WORK IN PROGRESS — NOT VERIFIED.

Internal working analysis pending editorial review; numbers may change; do not cite. Findings correspond to ledger entries with status “in progress”.

Working analysis: chronic absenteeism

Descriptives for NYC district schools (plus a charter cut on a different basis), 2026-06-10· computed read-only from the local replica built by this repo’s loaders · every number on this page is rendered from the committed packet JSON data/analysis/absenteeism-descriptives.json. Findings correspond to ledger entries A.14–A.17, A.21 and A.22(status “in progress”), with data-quality asterisks A.11 / A.13 / A.20. Revised 2026-06-11: Figure 5 now carries a district grades-1–12 series matching the charter grade basis; Figures 8–9 add a stability benchmark vs test proficiency and a pseudo-cohort decomposition.

1. Headline takeaways

Recovery is real but decelerating and incomplete. Student-weighted chronic absenteeism (CA) in district schools: 25.8% (2018-19) → 39.5% peak (2021-22) → 32.3% (2024-25); annual improvement has slowed −4.4 → −1.3 → −1.5pp, leaving 2024-25 still +6.5pp above baseline. [Fig 1; A.14]
The whole school distribution shifted up, not just the average. The median school went 27.9% → 35.7%; the 90th-percentile school 49.2% → 54.1%; schools above 50% CA: 137 (2018-19) → 520 (2021-22) → 231 (2024-25). [Figs 1, 6]
Persistent damage is concentrated in younger grades. Grade 12 is the only grade back to its pre-pandemic level (−0.1pp); grades 1–8 remain +7.1 to +8.9pp elevated.* [Fig 2; A.14, * = A.11 asterisk on 2024-25 grades 1–8]
High schools have nearly recovered; every other band has not. Vs 2018-19: HS +3.3pp, ES +7.1, MS +8.5, K8 +9.5— a reversal of the usual “HS is the problem” framing. [Fig 3]
CA amplifies the same attendance loss ≈ 4× vs ADA. At the 2021-22 trough, replica weighted ADA fell 3.3pp while CA rose 13.7pp; the two metrics rank schools nearly identically (r ≈ −0.91 every year). Official citywide figures (CA 26.5→40.3→33.3; ADA 91.5→88.1→89.8) sit ≤0.8pp from our replica equivalents — a universe difference, not a data problem. [Fig 4]
New: the charter cut (different basis — not comparable to district levels). All-reporting charter weighted CA (NYSED, grades 1–12): 41.8% (2021-22) → 36.7% → 44.6% → 37.9% (2024-25); but n swings 129→246→206 and Success Academy (31 schools, low-CA) vanishes from NYSED data 2023-24+.‡ The constant panel (n=183) reads 44.1 → 38.7 → 43.2 → 38.0 — the 2023-24 worsening is broad-based (58.5% of same-school pairs rose), is NYSED-only, and cannot be verified against a second publisher. COVID-era charter HS values additionally inherit the unexplained NYSED HS tail.§ [Fig 5; A.21, ‡ = A.13, § = A.20]
School-level CA is signal, not noise. Consecutive-year correlation 0.892 → 0.928 → 0.936; median year-over-year |change| 2.9pp — but small schools are ≈2× as volatile (SD of change 6.8pp at 100–249 students vs 3.3pp at 1000+), so single-year swings at small schools need a size-aware gate. [Fig 7; A.15]
New: CA is about as stable as test proficiency — and pseudo-cohorts carry their attendance into the next grade. On identical school sets, consecutive-year r runs CA 0.917 → 0.930 → 0.943 vs ELA 0.925 → 0.941 → 0.894 and Math 0.932 → 0.935 → 0.920— raw levels, not value-added, so the literature’s attendance-beats-tests stability gap does not transfer. Within schools, the cohort correlation (grade g → g+1, next year) exceeds the same-grade correlation in 31 of 36 transition × year-pair cells (g3 2022-23 → g4 2023-24: 0.862 vs 0.758); the exceptions are late-HS transitions. [Figs 8–9; A.22]
Trend hygiene: 2019-20 (COVID-truncated) and 2020-21 (remote attendance) are shown but excluded from all trend statements; every figure marks them. [all figures, †]

* A.11 (publisher-dependent): 2024-25 grades 1–8 are publisher-dependent — NYSED shows this band rising (DOE 30.5→29.0% weighted vs NYSED 30.0→30.8% on the same schools); see the cross-publisher page. Carry source attribution on any 2024-25 elementary/middle improvement claim.

2. Figures

Figure 1. Citywide chronic absenteeism, 2018-19 → 2024-25: weighted rate over the school-level distribution

Line: student-weighted citywide CA. Dashed: median school. Bands: school-level p25–p75 (dark) and p10–p90 (light). The entire distribution shifted up after COVID and has only partly come back.

† Hatched years excluded from trend interpretation: 2019-20 was truncated by the COVID closure; 2020-21 counted remote attendance. Source: local replica of NYC DOE InfoHub attendance data; district universe (DOE public schools in default comparisons; no charters/D75/D79/alt), subgroup ALL, non-suppressed; n schools shown per year. Percentiles unweighted across schools. Series: appendix A1 / citywide_trend in the packet JSON.

Figure 2a. The grade U-shape: weighted CA by grade — baseline, peak, and latest year

PK and grade 12 are always highest; grades 5–6 lowest. The U held its shape through the pandemic; the whole curve moved up and has settled back unevenly.

* 2024-25 grades 1–8 are publisher-dependent — NYSED shows this band rising; see A.11. Open markers = A.11-affected values. Source: per-grade CA keys, district universe, subgroup ALL, non-suppressed; 2024-25 per-grade n students range 32,289 (PK) to 74,196 (gr 9). Series: appendix A2 / per_grade in the packet JSON.

Figure 2b. Recovery scorecard by grade: 2024-25 minus 2018-19 (pp)

Grade 12 is the only grade fully back to baseline (−0.1pp). Grades 1–8 (orange) remain +7–9pp elevated — the most persistent residual damage is in elementary and middle grades.

* 2024-25 grades 1–8 are publisher-dependent — NYSED shows this band rising; see A.11. Under NYSED’s reading these bars would be larger, not smaller — the “younger grades stuck” finding is robust to the publisher question; exact magnitudes are not. Bars are the recovery_vs_2018_19 fields committed in the packet JSON (changes computed at full precision, then rounded once).

Figure 3. Grade-band trends: high schools nearly recovered, K8 schools worst off

Weighted CA by school grade band. HS fell the most from the peak (−8.9pp) and sits closest to baseline (+3.3pp); K8 (+9.5), MS (+8.5) and ES (+7.1) remain far above 2018-19.

† Hatched years excluded from trends. K12 band (~46 schools) omitted, so bands do not sum to citywide. N schools per band (2024-25): ES 638, MS 238, K8 164, HS 406. Source: district universe, subgroup ALL, non-suppressed, denominator-weighted. Series: appendix A3 / grade_band_trend in the packet JSON.

Figure 4. CA vs attendance (ADA): the same loss, amplified ≈ 4×

Change vs 2018-19 in percentage points, replica weighted series. ADA never fell more than 3.3pp; CA rose 13.7pp — because CA is a threshold measure (share of students missing ≥10% of days), it amplifies a small average-attendance shift into a large headline change.

Replica vs official citywide:
Series	2018-19	2021-22	2024-25
Official citywide CA %	26.5	40.3	33.3
Replica weighted CA %	25.8	39.5	32.3
Official citywide ADA %	91.5	88.1	89.8
Replica weighted ADA %	92.0	88.8	90.4

† Hatched years excluded from trends. Replica figures run 0.7–1.0pp below official CA (and ~0.5pp above official ADA) because the replica aggregates school-level rates over the default district universe — excluding D75 (weighted CA 52–68%), D79 and alternative programs, plus suppressed cells — while official citywide rates cover all DOE-run programs. School-level CA–ADA correlation: r −0.907 to −0.948 every year (appendix A4).

Figure 5. Charter schools (NYSED basis) next to the district series — a different measuring stick

New since memo 01: charter chronic absenteeism, from NYSED, on a grades-1–12, enrollment-denominator basis. Two district lines: school-wide PK–12 (slate — the series used everywhere else on this page) and a grades-1–12 re-aggregation (cyan, rev 2026-06-11) that matches the charter grade coverage — compare the charter lines against the district 1–12 line. Solid orange: all charters reporting each year. Dashed: constant panel of 183 charters present in all four years 2021-22–2024-25 — the honest trend line.

Charter caveats — read before citing any charter number:

Different basis. NYSED grades-1–12 rates with enrollment denominators are not level-comparable to the DOE school-wide series (which includes PK/K — high-CA grades — and uses register denominators). The district grades-1–12 line removes the grade-coverage difference (worth 1.0–1.7pp in trend years) — use it for the charter comparison; the denominator difference remains, so treat levels as approximate.
Composition shifts. Reporting n swings 129 (2017-18) → 246 (2021-22) → 206 (2024-25) with suppression heaviest pre-COVID — year-over-year comparisons of the all-reporting line are not apples-to-apples.
Success Academy hole (A.13). SA stopped reporting attendance to NYSED network-wide in 2023-24 (31 schools, ~14% of 2022-23 charter enrollment, weighted CA 24.7 vs 38.7 for other charters) — its exit alone mechanically raises the all-reporting line ~2pp in 2023-24.
The 2023-24 bump survives in the constant panel (38.7 → 43.2 → 38.0) and is broad-based (58.5% of same-school pairs worsened, median +1.3pp) — but it is NYSED-only, its cause is not established, and there is no second publisher to verify charters against (A.21).
A.20. Charter HS values for 2020-21–2022-23 inherit whatever affected NYSED’s HS data in those years (the unexplained one-sided HS tail vs DOE on district schools).

† Hatched years: 2019-20 has no NYSED data (and is COVID-truncated); 2020-21 counted remote attendance. Charter universe: school_type = CHARTER, subgroup ALL, non-suppressed; n shown under each point. Constant panel: non-suppressed in all of 2021-22–2024-25 (Success Academy excluded by construction); interior panel labels omitted for legibility — full series (44.1 → 38.7 → 43.2 → 38.0, and all-reporting 2022-23 = 36.7) in the caveat box and appendix A5. District grades-1–12 line: sum of per-grade numerators/denominators over grades 1–12 (appendix A9). PK/K inclusion adds 1.0–1.7pp to the school-wide line in trend years (2024-25: 32.3 school-wide vs 31.2 grades 1–12, +1.1pp); the residual basis difference vs charters is the denominator (DOE attendance register vs NYSED enrollment), so levels remain approximate even on matched grades. 2024-25 grades 1–8 carry the A.11 asterisk.

Figure 6. The distribution shift: school-level CA, 2018-19 vs 2024-25

Overlaid histograms (5pp bins). In 2018-19 the mode sat in the 10–30% range; by 2024-25 the mass has moved right and a thicker right tail remains: 231 schools above 50% CA vs 137 pre-pandemic (peak: 520 in 2021-22).

Schools above 50% CA, by year:
	2018-19	2019-20†	2020-21†	2021-22	2022-23	2023-24	2024-25
Schools above 50% CA	137	77	205	520	332	277	231
% of universe	9.3%	5.2%	14.0%	35.5%	22.7%	19.0%	15.9%

† COVID-measurement years. Source: district universe, subgroup ALL, non-suppressed school-level CA values; bins closed on the left. Full bin counts in appendix A6.

Figure 7. Year-over-year stability: school CA is mostly signal — but noise scales with size

Left: each district school’s CA in 2023-24 vs 2024-25 (r = 0.936, n = 1,450); most schools sit just below the no-change line (median change −1.4pp, median |change| 2.9pp). Right: small schools are ≈2× as volatile as large ones.

Source: district universe, subgroup ALL, schools non-suppressed in both years. The <100-student bucket (n=4) is omitted from the bar chart as unreliable. Consecutive-year r for earlier pairs: 0.892 (2021-22→2022-23), 0.928 (2022-23→2023-24); long arc 2018-19→2024-25: r = 0.777. Reused unchanged from memo 01 (recomputed and matching): 58 schools sat in the worst within-year decile and 83 in the best in all four post-COVID years; the worst 2022-23 decile improved 3.4pp more than average by 2024-25 (regression to the mean). A 5pp single-year swing at a 150-student school is within ~1 SD of noise; the same swing at a 1000+ school is ~1.5 SD. (A.15)

Figure 8. Stability benchmark: CA is about as stable year-over-year as test proficiency — not more (rev 2026-06-11)

On identical school sets (schools with CA, ELA and Math values in both years of each pair), school-level CA is roughly as stable as test proficiency — r 0.917–0.943 vs 0.894–0.941 — and is the most stable of the three metrics only in the latest pair. These are raw levels, not value-added: the literature’s stability gap favoring attendance (Liu 2022: attendance value-added r = 0.878 vs test value-added 0.774) concerns value-added estimates and does not transfer to levels, which are dominated by persistent school composition.

Consecutive-year school-level Pearson r, same school set per pair:
Pair	CA r	ELA proficiency r	Math proficiency r	N schools
2021-22 → 2022-23¶	0.917	0.925	0.932	1,024
2022-23 → 2023-24	0.930	0.941	0.935	1,038
2023-24 → 2024-25	0.943	0.894	0.920	1,046

¶ The 2021-22 → 2022-23 pair spans the NYS test recalibration (NextGen-standards scale introduced in 2022-23) — levels are not comparable across that pair even though the school-level correlation survives; read it with care. Tests were not administered in 2019-20/2020-21, so these three pairs are the only consecutive pairs in the analysis window (2018-19 test data exists but has no usable consecutive partner; 2017-18 predates the window). Source: ELA/Math All-Grades % Level 3+4 computed from the ingested InfoHub test files with the loader’s parsing rules; district universe, subgroup ALL, non-suppressed. Each metric’s own (unrestricted) both-year set gives nearly identical r — appendix A10.

Figure 9. Pseudo-cohorts vs the grade environment: cohorts carry their attendance into the next grade (rev 2026-06-11)

For each grade transition, two within-school correlations: the cohort path (grade g CA this year vs grade g+1 CA next year — following the same student body) and the same-gradepath (grade g vs grade g — the school’s grade environment with a new student body). The cohort correlation is the higher one in 31 of 36 transition × year-pair cells — by ~+0.04 to +0.08 (medians across pairs) at every transition from K→1 through 9→10 except the structural 8→9. The user-flagged example: grade-3 2022-23 → grade-4 2023-24 cohort r = 0.862 (n = 772, median |change| 5.0pp, cohort-size churn >20% in 6.6% of schools) vs same-grade r = 0.758. Both correlations are high everywhere (≈0.74–0.89): most persistence is school-level; the consistent cohort margin is the part that travels with the cohort. The exceptions are late high school — 11→12 in all three pairs (grade 12 runs its own dynamics) plus one 10→11 and one 8→9 cell.

School-level pseudo-cohorts, not linked students — individual students move in and out even where cohort size is stable, and school-level data cannot separate cohort composition from other mechanisms, so “carry” describes the correlation pattern, not an established mechanism. ° = structural school-change transitions (5→6, 8→9): only the minority of schools spanning both grades enter (n 80–141vs ~770 at elementary transitions) and cohort-size churn >20% hits 44.0–71.3% of them, so those cells describe an unusual subset. Elsewhere churn >20% is rare (e.g. 6.6% at g3→g4 2022-23→2023-24). Chart shows medians across the three post-COVID pairs; full matrix (per-pair r, n, median |change|, churn) in appendix A11. K→1 included, PK excluded; pairs involving 2019-20/2020-21 excluded per trend conventions. 2024-25 grades 1–8 values carry the A.11 asterisk.

3. Methods & caveats

Data. All numbers computed read-only from the local Postgres replica (edu_data_local), built by the repo’s own loaders from the official DOE InfoHub attendance file (district) and NYSED (charters); cell-level fidelity of the replica was verified previously. The committed companion file data/analysis/absenteeism-descriptives.json contains every series shown.

District universe. school_type != 'CHARTER' with include_in_default_comparisons = true (equivalently DOE public schools; excludes charters, D75 special-ed, D79 transfer, alternative programs), subgroup ALL, non-suppressed rows with non-null value and denominator. N = 1,450–1,470 schools per year; counts reported on every figure/table. Year-over-year comparisons are repeated cross-sections, not a fixed panel (except where labeled).

Charter universe and basis difference. school_type = 'CHARTER'; values are NYSED-based, grades 1–12 only, denominator = enrollment. The district series includes PK/K (the highest-CA elementary grades) and grades from DOE attendance registers — so charter and district levels are not comparable; only within-series shapes are meaningful. Charter reporting n varies 129→246→206 across years (suppression, openings, A.13), so the all-reporting series confounds trend with composition; the constant-panel series (n=183, present all of 2021-22–2024-25) is the honest trend line. Success Academy reports no NYSED attendance from 2023-24 (A.13). Charter HS 2020-21–2022-23 inherits the unexplained NYSED HS tail (A.20) and no second publisher exists for charters.

District grades-1–12 series (rev 2026-06-11, Figure 5). Re-aggregation of the per-grade CA keys (chronic_absenteeism_rate_grade_1..12): estimated absences = Σ value/100 × denominator, N = Σ denominator, over the district universe — aligning grade coverage with the charter NYSED basis. It is not a perfect school-wide complement: ~3.9% of grade-level cells are suppressed (small grades dropped), and the denominator is still the DOE attendance register, not NYSED enrollment. The gap to the school-wide line (1.0–1.7pp in trend years) is therefore the PK/K contribution plus a small suppression residual. Full series: appendix A9 / district_grades_1_12 in the packet JSON.

Test-score stability benchmark (rev 2026-06-11, Figure 8). The local replica carries no test rows, so ELA/Math All-Grades % Level 3+4 were computed directly from the ingested InfoHub files (data/reference/sources/nycdoe_infohub_ela|math/2026-06-01, the vintages whose DB load was base-case-verified exact) using the test loader’s own parsing and suppression rules, restricted to the district universe, subgroup ALL. Per pair, one school set — schools with CA, ELA and Math in both years — so the three r values are computed on identical schools. Correlations are on raw levels; no value-added or demographic adjustment. The 2021-22 → 2022-23 pair spans the NextGen recalibration (flagged ¶). As a check, the school-wide CA consecutive r on the full universe reproduces the committed values exactly (0.892/0.928/0.936).

Pseudo-cohorts (rev 2026-06-11, Figure 9).Within school, cohort r correlates grade-g CA in year y with grade-(g+1) CA in year y+1 across schools with both cells non-suppressed; same-grade r correlates grade g with itself across the same year pair. These are school-level aggregates, not linked students. Cohort-size churn (|ΔN|/N > 20% on metric denominators) flags school-cohorts where the student body changed substantially; outside the structural 5→6 / 8→9 transitions it is rare. The memo-01 spot value (g3 2023-24 → g4 2024-25: r 0.868, n 770, median |change| 4.4pp, churn 5.2%) reproduces exactly as one cell of the matrix.

Weighting.“Weighted” = denominator-weighted (student-weighted) aggregation of school-level rates. Percentiles, SDs, histograms, correlations and change distributions are unweighted across schools, per project convention. Weighted and unweighted means diverge (32.3 vs 35.7 in 2024-25) because smaller schools tend to have higher CA — both are reported.

Excluded years. 2019-20 (year truncated by the March 2020 closure) and 2020-21 (remote attendance counted) are plotted for completeness but excluded from all trend statements; every figure hatches them. Trend language compares 2018-19 → 2021-22 → 2024-25.

Publisher dependence (A.11). 2024-25 grades 1–8 CA is publisher-dependent: DOE shows the band falling while NYSED shows it rising on the same schools with near-identical denominators; document review rules out any documented definition change and the cause is not established (NYSED inquiry pending). All 2024-25 grades 1–8 values here are DOE-based and carry the asterisk; HS publisher agreement is excellent, so HS findings are unaffected.

Replica vs official citywide.Our “citywide” is a school-level aggregation over the default district universe; DOE’s official citywide rates cover all DOE-run programs (incl. D75/D79/alt, which run far higher CA). This explains the ≤0.8pp CA and ≤0.6pp ADA gaps in Figure 4’s table; the replica is internally consistent and preferred for school-level analysis.

Cross-checks vs memo 01. Every statistic shared with docs/analysis/absenteeism/01_descriptives.md was recomputed: 168 checks, 0mismatches > 0.1pp (row-level log in the packet JSON under crosschecks_vs_memo01). Underlying levels match exactly. Derived change figures on this page are the change fields committed in the packet JSON, computed from full-precision levels and rounded once (MS band recovery +8.5, grade-9 recovery +4.3); the packet HTML’s prose quotes two of these 0.1pp differently from pre-rounding arithmetic — the JSON values are canonical here.

Precision. Rates shown to 1dp (4dp in the JSON); correlations to 3dp; percentiles via linear interpolation (numpy default). Suppressed cells are excluded wherever they occur, which removes small cells and can slightly bias small-grade estimates.

4. Numbers appendix

Every figure’s underlying series. The same numbers (plus full precision and the 168-row cross-check log) are in data/analysis/absenteeism-descriptives.json. † = COVID-measurement year, excluded from trends. * = A.11 publisher-dependent (2024-25 grades 1–8). ¶ = pair spans the NextGen test recalibration. ° = structural school-change transition. A9–A11 are rev 2026-06-11 additions (see meta.revisions in the JSON).

A1. Citywide CA by year (Figure 1)
Year	Weighted CA %	Unweighted school mean	p10	p25	p50	p75	p90	SD	N schools	N students
2018-19	25.8	29.0	10.5	17.1	27.9	39.7	49.2	14.7	1,469	957,623
2019-20†	24.2	27.3	10.8	16.8	25.9	36.7	45.8	13.3	1,470	930,492
2020-21†	28.5	30.9	8.4	17.2	30.1	43.3	53.1	17.2	1,465	881,538
2021-22	39.5	43.1	20.1	30.7	43.6	55.4	63.5	16.3	1,463	860,089
2022-23	35.1	38.7	19.7	28.5	38.4	48.7	57.3	14.2	1,462	852,437
2023-24	33.8	37.2	18.1	26.7	36.9	46.7	56.7	14.5	1,459	853,621
2024-25	32.3	35.7	16.8	25.4	35.7	45.7	54.1	14.1	1,450	836,268

A2. Weighted CA by grade (Figures 2a/2b)
Grade	2018-19	2021-22 (peak)	2024-25	Chg vs 2018-19 (pp)	Chg vs peak (pp)	N schools 2024-25	N students 2024-25
PK	43.5	60.3	48.2	+4.7	−12.2	632	32,289
K	31.6	45.8	37.8	+6.3	−8.0	782	57,489
1*	25.6	39.9	33.2	+7.6	−6.7	786	58,588
2*	22.7	36.8	30.5	+7.8	−6.3	779	58,280
3*	21.0	34.4	28.1	+7.1	−6.3	780	59,764
4*	19.7	33.1	27.3	+7.6	−5.8	779	59,025
5*	19.3	32.9	26.6	+7.3	−6.4	779	59,349
6*	18.1	33.2	26.7	+8.6	−6.5	440	52,892
7*	19.8	33.3	28.7	+8.9	−4.6	455	59,531
8*	23.1	35.7	31.2	+8.1	−4.5	451	59,134
9	29.3	39.1	33.6	+4.3	−5.5	411	74,196
10	30.4	41.7	33.7	+3.3	−8.1	408	70,749
11	28.1	41.8	31.5	+3.4	−10.3	402	62,368
12	40.2	53.6	40.1	−0.1	−13.5	399	59,385

A3. Weighted CA by grade band (Figure 3)
Year	ES weighted CA % (n)	MS weighted CA % (n)	K8 weighted CA % (n)	HS weighted CA % (n)
2018-19	24.6 (643)	20.7 (244)	23.7 (164)	30.8 (414)
2019-20†	23.4 (644)	18.5 (244)	22.1 (164)	28.8 (414)
2020-21†	25.5 (641)	27.2 (244)	25.7 (164)	33.8 (412)
2021-22	38.8 (641)	34.0 (243)	38.8 (164)	43.0 (411)
2022-23	35.7 (640)	30.7 (243)	36.1 (164)	36.0 (411)
2023-24	33.7 (639)	30.2 (241)	34.9 (164)	35.3 (411)
2024-25	31.7 (638)	29.2 (238)	33.2 (164)	34.1 (406)
Chg 2024-25 vs 2018-19 (pp)	+7.1	+8.5	+9.5	+3.3
Chg 2024-25 vs 2021-22 peak (pp)	−7.2	−4.9	−5.6	−8.9

A4. CA vs ADA, replica and official (Figure 4)
Year	Replica weighted CA %	Replica weighted ADA %	School-level r (CA, ADA)	R²	N schools	Official citywide CA %	Official citywide ADA %
2018-19	25.8	92.0	−0.917	0.841	1,469	26.5	91.5
2019-20†	24.2	92.3	−0.918	0.843	1,470	—	—
2020-21†	28.5	90.2	−0.948	0.898	1,465	—	—
2021-22	39.5	88.8	−0.923	0.853	1,463	40.3	88.1
2022-23	35.1	90.0	−0.909	0.826	1,462	—	—
2023-24	33.8	90.2	−0.913	0.834	1,459	—	—
2024-25	32.3	90.4	−0.907	0.822	1,450	33.3	89.8

A5. Charter series, both variants (Figure 5)
Year	All-reporting weighted CA %	N charters reporting	N students (enrollment)	Constant-panel weighted CA %	Panel n	Success Academy schools reporting
2017-18	12.4	129	62,854	—	—	1
2018-19	17.2	164	86,213	—	—	1
2020-21†	28.7	241	114,616	—	—	30
2021-22	41.8	246	125,208	44.1	183	31
2022-23	36.7	239	120,902	38.7	183	31
2023-24	44.6	199	102,302	43.2	183	0
2024-25	37.9	206	106,387	38.0	183	0

Charter basis: NYSED, grades 1–12, denominator = enrollment. NYSED published no 2019-20 data. Constant panel = 183 charters non-suppressed in all of 2021-22–2024-25. Success Academy ran ~14% of 2022-23 charter enrollment at weighted CA 24.7 (vs 38.7 for other charters) before leaving the NYSED data (A.13).

A6. School-level CA histogram counts, 5pp bins (Figure 6)
CA bin (%)	N schools 2018-19	N schools 2024-25
0–5	30	3
5–10	108	33
10–15	170	77
15–20	157	100
20–25	158	134
25–30	180	171
30–35	161	181
35–40	148	193
40–45	124	180
45–50	96	142
50–55	79	107
55–60	26	60
60–65	24	43
65–70	5	17
70+	3	9

A7. Year-over-year stability (Figure 7)
Statistic	Value
Pearson r, 2021-22 → 2022-23	0.892 (n=1,462)
Pearson r, 2022-23 → 2023-24	0.928 (n=1,459)
Pearson r, 2023-24 → 2024-25	0.936 (n=1,450)
Pearson r, 2018-19 → 2024-25 (long arc)	0.777 (n=1,448)
Median change, 2023-24 → 2024-25	−1.4pp
Median \|change\|, 2023-24 → 2024-25	2.9pp
Improved >2pp / >5pp	42.7% / 19.5%
Worsened >2pp / >5pp	21.1% / 7.4%

A7b. Volatility by school size, 2023-24 → 2024-25 (Figure 7, right)
Size bucket (2023-24 denominator)	N schools	SD of change (pp)	Median \|change\| (pp)	Mean change (pp)
<100	4	4.8	2.8	−2.5
100-249	186	6.8	4.4	−1.0
250-499	612	5.3	3.2	−1.2
500-999	498	4.5	2.7	−1.8
1000+	150	3.3	1.7	−1.0

A8. Cross-check log vs memo 01

168 statistics recomputed against memo 01 (citywide table, grade-band table, CA/ADA table, per-grade table, stability headline numbers): 0mismatches > 0.1pp. Full row-level log in the committed packet JSON under crosschecks_vs_memo01.

A9. District grades-1–12 weighted CA vs school-wide (Figure 5; rev 2026-06-11)
Year	Grades 1–12 weighted CA %	School-wide weighted CA %	Gap (PK/K + suppression, pp)	N schools	N students (grades 1–12)
2018-19	24.7	25.8	+1.0	1,469	859,714
2019-20†	23.0	24.2	+1.2	1,470	833,924
2020-21†	28.1	28.5	+0.4	1,465	796,316
2021-22	38.1	39.5	+1.3	1,463	769,552
2022-23	33.4	35.1	+1.7	1,462	761,918
2023-24	32.5	33.8	+1.3	1,459	761,667
2024-25	31.2	32.3	+1.1	1,450	746,490

A10. Stability benchmark: CA vs ELA/Math proficiency (Figure 8; rev 2026-06-11)
Pair	Same school set: CA r / ELA r / Math r (n)	Unrestricted: CA r (n)	Unrestricted: ELA r (n)	Unrestricted: Math r (n)
2021-22 → 2022-23¶	0.917 / 0.925 / 0.932 (1,024)	0.892 (1,462)	0.926 (1,080)	0.932 (1,029)
2022-23 → 2023-24	0.930 / 0.941 / 0.935 (1,038)	0.928 (1,459)	0.941 (1,089)	0.935 (1,043)
2023-24 → 2024-25	0.943 / 0.894 / 0.920 (1,046)	0.936 (1,450)	0.893 (1,094)	0.920 (1,054)

¶ = pair spans the NextGen test recalibration (new scale from 2022-23). “Unrestricted” = each metric’s own both-year school set (CA includes schools without tested grades). Raw levels, not value-added.

A11a. Pseudo-cohort vs same-grade Pearson r, full matrix (Figure 9; rev 2026-06-11)
Transition	2021-22 → 2022-23: cohort r (n)	same-grade r (n)	2022-23 → 2023-24: cohort r (n)	same-grade r (n)	2023-24 → 2024-25: cohort r (n)	same-grade r (n)
K→1	0.822 (778)	0.774 (770)	0.834 (773)	0.755 (770)	0.869 (778)	0.795 (774)
1→2	0.852 (773)	0.779 (775)	0.862 (780)	0.787 (780)	0.886 (775)	0.799 (782)
2→3	0.847 (773)	0.787 (778)	0.847 (769)	0.791 (781)	0.881 (776)	0.796 (776)
3→4	0.870 (772)	0.794 (776)	0.862 (772)	0.758 (771)	0.868 (770)	0.790 (772)
4→5	0.838 (771)	0.817 (770)	0.857 (772)	0.813 (769)	0.861 (773)	0.775 (773)
5→6°	0.805 (139)	0.759 (766)	0.800 (141)	0.753 (773)	0.884 (141)	0.789 (775)
6→7	0.837 (419)	0.777 (414)	0.863 (430)	0.795 (408)	0.884 (414)	0.821 (399)
7→8	0.831 (439)	0.772 (438)	0.824 (447)	0.792 (447)	0.852 (445)	0.789 (448)
8→9°	0.794 (80)	0.771 (445)	0.841 (80)	0.785 (447)	0.770 (87)	0.795 (445)
9→10	0.833 (409)	0.784 (409)	0.880 (409)	0.844 (409)	0.891 (405)	0.854 (402)
10→11	0.815 (404)	0.795 (410)	0.864 (408)	0.838 (410)	0.846 (400)	0.853 (405)
11→12	0.736 (400)	0.766 (401)	0.802 (393)	0.822 (402)	0.758 (396)	0.799 (399)

A11b. Pseudo-cohort path detail: median |change| (pp) / cohort-size churn >20% (%) / median cohort N
Transition	2021-22 → 2022-23	2022-23 → 2023-24	2023-24 → 2024-25
K→1	8.8 / 12.0% / 69	9.8 / 10.0% / 69	7.7 / 7.1% / 69
1→2	7.6 / 6.5% / 67	6.8 / 8.2% / 69	5.6 / 5.4% / 68
2→3	7.2 / 8.9% / 68	5.7 / 8.3% / 67	5.2 / 5.2% / 69
3→4	6.2 / 6.1% / 71	5.0 / 6.6% / 68	4.4 / 5.2% / 68
4→5	6.3 / 5.7% / 71	4.6 / 5.3% / 71	4.6 / 4.5% / 69
5→6°	5.9 / 46.0% / 62	5.6 / 55.3% / 64	4.5 / 44.0% / 64
6→7	5.6 / 8.4% / 92	4.2 / 10.0% / 88	3.7 / 4.1% / 92
7→8	5.3 / 5.2% / 95	5.0 / 5.1% / 93	4.7 / 3.1% / 91
8→9°	7.6 / 66.3% / 79	7.9 / 58.8% / 82	6.8 / 71.3% / 82
9→10	5.9 / 11.5% / 118	5.0 / 9.8% / 118	4.8 / 13.8% / 119
10→11	7.1 / 22.0% / 108	5.6 / 20.6% / 110	5.0 / 22.5% / 111
11→12	8.2 / 12.8% / 99	8.9 / 9.9% / 96	9.4 / 8.8% / 100

° = structural school-change transition (few schools span both grades; high churn — those cells describe an unusual subset). School-level pseudo-cohorts, not linked students. Cohort r: grade g year y vs grade g+1 year y+1, same school; same-grade r: grade g both years. District universe, subgroup ALL, non-suppressed; series in pseudo_cohort in the packet JSON.