2% of applicants at each school.

### 4.4 Demonstrated Interest

**What colleges track:**

* Campus visits (sign-in sheets)

* Email open rates and click-through rates

* Virtual event attendance and duration

* Interview requests

* Applicant portal login frequency

* Information session attendance

* College fair booth visits

**Who tracks it:**

* **Do NOT track (yield too high):** Harvard, Yale, Princeton, Stanford, MIT, and most Ivies

* **Do track (need yield optimization):** Tufts, Tulane, Emory, Lehigh, American, Case Western, Northeastern, Boston University, most LACs, most schools outside the top ~25

**Simulation parameter:**

```javascript proof:W3sidHlwZSI6InByb29mQXV0aG9yZWQiLCJmcm9tIjowLCJ0byI6MTgwLCJhdHRycyI6eyJieSI6ImFpOmNsYXVkZSJ9fV0=
college.tracksDemonstratedInterest = true | false;
// If true, students who mark the school as a top choice get a boost
// Typical boost: 1.1x to 1.3x multiplier on admission score

5. Institutional Priority Buckets (ALDC)

5.1 What Is ALDC?

ALDC stands for:

A = Athletes (recruited varsity athletes)
L = Legacies (children of alumni)
D = Dean's Interest List (connected to major donors or institutional priorities)
C = Children of faculty/staff

These categories receive significant admissions advantages at elite schools.

5.2 Harvard ALDC Data (SFFA v. Harvard Trial)

From the Arcidiacono et al. NBER study using 6 years of Harvard admissions data (2014-2019):

Admission Rates by Category

Category	Admit Rate	vs. Non-ALDC Baseline (~6%)
Recruited Athletes	86%	14x baseline
Children of Faculty/Staff	47%	8x baseline
Dean's Interest List	42%	7x baseline
Legacy	33%	5.7x baseline
Non-ALDC	~6%	1x (baseline)

Class Composition

Category	% of Applicant Pool	% of Admitted Class
ALDC (combined)	~5%	~30%
Recruited Athletes	< 1%	~10-12%
Legacies	~5%	~14%
Dean's Interest List	< 1%	~3-5%
Faculty/Staff Children	< 1%	~2-3%

Racial Composition of ALDC

Among white admitted students: 43% are ALDC
Among African American admitted students: 16% are ALDC
Among Asian American admitted students: 16% are ALDC
Among Hispanic admitted students: 16% are ALDC

The Counterfactual Finding

Roughly 75% of white ALDC admits would have been rejected absent their ALDC status. This means three-quarters of legacy/athlete/donor-connected white students only got in because of their ALDC boost.

5.3 SAT-Equivalent Advantages

Research estimates the admissions advantage of ALDC status in SAT-point equivalents:

Category	Estimated SAT Advantage
Recruited Athlete	~200+ points
Legacy	~160 points
Dean's Interest List (Donor)	~150+ points
First-Generation	~40-60 points

5.4 Modeling ALDC as Capacity Constraints

In the simulation, ALDC admits should be modeled as reserved capacity that is filled before general admissions:

```javascript proof:W3sidHlwZSI6InByb29mQXV0aG9yZWQiLCJmcm9tIjowLCJ0byI6MTI1NywiYXR0cnMiOnsiYnkiOiJhaTpjbGF1ZGUifX1d function allocateClass(college) { const totalSlots = college.targetClassSize;

// Reserved bucket allocation (approximate) const buckets = { recruitedAthletes: Math.round(totalSlots * 0.12), // ~12% legacy: Math.round(totalSlots * 0.14), // ~14% deansInterest: Math.round(totalSlots * 0.04), // ~4% facultyChildren: Math.round(totalSlots * 0.02), // ~2% // Total reserved: ~32% of class };

const generalSlots = totalSlots - buckets.recruitedAthletes - buckets.legacy - buckets.deansInterest - buckets.facultyChildren;

return { reserved: buckets, // ~32% of class generalAdmission: generalSlots // ~68% of class }; }

// Each bucket has its own admission threshold function getAdmitThreshold(college, bucket) { const baseThreshold = college.generalAdmitThreshold;

switch (bucket) { case 'recruitedAthlete': return baseThreshold * 0.50; // Much lower bar case 'legacy': return baseThreshold * 0.70; // Significantly lower bar case 'deansInterest': return baseThreshold * 0.65; // Very low bar case 'facultyChild': return baseThreshold * 0.75; // Lower bar default: return baseThreshold; // General admission } }

### 5.5 School-Specific ALDC Percentages

| School        | Athletes % | Legacy % | Total ALDC % | Notes                  |
| -------------------------------------------------------------------- | ----------------------------------------------------------------- | --------------------------------------------------------------- | ------------------------------------------------------------------- | ----------------------------------------------------------------------------- |
| Harvard       | 10-12%     | 14%      | ~30%         | SFFA trial data        |
| Princeton     | ~15%       | ~11%     | ~28%         | Athletic powerhouse    |
| Stanford      | ~12%       | ~10%     | ~25%         | Estimate               |
| Yale          | ~13%       | ~12%     | ~28%         | Estimate               |
| MIT           | ~12%       | ~5%      | ~20%         | Less legacy emphasis   |
| Duke          | ~10%       | ~12%     | ~25%         | Estimate               |
| Northwestern  | ~8%        | ~10%     | ~22%         | Estimate               |
| Johns Hopkins | ~8%        | ~4%      | ~15%         | Dropped legacy in 2014 |
| Amherst       | ~15%       | ~8%      | ~25%         | NESCAC athletics       |

**Note:** Many of these percentages are estimates based on available data and news reports. Only Harvard has had its data fully exposed through litigation. Schools that have dropped legacy preferences (JHU, Amherst post-2023, MIT) will have lower ALDC percentages.

***

## 6. Simulation Integration: Complete College Decision Algorithm

### 6.1 End-to-End Admission Decision Flow

```javascript proof:W3sidHlwZSI6InByb29mQXV0aG9yZWQiLCJmcm9tIjowLCJ0byI6MTkzMywiYXR0cnMiOnsiYnkiOiJhaTpjbGF1ZGUifX1d
function collegeDecision(college, applicant, round) {
  // Step 1: Calculate Academic Index
  const ai = calculateAcademicIndex(applicant.gpa, applicant.sat, applicant.act);

  // Step 2: Check ALDC status and assign to bucket
  const bucket = getApplicantBucket(applicant, college);

  // Step 3: Check minimum AI threshold
  const minAI = (bucket === 'recruitedAthlete') ? 171 : college.minAI;
  if (ai < minAI && bucket === 'general') {
    return { decision: 'REJECT', reason: 'Below minimum AI' };
  }

  // Step 4: Calculate holistic score (simulated reader rating)
  const academicScore   = scoreAcademic(ai, college);       // 1-6 scale
  const ecScore         = scoreExtracurricular(applicant);   // 1-6 scale
  const personalScore   = scorePersonal(applicant);          // 1-6 scale
  const athleticScore   = scoreAthletic(applicant);          // 1-6 scale

  // Step 5: Compute composite score
  let composite = (
    academicScore * 0.40 +
    ecScore * 0.25 +
    personalScore * 0.25 +
    athleticScore * 0.10
  );

  // Step 6: Apply ALDC multiplier
  const aldcMultiplier = getALDCMultiplier(bucket);
  composite *= aldcMultiplier;

  // Step 7: Apply round multiplier
  const roundMultiplier = getRoundMultiplier(round);
  composite *= roundMultiplier;

  // Step 8: Apply yield protection (non-HYPSM only)
  composite = applyYieldProtection(college, applicant, composite);

  // Step 9: Add randomness (holistic review uncertainty)
  composite *= (1 + (Math.random() - 0.5) * 0.25); // +/- 12.5%

  // Step 10: Compare against threshold and capacity
  const threshold = getAdmitThreshold(college, bucket);
  if (composite >= threshold && hasBucketCapacity(college, bucket)) {
    return { decision: 'ADMIT', score: composite, bucket: bucket };
  } else if (composite >= threshold * 0.85) {
    return { decision: 'WAITLIST', score: composite };
  } else {
    return { decision: 'REJECT', score: composite };
  }
}

6.2 ALDC Multipliers for Simulation

```javascript proof:W3sidHlwZSI6InByb29mQXV0aG9yZWQiLCJmcm9tIjowLCJ0byI6NDYwLCJhdHRycyI6eyJieSI6ImFpOmNsYXVkZSJ9fV0= function getALDCMultiplier(bucket) { switch (bucket) { case 'recruitedAthlete': return 3.5; // ~14x admit rate vs baseline case 'legacy': return 2.5; // ~5.7x admit rate case 'deansInterest': return 4.0; // Donor connections, ~7x admit rate case 'facultyChild': return 2.0; // ~8x rate, but smaller pool case 'firstGen': return 1.4; // Modest boost default: return 1.0; // No boost } }

### 6.3 Round Multipliers

```javascript proof:W3sidHlwZSI6InByb29mQXV0aG9yZWQiLCJmcm9tIjowLCJ0byI6NDA1LCJhdHRycyI6eyJieSI6ImFpOmNsYXVkZSJ9fV0=
function getRoundMultiplier(round) {
  switch (round) {
    case 'ED':   return 1.5;  // Binding = guaranteed yield = school loves it
    case 'EDII': return 1.4;  // Slightly less advantage than ED
    case 'REA':  return 1.2;  // Restrictive EA shows strong interest
    case 'EA':   return 1.1;  // Some demonstrated interest
    case 'RD':   return 1.0;  // Baseline
    default:     return 1.0;
  }
}

6.4 Yield Management: How Many Offers to Send

```javascript proof:W3sidHlwZSI6InByb29mQXV0aG9yZWQiLCJmcm9tIjowLCJ0byI6NzQ3LCJhdHRycyI6eyJieSI6ImFpOmNsYXVkZSJ9fV0= function calculateOffers(college, round) { const target = college.targetClassSize; const alreadyCommitted = college.committedStudents; // ED/EDII commits const remaining = target - alreadyCommitted;

// Adjust yield expectation by round let expectedYield; switch (round) { case 'ED': expectedYield = 1.00; break; // Binding case 'EDII': expectedYield = 1.00; break; // Binding case 'EA': expectedYield = college.eaYieldRate || 0.45; break; case 'RD': expectedYield = college.rdYieldRate || 0.35; break; }

const offersNeeded = Math.ceil(remaining / expectedYield);

// Add safety margin (5-10%) to account for yield uncertainty const safetyMargin = 1.07; return Math.ceil(offersNeeded * safetyMargin); }

### 6.5 Waitlist Resolution Algorithm

```javascript proof:W3sidHlwZSI6InByb29mQXV0aG9yZWQiLCJmcm9tIjowLCJ0byI6ODE0LCJhdHRycyI6eyJieSI6ImFpOmNsYXVkZSJ9fV0=
function resolveWaitlist(college) {
  const committed = college.committedStudents;
  const target = college.targetClassSize;
  const gap = target - committed;

  if (gap <= 0) return; // Class is full or over-enrolled

  // Sort waitlisted students by score, with adjustments for
  // diversity needs, geographic balance, intended major gaps
  const waitlist = college.waitlistedStudents
    .filter(s => s.stillInterested) // Some students committed elsewhere
    .sort((a, b) => b.compositeScore - a.compositeScore);

  // Pull students off waitlist to fill gap
  // Assume ~30-50% of waitlisted students will accept
  const pullCount = Math.ceil(gap / 0.40);
  const pulled = waitlist.slice(0, Math.min(pullCount, waitlist.length));

  pulled.forEach(student => {
    college.admitFromWaitlist(student);
  });
}

7. Key Parameters Summary

For College Agent Configuration

Parameter	Description	Typical Range
`targetClassSize`	Target enrollment	400-3,500
`overallAcceptRate`	Overall acceptance rate	3-30%
`yieldRate`	Historical overall yield	30-88%
`edYieldRate`	ED yield (always 100%)	1.0
`eaYieldRate`	EA/REA yield	0.40-0.60
`rdYieldRate`	RD yield	0.25-0.50
`medianAI`	Median Academic Index of admits	200-235
`minAI`	Floor AI for general admits	170-215
`athleteMinAI`	Floor AI for recruited athletes	160-185
`yieldProtectionStrength`	How aggressively school yield-protects	0.0-0.5
`tracksDemonstratedInterest`	Whether school uses DI in decisions	true/false
`aldcPct`	% of class pre-allocated to ALDC	15-32%
`athletePct`	% of class that are recruited athletes	8-15%
`legacyPct`	% of class that are legacies	4-14%
`donorPct`	% of class from dean's list / donors	2-5%

Sources

Arcidiacono, P., Kinsler, J., & Ransom, T. "Legacy and Athlete Preferences at Harvard." NBER Working Paper 26316 / Journal of Labor Economics, Vol 40, No 1 (2022).
Harvard Crimson, "Harvard Ranks Applicants on 'Humor' and 'Grit,' Court Filings Show" (2018).
Harvard Crimson, "Legacy, Athlete, and Donor Preferences Disproportionately Benefit White Applicants" (2019).
IvyWise, "Yield Rates for the Class of 2029."
IvyCoach, "Ivy League Yield Rates, Class of 2029."
CollegeVine, "What Is the Academic Index?"
COR Athletics, "Understanding the Academic Index for Ivy League Recruiting."
Top Tier Admissions, "How to Calculate Your Academic Index."
BestColleges, "Is Yield Protection Real?"
CollegeVine, "What is Yield Protection/Tufts Syndrome?"
IvyBound, "The Harvard Admissions Process: The Ratings Scale."
IvyWise, "Waitlist Admission Rates."
SFFA v. Harvard trial documents and reading procedures.
Solomon Admissions, "Ivy League Academic Index Explained."
Spark Admissions, "What a Likely Letter Means for Ivy League Applicants."

Unweighted GPA	CGS (Converted Grade Score)
4.0	80
3.9	79
3.8	78
3.7	77
3.6	73
3.5	71
3.4	68
3.3	66
3.2	64
3.1	62
3.0	60
2.9	58
2.8	56
2.7	54
2.6	52
2.5	50

ACT Composite	Conversion Value	AI Contribution (Value * 2)
36	80	160
35	78	156
34	76	152
33	74	148
32	72	144
31	70	140
30	68	136
29	66	132
28	64	128

Unweighted GPA	CGS (Converted Grade Score)
4.0	80
3.9	79
3.8	78
3.7	77
3.6	73
3.5	71
3.4	68
3.3	66
3.2	64
3.1	62
3.0	60
2.9	58
2.8	56
2.7	54
2.6	52
2.5	50

ACT Composite	Conversion Value	AI Contribution (Value * 2)
36	80	160
35	78	156
34	76	152
33	74	148
32	72	144
31	70	140
30	68	136
29	66	132
28	64	128

Unweighted GPA	CGS (Converted Grade Score)
4.0	80
3.9	79
3.8	78
3.7	77
3.6	73
3.5	71
3.4	68
3.3	66
3.2	64
3.1	62
3.0	60
2.9	58
2.8	56
2.7	54
2.6	52
2.5	50

ACT Composite	Conversion Value	AI Contribution (Value * 2)
36	80	160
35	78	156
34	76	152
33	74	148
32	72	144
31	70	140
30	68	136
29	66	132
28	64	128