Technical methodology

The verification story, in full: for the plain-language explainer of each mechanism, see the methodology page.

This project's central claim is not that it has the numbers — the numbers are the state's own, from DESE's published State Aid Notices. The claim is that it can show you how each number is produced and prove the reconstruction against the state's records. Here is what that means, precisely.

What "verified" means, exactly

The verification method is formula-fitting: for each mechanism, a candidate formula (rate × the right count, on the right year's basis) is applied to every district's actual paid dollars, and the residual is measured district by district, year by year. Three classes result, an ordinal confidence scale:

• verified_exact — the formula reproduces every district's payment to the printed-ADM rounding floor (median residual near $0). About 94% of all additive dollars sit here: foundation funding, EETF, ESA (most years), student growth, ALE, declining enrollment, PD, and ELL.
• verified_applied — the statutory per-student rate is confirmed against payments, but one input (a headcount that lives in the notice's DATA section, or a rate table not yet fully parsed) is proxied, so the full per-district fit is still pending.
• observed_actuals — the dollars are real and reconciled, but the allocation formula depends on inputs this project does not hold (per-district assessed valuation, federal hold-harmless factors), so the published figure is reported as-is.

The four load-bearing results

• Every payment line item is classified, none unexplained. All 24 additive state-aid line items across 2019-20–2025-26 carry a class; nothing in the payment record is left undocumented.
• Internal additivity is exact, every year. For every district-year, the sum of the ledger's line items equals DESE's published total state aid to $0.00. That reconciliation is what guarantees the grouping in the ledger neither drops nor double-counts a dollar — it is re-checked by scripts/spot_check.py before any view ships.
• The one-year lag is verified to 0.009% median error across n = 1,372 district-years: this year's funding basis is precisely last year's three-quarter ADM.
• The exact core was independently re-derived by a blind session. A fresh pass — new parser, new formula fits — reproduced the verified_exact classifications with zero cell mismatches (qa/registry_reverification.md). Foundation and the major categoricals also reproduce the figures in the BLR 2026 Adequacy Study to the dollar.

The honest walls

Verification is not complete, and the site says so on every line. The remaining gaps, and the specific document that would close each:

• Teacher Salary Equalization (verified_applied) — the $185 rate reproduces payments, but the allocation formula is absent from the current funding guide; closing it needs an older DESE rule. The appropriation folds into the matrix from FY2027, so the gap is self-limiting.
• Isolated & small-district funding (verified_applied) — the two-tier district-specific formula reproduces the paid dollars for the districts where the notice's item-24 rate field parsed cleanly; the remainder is an item-24 parse edge case on a few of ~14 small districts, not a formula error.
• ESA, 2022-23 only (verified_applied) — the ladder's first transitional year uses intermediate rates not on clean statutory rungs. Dollars still equal rate × count exactly; the rate itself is what's unexplained for that one year. Every other ESA year is exact.
• Bonded Debt Assistance, URT offset, federal streams (observed_actuals) — these depend on per-district assessed valuation and federal factors this project does not hold, so the published dollar is reported directly.
• 2018-19 (fy29) is a separate, pre-2020 vintage (verified_applied) — its foundation chain is now fully itemized against the $6,781 statutory rate for all 235 districts plus the 22 open-enrollment charters that operated that year (charter figures from DESE's 2018-19 mid-year notice; no Final survives). It is shown per district in the drill-down but held out of the statewide sums (FY_MIN = 30, so the trend series begin 2019-20): its charter panel is partial — 22 of 25, the 3 that closed before 2019-20 being unfillable — and mid-year vintage, so its totals are not apples-to-apples with the audited fy30+ full panel.

The 2025-26 preliminary caveat

2025-26 (fy36) figures are DESE's preliminary run (Prelim #4). Where a verified formula applies, it reproduces the preliminary numbers exactly, computed from final enrollment; where it does not, the preliminary published dollar is shown and tagged. The data layer is built so the final notice, when it posts, swaps in cleanly — one regenerate, no changes to any view. One rate caveat worth flagging: the 2026-27 foundation rate ($8,037) excludes employer insurance that Act 909 of 2025 moves off the rate, so it is not directly comparable to 2025-26's $8,162 without adjustment.

Why there is no on-behalf layer for Arkansas (and why Census figures disagree with Arkansas's own)

The revenue picture on this site is four layers — state, required local (the 25-mill URT), additional local, and federal — all from districts' audited annual reports (ASR). It deliberately does not add a fifth "state-paid benefits on behalf" layer. That layer was built, tested, and dropped, because for Arkansas it double-counts.

The tempting source is the U.S. Census F-33 school-finance survey, whose reported "state revenue" for Arkansas runs far above the state's own books. On the 234 districts matched to both sources in 2023-24, F-33 state revenue exceeds ASR state revenue by +$2.20B. In many states that gap is real — the state centrally pays teacher retirement on behalf of districts, money that never touches a district's books. But in Arkansas it is almost entirely a classification difference, not on-behalf money:

• F-33 local revenue is −$1.63B below ASR local — because Census classifies the 25-mill Uniform Rate of Tax (a uniform statewide property tax) as state revenue, while Arkansas books it as local. That $1.63B ≈ the URT, and corr(F-33 local, ASR local − URT) = 0.915.
• F-33 and ASR total revenue differ by only +$0.12B — so there is no large pool of extra state money hiding in the Census totals. Arkansas districts pay their own Arkansas Teacher Retirement System employer contributions (~$0.45B), which already sit inside the audited ASR figures.
• The genuine remainder — state support beyond the audited operating pie — is about $0.5B (~$1,160/pupil), and even that is not cleanly "retirement/insurance": the one truly state-paid benefit, a share of employee health insurance (~$0.15–0.4B), is small and not separately reported.

So adding F-33 state − ASR state as an "on-behalf" layer would have inflated it ~4× by re-counting the URT already shown in layers 1–2. Anyone comparing Census F-33 to Arkansas's own funding figures should expect this state↔local reclassification and not read the gap as hidden benefits. Full reconciliation and re-derivation: qa/registry_revenue_layers.md. (Added 2026-07-21.)

Looking ahead — Act 909 of 2025. From 2026-27, the employer health-insurance contribution moves off the foundation rate and becomes a DESE-paid benefit. At that point Arkansas will have a measurable state-paid insurance amount: the plan is to add it then as a properly-measured "State-paid insurance (from 2026-27)" layer — drawn as a distinct, clearly-labeled cap above the audited total — using published DESE/EBD amounts, never an estimate. It also means the 2026-27 foundation rate ($8,037) is a channel-shift lower, not a cut, and is not comparable to 2025-26's $8,162 without adding the moved insurance back.

Mechanism-by-mechanism

Every state-aid mechanism, its verification class over the panel, and the formula-fit residual where one was measured. Rendered live from data/registry_meta.json.

Reproducing this

Every figure in every view derives at render time from four JSON files under data/, built by the scripts in scripts/ from the upstream mart and the verified registry (data/analysis/registry.csv). scripts/spot_check.py independently re-derives sampled view numbers from the mart and asserts equality — roster membership, $0.00 additivity, JSON == mart totals and line items, statewide == sum of districts — and must pass before any view is declared done. The upstream pipeline and its per-mechanism verification memos (qa/registry_fy*.md) are available on request.