NYC Reads — ELA Results Explorer

Interactive dashboard of NYCPS public ELA results (grades 3–8, 2018–2026) for CPRL / the Robertson Foundation ask.

Deliverable: NYC_Reads_ELA_Dashboard.html — one file, 1.5 MB, opens by double-clicking. No server, no install, no network needed (Chart.js and all data are embedded; only webfonts load from Google and fall back cleanly to system serif/sans).

Pages

Page	What it answers
Citywide overview	Where the city stands in 2026 and how far it has moved from a baseline you pick (2023, 2024 or 2025). Level distribution by year and by grade.
District explorer	All 32 districts on any grade band and any student group. Names the districts that improved on both measures and those that got worse on both, each tagged with its NYC Reads wave. Scatter of Δ Level 1 against Δ proficiency with the "improved on both" quadrant shaded, ranked bars, sortable table, CSV.
Boroughs	Borough trends and distributions; every district as a bar coloured by borough; borough × grade matrix.
NYC Reads phases	Elementary Phase 1 vs Phase 2 on tested grades 3–5, the middle-school wave on grades 6–8, group-contrast table, and a placebo check.
Providers & curriculum	Districts grouped by the NYC Reads professional learning provider they work with and by the elementary curriculum they adopted, with ELA results for each group.
Subgroups & gaps	Every reported student group at city / borough / district level, plus five paired proficiency gaps over time.
Test format	The paper-to-computer transition, grouped by the year each grade moved. Each line is dashed with hollow points while those grades were on paper and solid with filled points once they moved, so the switch is visible without annotations.

Every chart has a PNG button; every page has a Download CSV button that exports exactly the view on screen (so results can be dropped straight into a deck).

Phase markers. Time-series charts carry dashed rules at the first year each NYC Reads wave could appear in results. A spring test sits at the end of that school year, so a wave launched in SY 2023–24 first shows up in the 2024 results — the marker goes on 2024, not 2023. Years before any wave reached a tested grade are tinted and labelled "before NYC Reads". Each chart shows only the waves that reach the grades it plots: the grades 3–5 charts carry the elementary waves, the grades 6–8 charts carry the middle-school wave. The test-format page marks the paper-to-computer transitions instead. A Phase markers checkbox on each page turns them off for a clean export.

How the numbers are produced

• The source files publish both counts and percentages. This tool loads the counts and derives every percentage as count / tested. Nothing is copied from a published percentage column, so no figure is a rounded value of a rounded value.
• Grade bands and district groups are aggregated by summing student counts, then dividing — every aggregate is weighted by students tested. District percentages are never averaged.
• Changes are simple differences in percentage points, and are never taken across the 2023 standards change.

Rebuilding

python3 scripts/extract.py   # .xlsx -> data/payload.json, asserts every invariant
python3 scripts/build.py     # template + app.js + payload + Chart.js -> single HTML

extract.py refuses to emit a row that violates its invariants, so a bad build fails loudly rather than shipping wrong numbers.

Verification

Two independent audits, both run against the shipped file with the browser driving the real UI. Each re-reads the original .xlsx files from scratch rather than reusing any intermediate artefact.

python3 scripts/audit_engine.py   # 40,698 combinations, numeric with tolerance
python3 scripts/audit_digest.py   # 40,698 combinations, exact by hash
python3 scripts/audit_render.py   # 8,324 rendered values
python3 scripts/audit_pass2.py    # 4,030 chart, table and CSV values

Audit	Coverage	Result
audit_raw.py	Internal consistency of all 30,728 source rows	counts sum to tested; % = count/tested; L3+4 = L3+L4; max deviation 5.7e-6 pp
audit_engine.py	Every level × geography × grade band × year × student group = 40,698 combinations pulled out of the live page	0 mismatches. Max deviation 5e-7 pp on percentages, 9.8e-7 on mean scale score. Also cross-checks all 28,523 cells against the files' own published % columns (agreement to 4.3e-6 pp, the source's float precision).
audit_digest.py	All 40,698 combinations again, by SHA-256 over the aggregated student counts. Exact integer comparison, no dependence on decimal formatting	identical hash
audit_render.py	9,560 values actually rendered — KPI cards, chart datapoints, table cells, signs and directions of every change — across 130 filter states	0 mismatches
audit_pass2.py	4,062 more: the diverging grade chart, the 2022 reference chart, the district scatter and ranked bars, the borough matrix and district bars, the phase contrast table, the gap chart, the per-grade format table, and all six CSV exports in full	0 mismatches

Claims in the team's working doc were reproduced exactly: citywide Level 1 26.1% → 24.3% (2024→2026); 27 of 32 districts down on Level 1; largest falls in Districts 23, 7, 16 and 5; and 23, 16, 5 confirmed as Elementary Phase 1.

Things the data will not support

• 2023 is a wall. NYSED re-aligned the test in 2023 and rescaled the mean scale score (601.6 in 2022 → 449.6 in 2023). 2022 appears once, as an isolated reference point.
• The three files do not reconcile. In 2026 the districts total 299,906 tested, the boroughs 301,227 and the citywide file 304,450 — D75 and out-of-district placements are attributed differently, and the rule changed in 2025. Each level is read only from its own file.
• 2025 is off-trend everywhere, up sharply then partly reversing in 2026. That is why the baseline is a control rather than a fixed choice.
• Suppression is not only about small groups. Where one category is too small to publish, NYSED also withholds the next smallest so it cannot be recovered by subtraction. This removes some very large cells — citywide Female, all grades, 2025 is suppressed across 149,821 tested students. Affected views carry an explicit warning and lines break rather than bridging the gap.
• No causal claims. Phase groups were not randomly assigned and did not start level. On elementary grades 3–5 the Phase 1 and Phase 2 waves move almost identically (−2.0pp vs −2.4pp from 2023), and the placebo contrast on grades 6–8 is about the same size — so the elementary contrast is not distinguishable from a general difference between those sets of districts. That is reported as found.

Provider and curriculum data

Assignments are for school year 2025-26, the year the spring 2026 test measures, transcribed from the curriculum-and-JESP chart the team supplied on 11 Aug 2026 into data/vendors_sy2025_26.json. JESP is the job-embedded support provider, i.e. the professional learning vendor. K-5 and grades 6-8 are held separately because they differ.

The 2026-27 draft workbook is no longer used for provider or curriculum: those assignments begin after the 2026 test, and they differ from SY2025-26 for 21 of the 32 districts. That workbook also carried superintendent names and 94 email addresses, so it was never committed; extract.py still asserts that no @ can reach the payload.

Unresolved: the supplied chart's own Phase column disagrees with the NYC Reads launch timeline for seven districts (10, 13, 14, 24, 28, 29, 30). The tool keeps the launch timeline, because it is independently corroborated by the "year joined" columns of the 2026-27 workbook for all 32 districts. The disagreement is displayed on the providers page rather than hidden.

Cohort size

The number of students tested is not stable. Between 2024 and 2026 the district totals fell 2.6%, but individual districts range from -15.4% (D05) to +10.0% (D14). Because a district's result is a percentage of whoever sat the test, that is a competing explanation for a measured gain, and it is not a small one: the change in cohort size correlates with the change in proficiency at r = -0.48 and with the change in the Level 1 share at r = +0.60. The 14 districts whose cohort shrank by more than 5% averaged +4.2pp on proficiency against +0.4pp for those that held steady or grew.

The district explorer carries a "Δ tested %" column and a dedicated panel plotting cohort change against proficiency change, so the two are read together.

Not included

Mathematics, Science and charter schools are out of scope. NYC Solves provider and curriculum are carried in the district table for reference but no ELA result is ever grouped by them, since they concern mathematics.

School level. The meeting scoped this to citywide, borough and district for a first cut, so the school file is not loaded. It is present and usable: 1,129 schools over 33,793 rows in its All Students tab, with the same columns plus DBN and school name. Adding a school page is straightforward — the same engine handles it — and would let you see which schools drive a district's movement, which is the natural next question for targeting support.

Layout

NYC_Reads_ELA_Dashboard.html   the deliverable
src/template.html              markup + CSS (CPRL palette, Crimson Text / Hind)
src/app.js                     aggregation engine + all seven pages
scripts/extract.py             .xlsx -> payload.json, with invariant assertions
scripts/build.py               assembles the single file
scripts/audit_*.py             the audits above
data/                          source .xlsx, payload.json, CPRL logo
audit/                         captured dumps from the live page