Scenario models Ohio

Ohio’s FY2027 calculator carries a note on its own simulation sheet:

Note: The statewide average values (such as economically disadvantaged percentage, median weighted wealth per-pupil or federal median income) are NOT recalculated based on the data changes specific to your district.

That is correct behavior for the tool it describes. A user changes one district’s inputs and asks what happens to that district; the statewide average would move by a rounding error, and recomputing it would obscure the answer rather than improve it.

It is not correct behavior here. crates/project::policy moves all 609 districts at once, and under a change of that kind the statewide average is the first thing that would move — it is an average of the thing being changed.

The exposure is concentrated and was measured before the decision was taken. $858m of categorical funding is denominated in the statewide average base cost per pupil: special education, English learners, career-technical and the weighted half of preschool special education are each weight x $8,241.61 x count x state share. Of that, $812.5m is inside [H] Foundation Funding and reachable; the $45.7m weighted half of preschool special education is not.

Sized against a 3.1% input-year refresh, base cost aid carried $113.0m of arithmetic and the reachable categoricals another $25.2m — a fifth again, which is the ratio the decision turns on. Both are statewide totals times a rate rather than model runs: the formula absorbs the per-pupil increase dollar-for-dollar and then the guarantee clips it, so what the same refresh delivers is $169.1m. This paragraph read “moved $113.0m in the model and would have moved about $26.6m more in fact” until #251 — stating a proportional product as a measurement, and pricing the preschool residue the sentence above excludes.

The two largest categoricals are not exposed, and that is a property of the mechanisms rather than a convenience. DPIA divides a district’s poverty share by the state’s; targeted assistance divides by a median. Under a change that moves everything, numerator and denominator move together and largely cancel. Those two are $1.89bn — larger than the exposed programs — which is why the exposure is a quarter and not three quarters.

Model Ohio. Policy::base_cost_scale scales the base-cost-denominated categoricals along with base cost, and the site diverges from the department’s simulator deliberately and says so on the page.

Scale exactly the three programs inside [H] Foundation Funding that are priced in the statewide average: special education, English learners, career-technical — $812.5m, exposed via DistrictRecord::base_cost_denominated_categoricals.

Do not scale gifted. Its $54.4m is mostly unit funding at stated salary prices — $85,776 a coordinator, $89,378 and $80,974 for the two specialist bands. A cost-input refresh would raise those in reality, but by their own amount and not by this factor; nothing in the calculator links them. Scaling them here would be an assumption wearing the shape of an identity.

Emit the figure as a bundle field rather than re-deriving it in TypeScript, so that which programs count is decided once, in project::panel, and the two implementations of apply cannot disagree about it.

Every reference checkpoint was recomputed, and the TypeScript apply() reproduces them. That is what the checkpoints exist for: a divergence between the Rust model and the browser’s reimplementation is exactly the failure this change could have introduced silently.

What a $465.0m computed refresh does, before and after:

computed base cost increase       465.0     465.0
categorical increase                  -      32.1
less ADM denominator mismatch      -7.6      -7.6
less minimum state share          -118.7   -118.7
less guarantee absorption         -141.6   -150.2
delivered                          197.1     220.6
share of what it computes           42%       44%

The floor line is unchanged at $118.7m, which is the check that the decomposition is separated correctly. who_a_change_reaches.rs briefly netted the categorical increase against it and still summed to the right total while reporting a floor $32m too small.

41 districts are lifted off the guarantee rather than 38, and a refresh reaches 356 rather than 353. A bigger increase reaches further, which is the mechanism by which any increase reaches a held district at all.

The incidence across state share stopped being monotone. The fourth band now collects slightly more per pupil than the fifth ($324.06 against $318.42), where the fifth was previously highest. Base cost aid rises with the state share because the state pays the residual; the categorical term does not, because it is paid on a district’s own special education, English learner and career-technical counts and those do not sort by state share. One lever, two mechanisms, two orderings.

A page that confessed a gap now declares a divergence, which is a different kind of claim and a heavier one. It is stated above the controls rather than below the results, and it names the department’s note it is departing from.

$45.7m remains unreached and is not held fixed. Preschool special education’s weighted half is denominated identically but sits outside [H] Foundation Funding, so it is outside categorical_funding and outside everything the scenario computes. Recording it as “held fixed” would misdescribe it: the scenario has never modeled the program at all.

Mirror the department’s tool exactly. The status quo, and it has a real argument: a reader can check this site against the department’s simulator, and a number that reproduces an official one is worth more than a number that is merely more nearly right. Rejected because the two tools answer different questions. The department’s changes one district; every lever here changes all 609, and holding the statewide average fixed under a statewide change describes a world where the department raised the average and then declined to use it in three formulas written in terms of it. Reproducibility against the official tool is preserved where it is meaningful — current law is still exactly the department’s published dollars, to the cent.

Scale gifted too, for completeness. Rejected. It would add about $1.7m to a 3.1% refresh and would substitute a guess for a gap. The distinction between a quantity denominated in base cost and one priced in a salary that happens to move for related reasons is exactly what the separate formula-component nodes were split to preserve, and collapsing it for tidiness would undo that.

Make it a lever the reader can toggle. Considered. Rejected because it offers a choice that is not the reader’s to make: one of the two behaviors is a better description of Ohio and the other is an artifact of a tool built for a different question. A site that cannot say which of its own two answers is right has moved the work onto the reader.

Recompute the poverty share and the wealth median too. Not rejected, unbuilt, and much harder. Those are indices whose effects largely cancel, so the payoff is small; and recomputing them properly means recomputing DPIA’s statewide share and targeted assistance’s median from perturbed district data, which is a different order of change from multiplying by a factor. If it is ever done, the measurement in what_a_scenario_holds_fixed.rs is where the case for it would have to be made.