# Adjudicated corrections ledger

## 2026-06-14 — Applications parse fix · ON-CHART
### George Washington — apps_trend column misalignment
| year | from | to | basis |
|---|---|---|---|
| 2022 | 2,671 | 9,021 | owner-adjudicated; 2,671 sat near that year's offers count (column misalignment). 2025 (9,718) already correct. |
| 2023 | 5,554 | 8,310 | owner-adjudicated; consistent with the 8,091 (2024) / 9,718 (2025) recovery. |
Registered in FIXES["george-washington-university-the"].apps_trend. Static
school page unaffected (shows the apps scalar 9,718, already correct).

### Admissions row misalignment — curated apps/offers/acc synced to gz source-of-record
32 points across 13 schools where the curated `*_trend` had drifted from the gz
(the ABA source of record). Confirmed by cross-check: GW's owner-given values
matched gz exactly. Synced curated -> gz. Rutgers (2013) and Widener
Commonwealth/Delaware HELD for owner review (pre-merge/split attribution +
shared-coordinates geo bug) — NOT synced.

| school | year(s) | correction |
|---|---|---|
| Boston University | 2025 | acc 8->12, apps 7,668->7,892, offers 650->953 |
| UC Law SF | 2025 | acc 42->28, apps 2,100->4,593, offers 890->1,262 |
| Chicago-Kent | 2024 | apps 5,366->1,781, offers 683->879 |
| Cooley | 2015 | apps 469->1,222, offers 400->1,072 |
| UDC | 2017/2023 | acc(2017) 0->40; apps(2023) 7,754->530, offers(2023) 948->235 |
| Emory | 2025 | acc 30->39, offers 1,173->1,505 |
| George Washington | 2022/2023 | acc 30->25 & 17->29; offers 809->2,255 & 953->2,387 (apps fixed earlier) |
| Loyola Marymount | 2022 | apps 843->3,414, offers 562->1,175 |
| Loyola Chicago | 2023 | apps 5,259->3,115, offers 672->1,368 |
| UMKC | 2024 | acc 38->62, apps 1,214->590, offers 465->365 |
| New England | 2023 | apps 5,650->2,937, offers 756->2,182 |
| Texas | 2012 | apps 1,735->4,316, offers 921->1,172 |
| Virginia | 2023 | apps 534->5,610, offers 241->643 |

### Scorched-earth scalar sweep (reconcile_gz.py)
- schol_none_pct clamped 0.0 (was a small negative from bucket rounding):
  Cincinnati, Dayton, Indiana-Indianapolis, Mitchell/Hamline, Penn State Dickinson.
- Wilmington tui/nrt 24,302 -> 25,140 (synced to gz source of record).
Post-sweep: reconcile_gz reports 0 inversions, 0 impossibilities, 0 gz drift
across all 210 schools. reconcile_gz --strict is now wired into build.sh/CI.

### Held merge/split cases — owner-adjudicated
- Rutgers 2013 (merged entity): apps 1,852 -> 3,109, offers 730 -> 1,082
  (Camden 1,257/352 + Newark 1,852/730 combined; curated had Newark-only, gz had
  Camden-only). Flagged ("rutgers-university", apps/offers) in overrides.py.
- Widener Commonwealth 2012: apps 1,869 -> 1,214, offers 1,033 -> 627. Curated had
  duplicated Widener Delaware's figure onto Commonwealth; gz (and owner) agree on
  1,214/627. Delaware's 2012 (1,869/1,033) was already correct, left as-is.
- Widener coordinates (both shared one wrong point 39.8521/-75.3877):
  Delaware (Wilmington) -> 39.81748, -75.54646; Commonwealth (Harrisburg)
  -> 40.32409, -76.85456. Fixes the shared-pin geo bug.


Curated/display-layer corrections that override a raw ABA cell in
`data/exhibit-data.js`. The raw gz (`data/exhibit_data.json.gz`) keeps the
original ABA value as the source of record; on a source re-upload + re-extract
the gz is regenerated, but curated overrides for existing schools persist.

Corrections flagged on the interactive chart (amber hollow ring + hover note)
are registered in `FIXES` in `index.html`; nonresident-only and percentile
fixes that don't surface on the resident chart are recorded here only.

## 2026-06-11 — anomaly sweep (Tier 3)

### Tuition — misreported per-semester halves (raised to fit trend) · ON-CHART
| school | year(s) | from (res/nr) | to (res/nr) | basis |
|---|---|---|---|---|
| Texas | 2017, 2018 | 17,857 / 26,517 | 35,714 / 53,034 | per-semester halves; trend is ~$34–38k throughout. Reverses the Tier-2 sheet's 2018 = 17,857 (that was the misreport). |
| Cincinnati | 2021 | 12,005 / 14,505 | 24,010 / 29,010 | lone half; every other year 2014–2025 = 24,010 / 29,010. |
| South Carolina | 2021 | 9,653 / 24,661 | 18,556 / 35,342 | half-value outlier; aligned to the 2022–23 plateau. |

### Tuition — Stanford fill (interpolated 2017→2021) · ON-CHART
| year | from | to | basis |
|---|---|---|---|
| 2018 | 0 | 61,290 | $0 text-marker corruption; interpolated. |
| 2019 | 41,450 | 62,310 | partial/quarter figure; interpolated. |
| 2020 | 42,900 | 63,330 | partial/quarter figure; interpolated. |
nr mirrors res (private); nr 2017 filled 60,270; nrt scalar 0 → 77,454.

### Tuition — Florida Coastal · ON-CHART
- 2019 resident 79,580 → 39,790 (private; res = nr).

### Enrollment — Montana · ON-CHART
- `enr_trend` 2021 = 3 → 248 (curated corruption; gz total JD enrollment is 248).
  Note: `enr` = total JD enrollment (ABA "Total Grand Total"); `enr_1l` = entering 1L class.

### LSAT percentiles — South Dakota (ledger only)
- 2011 25th/75th were inverted (152/148) → restored to 148 / 152 (50th = 150 unchanged).

### Nonresident = resident (private; remove divergent nonres) (ledger only)
- Detroit Mercy: `nrt_trend` mirrored to resident for all years (2023–25 nr 45,037/45,936/46,864 dropped); nrt scalar 46,864 → 56,143. **2025 resident → $27,800** (tuition cut; res=nr; scalar updated). **2023 & 2024 still pending** owner values — they carry the old ~$53–54k sticker, inconsistent with a 2025 of $27,800.
- Northwestern 2025 nr 79,722 → 79,772 (+ scalar); Boston College 2016 nr 50,700 → 50,770; Denver 2012 nr 39,480 → 39,840; Willamette 2020 nr 46,530 → 47,130; Mitchell/Hamline 2016 nr 40,470 → 40,570.

### Public — nonresident correction (ledger only)
- Indiana-Indianapolis 2016 nr 22,614 → 45,227 (resident 26,379 unchanged).

## 2026-06-11 — E resolutions (apostrophe near-zero cells)

Owner-adjudicated. "null" = explicit missing key → interpolated open-circle on chart.
- Dayton: tui+nrt 2018, 2019, 2020 → null. (2021, 2022 still $0 — pending.)
- Indiana-Indianapolis: 2018 → null; **2019 → res $28,672 / nr $48,736**. (2023 still $0 — pending.)
- Washburn: 2018, 2019 → null. (2020, 2021 still $0 — pending.)
- Illinois-Chicago (UIC): 2018, 2019 → null. (2020 still $0 — pending.)
- Widener Delaware: 2018, 2019 → null. (2020 still $0 — pending.)
- Widener Commonwealth: 2018, 2019 → null. (2020 still $0 — pending.)
- Inter-American PR: 2019 → null.

Nebraska (F): explanatory note added to methodology.html (grant > resident tuition is
an artifact of residency-pooled, non-resident-sticker-scaled 509 aid reporting).

## 2026-06-11 — master-workbook reconciliation (owner-adjudicated to ABA)

Resolved the 21 genuine display-vs-master discrepancies to the owner's master values:
- **Montana** 2023/24/25 res 19,494 / 24,161 / 24,582; nr 44,822 / 51,161 / 51,582
  (was ~$12k; fixes the scalar-vs-trend drift warning too).
- **Texas Southern** 2024 res 18,582 / nr 25,632 (was 12,645 / 20,049).
- **Indiana-Indianapolis** 2017 res 28,631 / nr 48,711 (was 26,736).
- **Cornell** 2019 res=nr 67,748 (superseded the Tier-2 sheet's 65,541).
- **Florida Coastal** 2020 res=nr 19,895 (was doubled to 79,580; school winding down).
- **South Carolina** 2021 res 9,653 / nr 25,661 (reverted my fit-trend 18,556; ABA per-semester figure, shown as reported with a note).
- **Florida State** 2019 res 18,072 / nr 22,634 (nr was 36,144; owner flags a fee/tuition flip, low confidence).
- **Oklahoma** 2025 res 18,379 / nr 33,357 (was 18,739; fixes the scalar drift).

Reconciliation evidence: `master-reconciliation.md` / `.csv`. Removed the now-stale
South Carolina "$18,556 plateau" ring note.

## 2026-06-11 — North Dakota (owner-adjudicated)

- North Dakota 2020 resident 14,296 -> 15,296 (ABA figure; fits the 2019 $15,618 / 2021 $15,011 bracket). Nonresident 2018-2020 (29,644 / 42,836 / 42,888) confirmed correct by owner.

## Deferred / not changed
- **E — apostrophe near-zero cells** (Stanford 2018 prior, Dayton, Washburn, UIC, Widener ×2, Indiana-Indianapolis, Inter-American PR + closed schools): await the ABA 509 re-upload + re-extract (parser fix already in place). Closed schools stay missing.
- **F — Nebraska** grant_med > resident tuition: not a bug (grants scaled to nonresident/sticker, pooled school-wide).

2026-06-13 (v1.94.33) — anomaly-scan follow-ups:
- Marquette bar_trend 2024: 0% → 99.46% (WI diploma privilege; ABA Total First-Time incl. alt pathways; 1 of 186 grads took the exam). Source: marquette_2024 ABA 509.
- trans_gpa50 nulled where an LSAT-scale value sat in the GPA field and no source GPA exists: Georgia State 2014 (151), Mitchell/Hamline 2019 (149), St. Mary's 2015 (146). 2023-25 transfer file does not cover these years.
- schol_none_pct negatives (Cincinnati -0.3, Penn State Dickinson -0.4, Dayton -1.9, Indiana-Indy -0.8, Mitchell/Hamline -0.3): value KEPT (honest catch-all residual); now surfaces the "adjusted" note. No data change.
- Bar dips verified CORRECT vs ABA source, no change: DC 2021 (64.18%), Southern 2021 (41.66%), Vermont 2020 (45.76%).

2026-06-13 (v1.94.34) — Faculty rebuild 2011-2025 from ABA source files:
- Root cause: a 2017 ABA definition change (pre-2017 "Full-Time Fall Total" excludes deans/librarians who teach; 2017+ "Total FT" includes them) plus per-school source inconsistency (some carried total-incl-adjunct counts at 3-4x the real FT, e.g. Loyola Chicago ~200 vs real ~67).
- Rebuilt fac_trend (all 210 schools) to one consistent measure: 2011-2016 = Full-Time Fall Total + Deans/librarians-who-teach Fall Total; 2017-2022 = "Total FT"; 2023-2025 = "FTTotal". Boundary verified exact (Akron 21+11=32; Baylor 33; Loyola 67).
- Faculty 1-year V-dips/spikes 35 -> 1 (remaining = Arizona Summit's real pre-closure collapse). fac_ft scalars + gz in sync.

2026-06-13 (v1.94.35) — Transfer-GPA column transpositions from ABA source:
- 9 file cells report 25th > 75th (impossible) = a 25<->75 transposition at the source; corrected by swapping 25<->75 (50th unchanged): George Mason 2017, GWU 2017, Fordham 2019 & 2024, UNLV 2019, Rutgers 2020, SMU 2020, Florida State 2021, Michigan 2024. Fordham 2024 & Michigan 2024 were blank in our data, filled with corrected order.
- Mitchell/Hamline transfer 50th is a bad-scale LSAT value in the source (149 in 2019, 154 in 2025) -> kept null.
- Pre-2016 transfer files are counts-only (ABA added GPA ~2016), confirming the Georgia State 2014 / St. Mary's 2015 nulls. Transfer counts verified accurate (8/600 trivial diffs).

2026-06-13 (v1.94.36) — Enrollment + acceptance data-quality pass (owner master-adjudicated):

ENROLLMENT (enr_trend = total JD; reconciled vs master JD Enrollment & Ethnicity -> TotalGrandTotal):
- Howard 2021: 0 -> null (open-circle gap). Master TGT shows 467, but owner directed the cell be shown as not-reported; JS + gz both set null.
- Ohio Northern 2022: 0 -> 178 (master TGT; owner note said 467 but that is outside ON's 88-320 history and equals Howard's 2021 figure listed above it -> transcription slip, owner confirmed 178).
- Ohio Northern 2024: 0 -> 241 (master TGT; owner-flagged low-confidence/bad-data year, but master and gz agree).
- Penn State Dickinson 2017: 402 -> 193 (master TGT; was a total-vs-1L style spike). gz 2017 node created with enr=193.
- Wyoming 2021: 450 -> 229 (master TGT; the 450 spike was a stray ~2x value).
- Widener Commonwealth 2012: 811 -> 122 (owner value; pre-2015 split year, master's early "WIDENER-COMMONWEALTH" column is unreliable for the standalone Harrisburg campus).
- Cooley 2015: 423 -> 550 (owner: Lansing campus only). NOTE: neighboring Cooley years are all-campus totals (2014=1742, 2016=1209), so the 2015 cell is intentionally Lansing-only per owner.

ACCEPTANCE (acc_trend = acceptance %; reconciled vs master First Year Class -> AcceptanceRate. gz already held correct 1dp values; only the JS trend was corrupted):
- Virginia 2023: 45.13 -> 11.46. The 45.13 was West Virginia's 2023 rate (adjacent-school contamination); master UVA = 643 offers / 5610 apps = 11.46.
- Loyola Marymount 2022: 66.67 -> 34.42 (master).
- Chicago-Kent 2024: 12.73 -> 49.35 (master; 879/1781).
- Loyola Chicago 2023: 12.78 -> 43.92 (master; 1368/3115).
- New England Law 2023: 13.38 -> 74.29 (master; 2182/2937).
- DC 2023: 12.23 -> 44.34 (master; 235/530).
- Ohio Northern 2024 acceptance (11.53): VERIFIED REAL by master (1041 apps -> 120 offers -> 74 enrolled), NOT an error -> no change, despite the handoff flag.

Method: byte-safe S-object splice (all 210 objects round-trip identical) + gz compaction; reconcile-before-apply diff against Exhibit_509_Master_updated_20260613.xlsx. Staging tooling in pipeline/_staging/.
Observation (not changed, out of scope): the JS enr_trend year-labels are inconsistently aligned vs the gz/master across schools (e.g. Howard's series sits +1 year vs master; Cooley's is same-year) -- likely a disclosure-cycle vs fall-year labeling mix worth a future systematic pass.