# Exhibit 509 — Dataset Validation Report

Reconciliation-based validation across 94 metrics, 218 schools, 2011–2025. **Flags are for human review, never auto-corrected.** A flag is not proof of error: some are real reporting events (e.g. an acquisition year), some are structural ABA reporting gaps, some are genuine data defects.

**128 flags after deduplication** (mirror-bucket duplicates removed).

## How to read severity
- **logical** — hard contradiction (A>B where impossible). Highest priority.
- **bounds** — value outside its valid range (e.g. % >100). Check parsing.
- **ordering** — percentile out of order (25th>median). Check column mapping.
- **spike_review** — single-year reversal with NO sibling absorption. Likely defect.
- **reconcile** — a group that should sum/match is off. Often structural, verify.
- **spike_reconciled** — single-year reversal that sibling buckets absorb. Likely REAL event, kept for transparency.

## LOGICAL (4)

- **University of Nevada-Las Vegas** (2025) — `clinic_filled_gt_avail`: 51 filled > 48 available
- **University of Pennsylvania** (2025) — `clinic_filled_gt_avail`: 140 filled > 132 available
- **Washburn University** (2025) — `clinic_filled_gt_avail`: 49 filled > 40 available
- **Widener University Delaware Law School** (2025) — `clinic_filled_gt_avail`: 58 filled > 48 available

## SPIKE_REVIEW (1)

- **Columbia University** (2021) — `single_year_spike`: % grant < half tuition: 40.3->78.8->41.1 (reverses); no sibling absorption, REVIEW

## KNOWN_ARTIFACT (3)

- **Marquette University** (2021) — `diploma_privilege_small_n`: Two-year ultimate bar pass %: 99.0->0.0->100.0. Wisconsin diploma privilege: only the few grads who sit an out-of-state/UBE exam are counted, so the denominator is tiny and these values are REAL but not a quality signal.
- **Marquette University** (2022) — `diploma_privilege_small_n`: Two-year ultimate bar pass %: 0.0->100.0->57.1. Wisconsin diploma privilege: only the few grads who sit an out-of-state/UBE exam are counted, so the denominator is tiny and these values are REAL but not a quality signal.
- **Marquette University** (2023) — `diploma_privilege_small_n`: Two-year ultimate bar pass %: 100.0->57.1->100.0. Wisconsin diploma privilege: only the few grads who sit an out-of-state/UBE exam are counted, so the denominator is tiny and these values are REAL but not a quality signal.

## RECONCILE (111)

- **Appalachian School of Law** (2020) — `firm_buckets_vs_total`: firm buckets sum 28.8 vs Law firm jobs % 23.1 (off +5.7)
- **Appalachian School of Law** (2021) — `firm_buckets_vs_total`: firm buckets sum 27.2 vs Law firm jobs % 20.0 (off +7.2)
- **Atlanta's John Marshall Law School** (2021) — `firm_buckets_vs_total`: firm buckets sum 55.3 vs Law firm jobs % 48.7 (off +6.6)
- **Faulkner University** (2019) — `firm_buckets_vs_total`: firm buckets sum 40.4 vs Law firm jobs % 32.8 (off +7.6)
- **Lincoln Memorial University** (2025) — `firm_buckets_vs_total`: firm buckets sum 55.4 vs Law firm jobs % 49.1 (off +6.3)
- **Oklahoma City University** (2022) — `firm_buckets_vs_total`: firm buckets sum 52.6 vs Law firm jobs % 46.7 (off +5.9)
- **Southern University** (2020) — `firm_buckets_vs_total`: firm buckets sum 25.4 vs Law firm jobs % 18.7 (off +6.7)
- **Southern University** (2021) — `firm_buckets_vs_total`: firm buckets sum 25.6 vs Law firm jobs % 19.8 (off +5.8)
- **Texas Southern University** (2019) — `firm_buckets_vs_total`: firm buckets sum 37.7 vs Law firm jobs % 29.5 (off +8.2)
- **Texas Southern University** (2020) — `firm_buckets_vs_total`: firm buckets sum 33.5 vs Law firm jobs % 27.3 (off +6.2)
- **Texas Southern University** (2021) — `firm_buckets_vs_total`: firm buckets sum 38.0 vs Law firm jobs % 26.9 (off +11.1)
- **University of Idaho** (2020) — `firm_buckets_vs_total`: firm buckets sum 39.8 vs Law firm jobs % 33.7 (off +6.1)
- **University of Nebraska** (2021) — `firm_buckets_vs_total`: firm buckets sum 54.4 vs Law firm jobs % 49.1 (off +5.3)
- **University of Arkansas at Little Rock** (2020) — `grant_sum`: grant buckets sum to 107.5, expected ~100 (off by +7.5)
- **Case Western Reserve University** (2012) — `race_sum`: race buckets sum to 89.5, expected ~100 (off by -10.5)
- **Columbia University** (2011) — `race_sum`: race buckets sum to 90.3, expected ~100 (off by -9.7)
- **Columbia University** (2012) — `race_sum`: race buckets sum to 89.1, expected ~100 (off by -10.9)
- **Columbia University** (2013) — `race_sum`: race buckets sum to 88.1, expected ~100 (off by -11.9)
- **Columbia University** (2014) — `race_sum`: race buckets sum to 90.8, expected ~100 (off by -9.2)
- **Columbia University** (2015) — `race_sum`: race buckets sum to 89.6, expected ~100 (off by -10.4)
- **Columbia University** (2016) — `race_sum`: race buckets sum to 88.7, expected ~100 (off by -11.3)
- **Columbia University** (2017) — `race_sum`: race buckets sum to 86.2, expected ~100 (off by -13.8)
- **Columbia University** (2018) — `race_sum`: race buckets sum to 86.9, expected ~100 (off by -13.1)
- **Columbia University** (2019) — `race_sum`: race buckets sum to 87.3, expected ~100 (off by -12.7)
- **Columbia University** (2020) — `race_sum`: race buckets sum to 86.8, expected ~100 (off by -13.2)
- **Columbia University** (2021) — `race_sum`: race buckets sum to 83.5, expected ~100 (off by -16.5)
- **Columbia University** (2022) — `race_sum`: race buckets sum to 84.0, expected ~100 (off by -16.0)
- **Columbia University** (2023) — `race_sum`: race buckets sum to 81.6, expected ~100 (off by -18.4)
- **Cornell University** (2013) — `race_sum`: race buckets sum to 89.0, expected ~100 (off by -11.0)
- **Cornell University** (2014) — `race_sum`: race buckets sum to 86.2, expected ~100 (off by -13.8)
- **Cornell University** (2015) — `race_sum`: race buckets sum to 81.4, expected ~100 (off by -18.6)
- **Cornell University** (2016) — `race_sum`: race buckets sum to 88.9, expected ~100 (off by -11.1)
- **Cornell University** (2017) — `race_sum`: race buckets sum to 80.7, expected ~100 (off by -19.3)
- **Cornell University** (2018) — `race_sum`: race buckets sum to 84.3, expected ~100 (off by -15.7)
- **Cornell University** (2019) — `race_sum`: race buckets sum to 87.9, expected ~100 (off by -12.1)
- **Cornell University** (2020) — `race_sum`: race buckets sum to 82.8, expected ~100 (off by -17.2)
- **Cornell University** (2021) — `race_sum`: race buckets sum to 83.9, expected ~100 (off by -16.1)
- **Cornell University** (2022) — `race_sum`: race buckets sum to 77.5, expected ~100 (off by -22.5)
- **Cornell University** (2023) — `race_sum`: race buckets sum to 79.2, expected ~100 (off by -20.8)
- **Duke University** (2023) — `race_sum`: race buckets sum to 89.5, expected ~100 (off by -10.5)
- **Emory University** (2015) — `race_sum`: race buckets sum to 89.1, expected ~100 (off by -10.9)
- **Emory University** (2016) — `race_sum`: race buckets sum to 87.0, expected ~100 (off by -13.0)
- **Emory University** (2017) — `race_sum`: race buckets sum to 87.1, expected ~100 (off by -12.9)
- **Emory University** (2018) — `race_sum`: race buckets sum to 86.6, expected ~100 (off by -13.4)
- **Emory University** (2019) — `race_sum`: race buckets sum to 85.5, expected ~100 (off by -14.5)
- **Emory University** (2020) — `race_sum`: race buckets sum to 88.0, expected ~100 (off by -12.0)
- **Harvard University** (2015) — `race_sum`: race buckets sum to 90.2, expected ~100 (off by -9.8)
- **Harvard University** (2016) — `race_sum`: race buckets sum to 87.8, expected ~100 (off by -12.2)
- **Harvard University** (2017) — `race_sum`: race buckets sum to 86.2, expected ~100 (off by -13.8)
- **Harvard University** (2018) — `race_sum`: race buckets sum to 85.5, expected ~100 (off by -14.5)
- **Harvard University** (2019) — `race_sum`: race buckets sum to 86.6, expected ~100 (off by -13.4)
- **Harvard University** (2020) — `race_sum`: race buckets sum to 89.7, expected ~100 (off by -10.3)
- **Indiana University-Bloomington** (2018) — `race_sum`: race buckets sum to 90.5, expected ~100 (off by -9.5)
- **Indiana University-Bloomington** (2019) — `race_sum`: race buckets sum to 90.9, expected ~100 (off by -9.1)
- **Indiana University-Bloomington** (2023) — `race_sum`: race buckets sum to 89.8, expected ~100 (off by -10.2)
- **Inter American University of Puerto Rico** (2015) — `race_sum`: race buckets sum to 73.9, expected ~100 (off by -26.1)
- **New York University** (2017) — `race_sum`: race buckets sum to 90.7, expected ~100 (off by -9.3)
- **New York University** (2018) — `race_sum`: race buckets sum to 90.8, expected ~100 (off by -9.2)
- **Northwestern University** (2015) — `race_sum`: race buckets sum to 90.2, expected ~100 (off by -9.8)
- **Northwestern University** (2016) — `race_sum`: race buckets sum to 89.9, expected ~100 (off by -10.1)
- **Northwestern University** (2017) — `race_sum`: race buckets sum to 90.7, expected ~100 (off by -9.3)
- **Northwestern University** (2019) — `race_sum`: race buckets sum to 90.4, expected ~100 (off by -9.6)
- **Northwestern University** (2022) — `race_sum`: race buckets sum to 90.8, expected ~100 (off by -9.2)
- **Northwestern University** (2023) — `race_sum`: race buckets sum to 89.7, expected ~100 (off by -10.3)
- **University of Arizona** (2014) — `race_sum`: race buckets sum to 88.1, expected ~100 (off by -11.9)
- **University of Arizona** (2015) — `race_sum`: race buckets sum to 88.8, expected ~100 (off by -11.2)
- **University of Arizona** (2016) — `race_sum`: race buckets sum to 90.1, expected ~100 (off by -9.9)
- **University of Arizona** (2017) — `race_sum`: race buckets sum to 90.5, expected ~100 (off by -9.5)
- **University of Arizona** (2018) — `race_sum`: race buckets sum to 90.1, expected ~100 (off by -9.9)
- **University of Arizona** (2019) — `race_sum`: race buckets sum to 88.2, expected ~100 (off by -11.8)
- **University of Arizona** (2020) — `race_sum`: race buckets sum to 88.8, expected ~100 (off by -11.2)
- **University of Detroit Mercy** (2011) — `race_sum`: race buckets sum to 76.4, expected ~100 (off by -23.6)
- **University of Detroit Mercy** (2012) — `race_sum`: race buckets sum to 73.2, expected ~100 (off by -26.8)
- **University of Detroit Mercy** (2014) — `race_sum`: race buckets sum to 62.3, expected ~100 (off by -37.7)
- **University of Detroit Mercy** (2015) — `race_sum`: race buckets sum to 56.5, expected ~100 (off by -43.5)
- **University of Detroit Mercy** (2016) — `race_sum`: race buckets sum to 54.2, expected ~100 (off by -45.8)
- **University of Detroit Mercy** (2017) — `race_sum`: race buckets sum to 56.1, expected ~100 (off by -43.9)
- **University of Detroit Mercy** (2018) — `race_sum`: race buckets sum to 57.8, expected ~100 (off by -42.2)
- **University of Detroit Mercy** (2019) — `race_sum`: race buckets sum to 56.8, expected ~100 (off by -43.2)
- **University of Detroit Mercy** (2020) — `race_sum`: race buckets sum to 58.8, expected ~100 (off by -41.2)
- **University of Detroit Mercy** (2021) — `race_sum`: race buckets sum to 62.2, expected ~100 (off by -37.8)
- **University of Detroit Mercy** (2022) — `race_sum`: race buckets sum to 63.6, expected ~100 (off by -36.4)
- **University of Detroit Mercy** (2023) — `race_sum`: race buckets sum to 67.2, expected ~100 (off by -32.8)
- **University of Illinois** (2014) — `race_sum`: race buckets sum to 90.1, expected ~100 (off by -9.9)
- **University of Illinois** (2015) — `race_sum`: race buckets sum to 90.4, expected ~100 (off by -9.6)
- **University of Illinois** (2023) — `race_sum`: race buckets sum to 89.5, expected ~100 (off by -10.5)
- **University of Minnesota** (2013) — `race_sum`: race buckets sum to 90.4, expected ~100 (off by -9.6)
- **University of Minnesota** (2014) — `race_sum`: race buckets sum to 88.0, expected ~100 (off by -12.0)
- **University of Minnesota** (2015) — `race_sum`: race buckets sum to 88.8, expected ~100 (off by -11.2)
- **University of Minnesota** (2016) — `race_sum`: race buckets sum to 90.5, expected ~100 (off by -9.5)
- **University of Minnesota** (2017) — `race_sum`: race buckets sum to 89.8, expected ~100 (off by -10.2)
- **University of Minnesota** (2018) — `race_sum`: race buckets sum to 90.7, expected ~100 (off by -9.3)
- **University of North Dakota** (2013) — `race_sum`: race buckets sum to 89.7, expected ~100 (off by -10.3)
- **University of North Dakota** (2015) — `race_sum`: race buckets sum to 88.4, expected ~100 (off by -11.6)
- **University of North Dakota** (2016) — `race_sum`: race buckets sum to 85.1, expected ~100 (off by -14.9)
- **University of North Dakota** (2017) — `race_sum`: race buckets sum to 88.5, expected ~100 (off by -11.5)
- **University of North Dakota** (2018) — `race_sum`: race buckets sum to 87.7, expected ~100 (off by -12.3)
- **University of North Dakota** (2019) — `race_sum`: race buckets sum to 87.0, expected ~100 (off by -13.0)
- **University of North Dakota** (2020) — `race_sum`: race buckets sum to 84.4, expected ~100 (off by -15.6)
- **University of North Dakota** (2021) — `race_sum`: race buckets sum to 85.7, expected ~100 (off by -14.3)
- **University of North Dakota** (2022) — `race_sum`: race buckets sum to 88.1, expected ~100 (off by -11.9)
- **University of North Dakota** (2023) — `race_sum`: race buckets sum to 90.3, expected ~100 (off by -9.7)
- **Washington University (St. Louis)** (2013) — `race_sum`: race buckets sum to 90.8, expected ~100 (off by -9.2)
- **Washington University (St. Louis)** (2014) — `race_sum`: race buckets sum to 89.4, expected ~100 (off by -10.6)
- **Washington University (St. Louis)** (2015) — `race_sum`: race buckets sum to 90.0, expected ~100 (off by -10.0)
- **Washington University (St. Louis)** (2016) — `race_sum`: race buckets sum to 89.3, expected ~100 (off by -10.7)
- **Washington University (St. Louis)** (2017) — `race_sum`: race buckets sum to 90.5, expected ~100 (off by -9.5)
- **Washington University (St. Louis)** (2018) — `race_sum`: race buckets sum to 89.1, expected ~100 (off by -10.9)
- **Washington University (St. Louis)** (2019) — `race_sum`: race buckets sum to 88.9, expected ~100 (off by -11.1)
- **Washington University (St. Louis)** (2020) — `race_sum`: race buckets sum to 90.1, expected ~100 (off by -9.9)
- **Washington University (St. Louis)** (2023) — `race_sum`: race buckets sum to 89.1, expected ~100 (off by -10.9)

## SPIKE_RECONCILED (9)

- **City University of New York** (2023) — `single_year_spike`: % receiving NO grant: 59.5->18.3->83.2 (reverses); sibling buckets absorb it, likely REAL
- **Cleveland State University** (2015) — `single_year_spike`: % receiving NO grant: 61.7->8.6->51.2 (reverses); sibling buckets absorb it, likely REAL
- **Florida A&M University** (2021) — `single_year_spike`: % receiving NO grant: 63.9->10.6->53.0 (reverses); sibling buckets absorb it, likely REAL
- **Ohio Northern University** (2012) — `single_year_spike`: % receiving NO grant: 43.5->80.4->30.7 (reverses); sibling buckets absorb it, likely REAL
- **Texas A&M University** (2014) — `single_year_spike`: % receiving NO grant: 64.1->9.6->76.2 (reverses); sibling buckets absorb it, likely REAL
- **University of Puerto Rico** (2023) — `single_year_spike`: % receiving NO grant: 100.0->42.4->91.6 (reverses); sibling buckets absorb it, likely REAL
- **University of Puerto Rico** (2024) — `single_year_spike`: % receiving NO grant: 42.4->91.6->49.7 (reverses); sibling buckets absorb it, likely REAL
- **University of Wyoming** (2021) — `single_year_spike`: % receiving NO grant: 18.3->57.3->17.7 (reverses); sibling buckets absorb it, likely REAL
- **Yale University** (2014) — `single_year_spike`: % receiving NO grant: 42.4->79.0->39.8 (reverses); sibling buckets absorb it, likely REAL

## Priority actions

1. **Marquette bar-pass series (2021–2023)** is internally impossible (alternates 0/100). Confirmed defect. Exclude from all bar-pass analysis until source re-parsed. Already excluded from residual model.
2. **Four clinic over-capacity flags (2025)**: University of Pennsylvania (Penn Carey Law, the Philadelphia T14 — confirmed, not Penn State), UNLV, Washburn, and Widener Delaware show clinic seats filled > available (Penn 140>132). Internally impossible as parsed. Likely a fill/capacity column swap or legitimate over-enrollment above nominal capacity. Check source PDFs.
3. **Grant-bucket spikes** (A&M 2014, Yale 2014, CUNY 2023, etc.) all reconcile to ~100 across buckets. These are REAL grant-policy years, not errors. Keep as-is; annotate when charted.
4. **Race-sum gaps** cluster at −9 to −12 points: consistent with nonresident/international students sitting outside the race buckets in early ABA reports. Structural, not a parse error. 24 cases exceed 15pts and merit a spot-check.

## Human review notes (2026-06-28)

Reviewed and adjudicated; per the standing rule, none were auto-corrected.

- **Clinic over-capacity (2025) — University of Pennsylvania (Carey Law), UNLV, Washburn, Widener University Delaware Law School.** The filled-greater-than-available figures match what the schools disclosed (Penn: 140 filled / 132 available). The mechanism is not recoverable from the 509 form — possible mid-year clinic expansion, course-selection/graduation-placement adjustments, or a reporting-input quirk. Logged as a **data-input question that cannot be resolved without contacting the school directly.** Left exactly as disclosed and flagged; not corrected.
- **Columbia University (2021) — share of students on a less-than-half-tuition grant (~79%).** Confirmed as the reported value; a genuine single-year reversal in grant-size mix. Kept visible; not corrected.