Phasing Out the Temporary Transitional Aid Guarantee
RUN. Reproduce with cargo run -p project --bin edfund-project -- --guarantee <rule> [--backstop repealed|rebased]. Every figure below is a simulation, not a forecast: it
re-runs the department’s own FY2027 model with one lever moved, at observed enrollment. Where a
projection is used it is labeled and reported separately, per
skills/project.
scenario/guarantee-phase-out · 6 nodes point here · 6 corrections
This scenario, and Change the formula Contents
A worked scenario is a change argued here in prose, against the act as enacted, with the figures it produced computed by the crates and committed beside it. Change the formula is the live counterpart: the same model, re-run in your browser at whatever lever positions you set, from the same baseline.
Ohio’s FY2027 model pays $7,281 million in state foundation aid to 609 districts. $879
million of it — 12.1% — is the temporary transitional aid guarantee, paid to 294
districts holding just over half the state’s students, holding them at what they received
in FY2020: a Bridge formula year Ohio froze
rather than computed. verified crates/project, which reproduces the department’s model to
the cent
Retiring it is three bills. [K], the formula transition supplement of uncodified Section
265.225, tops a district up to its FY2021 base from a total that already contains the
guarantee — so a guarantee that stops being paid becomes a shortfall [K] makes good. Left
standing it absorbs most of the apparent saving; repealed alongside, it does not; recomputed
against a base with the old guarantee taken out of it, it absorbs a quarter. R.C. 3317.019 and
Section 265.225 are separate instruments in separate law, and a bill could plainly do any of the
three, so all three are priced below. See
FSFP Formula Transition Supplement
for the device.
This scenario generalizes FSFP Cost Input Refresh vs. Freeze from the single district that node refreshes to the whole state.
What this repository computed Contents
Not what Ohio publishes. Figures derived here from committed fixtures, each one citing the test that reproduces it.
Retiring the guarantee is three bills, and they price an order of magnitude apart Contents
Section 265.225 decides what a guarantee retirement costs, and no lever over the guarantee contains the answer. All three, against the FY2027 model:
foundation aid Sec. 265.225 stands repealed alongside recomputed
removal -$879.0M -$79.8M -$942.5M -$253.6M
half phase-out -$439.5M -$70.8M -$503.1M -$210.4M
rebase to 90% -$248.2M -$66.2M -$311.8M -$148.5M
verified Left standing, [K] absorbs $799.1M of the $879.0M a removal takes off
foundation aid, and the three policies stop being three policies: a gross range of 3.54 to
one collapses to 1.21 to one. The second column is current law, and it says a guarantee
retirement is nearly free and very nearly rule-invariant. The third is what retiring the
guarantee means as a policy rather than as a lever. Neither alone is the answer, and the fourth
column is neither of them.
The left column is the one this node published as the cost for eleven phases. It is the right
number for the wrong quantity — what comes off [H] + [I] — and no district’s cheque falls by
it. The measure matters more than the model here: biennium::Measure separates foundation aid
from total state support, and these two columns differ by nothing else.
The third reading: Section 265.225 recomputed against a base that held no guarantee Contents
A legislature that retires [I] and wants to keep Section 265.225 has a third option, and it is
the one this node carried as an open question. [K] tops a district up to [L1], its FY2021
funding base, and the department’s own Introduction sheet says that base “default[s] to the
funding base used for line A Base Cost” — which is to say the guarantee the previous formula
paid sits inside [L1], unallocated. Take it out and the supplement still holds a district at
its FY2021 total, but at a total that never held a guarantee. That saves $253.6M on a removal,
$210.4M on a half phase-out and $148.5M on a rebase to 90%: the middle answer every time,
and between a quarter and a half of what repeal saves — 26.9% of it on a removal, 41.8% on a half
phase-out, 47.6% on a rebase. verified
The reading was refused because [L1] is one published column and the department never breaks
the guarantee out of it. It does not have to — the split is a join, and both halves are openly
served. The FY2019 final settlement payment report carries a per-district TRANSITIONAL GUARANTEE column; the FY2021 funding base is built from the FY2019 total, because FY2020 state
foundation funding was flat to FY2019 for the categories it is built from; and
project::transition_base asserts the identity rather than assuming it, on both sides: the seven
published terms sum to [L1] for all 611 districts, and every district but two reproduces its
FY2019 total to the cent, the two being footnoted restatements the department names. $257.0M
of [L1] is guarantee, and up to 75.4% of a single district’s base. verified
Two things about the fourth column are worth stating plainly, because both are easy to assume
away. It is not a blend of the other two: the three guarantee rules still span 1.71 to
one under it, against 1.21 as enacted and 3.54 gross, so which rule a bill picks still moves the
price. And it is not a guarantee bill: recomputing the base with [I] left entirely alone
already saves $21.3M and takes the supplement from 144 districts to 63, all of it inside
[K] and none of it foundation aid. So the fourth column is two amendments priced together, and
$79.8M + $21.3M is not $253.6M — the interaction is the larger half of it, because the same
dollars cannot come out of the base and out of realized aid twice. verified
Whom it reaches sits between the other two as well. 261 districts cut: short of the 294 the
guarantee pays, because 44 are still held whole by a supplement measured against the smaller
base, and past it, by 11 that were never on the guarantee at all and lose the [K] the published
base was paying them. The supplement ends up paying 263 districts, against the 294 it would
pay on the base as published. verified
A guarantee-only lever reaches 167 districts, and 442 are out of reach Contents
Removal, a half phase-out and a rebase reach exactly the same districts — a floor that stops
holding somebody cannot start holding somebody else — and that set is 167 districts, not the
294 the guarantee pays. The other 127 are made whole by [K], and they are precisely the 127
guaranteed districts already drawing it: a district at its [L1] ceiling is held at a total,
so anything taken off its guarantee is put back pound for pound. verified
No policy applied to the guarantee alone reaches the 442 districts it does not decide.
That is this node’s structural claim and the scope is the correction: it was published
as 315 — the districts the formula pays — which left out the 127 the backstop catches. Extend the
bill to Section 265.225 and the reach is 311, past the 294 the guarantee pays, by the 17
districts that draw [K] and were never on the guarantee. verified So the guarantee question
and the adequacy question are disjoint, and the hold-harmless question and the adequacy
question are not: one of the two devices reaches districts the formula pays.
The largest per-pupil loser is a wealthy suburb, and the poorest district loses nothing Contents
This is the finding that inverted. Under a half phase-out East Cleveland City loses $6,489 per
pupil of foundation aid — more than twice the next district, on assessed valuation of $141,623
per pupil against a statewide median of $248,097. Its total state support falls by zero:
[K] holds it at [L1] to the cent, under a half phase-out and under outright removal alike,
where the narrow measure says $12,978 per pupil. verified
The largest real per-pupil loss on a removal belongs to West Geauga Local, at $797 per pupil on valuation of $604,523 — two and a half times the state’s median. Across the set the median loser sits at $318,480 of valuation per pupil and the median district it makes whole at $257,278. verified The backstop therefore filters a guarantee retirement up the wealth distribution: the districts it catches are the poorer half of the guaranteed population, so what is left to cut is the richer half.
That does not overturn the finding that the guarantee is a wealth mechanism — see Temporary Transitional Aid Guarantee, whose own quintile table already had the per-pupil loss rising with wealth. It strengthens it, and it withdraws the exception this node had recorded against it. Whether a district is on the guarantee is predicted by property wealth; how much of a retirement it actually bears is predicted by property wealth too, once the device standing behind it is in the model. The 50 most guarantee-dependent districts still have a median valuation of $349,582, well above the state’s. inference
The guarantee is a shock absorber that wears out, and the backstop bills for it up front Contents
Raising base cost costs the state more per point the higher it goes, because each increment lifts more districts off the guarantee and onto a formula that then has to pay them:
base cost foundation aid [K] clawback total state support on guarantee $/point
+2% +$106.6M -$5.0M +$101.7M 269 $50.8M
+5% +$288.1M -$20.9M +$267.1M 243 $55.1M
+10% +$639.7M -$42.4M +$597.3M 204 $66.0M
+20% +$1,443.3M -$51.4M +$1,391.8M 153 $79.5M
verified The gradient is monotone and steep: the first two points cost $50.8M each and the last ten cost $79.5M each. A fiscal note scoring a large increase by extrapolating from a small one understates the cost, and the error grows with the size of the proposal — which is the direction that matters, since it is the large proposals that get scored this way. The state is buying back its own guarantee, and the price rises as it does.
Here [K] runs the other way from a guarantee cut. Raising base cost raises realized aid,
which reduces the supplement, so the clawback makes an increase cheaper than the foundation
column says — and because it is proportionally largest on the smallest increase (4.7% of +2%
against 3.6% of +20%), it makes the gradient steeper rather than flatter. The warning is
therefore stronger on the measure a cost is quoted in than on foundation aid alone, which is the
one place in this node where the correction went that way. What it is not is half-price: the
first two points are about two thirds the price of the last ten, which was true of the figures
this node published as well and was overstated in the sentence beneath them.
The foundation column moved too, and [K] is not why. It reads +$106.6M at two per cent
where this node published +$95.4M, because base_cost_scale was given the categoricals
priced in base cost on 2026-08-09 — recorded at
“Scenario models Ohio”, forty days before the
backstop was modeled at all. The two corrections run opposite ways and nearly cancel:
+$11.2M of categorical scaling against -$5.0M of clawback, which is the whole of the
$6.3M between the figure this node published and the one it now reports. That is the hazard in
a net figure, and the reason the foundation column is bound here beside the total: a table
carrying only the net could absorb two unrelated corrections and look very nearly unchanged.
It makes the state’s cost predictable — and so does the backstop, on its own Contents
This is the projection leg and is reported apart from everything above. Carrying each district’s enrollment forward to FY2032 by a damped trend from its own FY2024-FY2026 history — statewide ADM falls 1.3% — foundation aid under current law is $7,233M, in a band of $6,958M to $7,582M (±4.3%). With the guarantee removed the same enrollment forecast gives $6,301M, in a band of $5,831M to $6,809M (±7.8%). inference a forecast, with the caveats below
Those are foundation aid, and on total state support the comparison disappears. Current law is
$8,021M at a half-width of 4.23%; the guarantee removed is $7,943M at 4.16% — not
wider, slightly narrower. [K] is itself a floor against a fixed FY2021 total, so it inherits the
absorbing the guarantee gives up: it is $926.8M at FY2032 against $63.6M at FY2027. Repeal
Section 265.225 beside the guarantee and the half-width is 7.77%, which is where the doubling
actually lives. verified
Retiring the hold-harmless system nearly doubles the state’s exposure to enrollment forecast error. Retiring the guarantee alone does not. That is an argument with nothing to do with equity or adequacy — it is a budget-stability instrument, working by transferring enrollment risk from the state’s appropriation to the districts’ own planning — and what is new is that the instrument is not the guarantee. A device paying $63.6m to 144 districts delivers nearly all of it, latently, because what it guarantees is a total. Nobody appears to defend either device on these grounds. inference
18 districts move onto the guarantee between FY2027 and FY2032 under current law, 294 to 312 — enrollment decline moves the count by about three percentage points of the state, because it is already nearly half.
This paragraph read “only two districts, 294 to 292” until the figures were rechecked against
the model. That was wrong at the damping of the day as well: at 0.85 the count went 294 to 360,
and the direction word contradicted its own numbers. It was the one figure on this node that no
test pinned, which is why it could be wrong in two ways at once and stay that way — it is bound
now, as project/districts-on-the-guarantee-fy2032, along with the six band figures beside it
that were in the same position. Every delta on this node is bound now too, which is the other
half of the same lesson and took two more corrections to learn.
Properties Contents
| Name | Phasing out the temporary transitional aid guarantee |
|---|---|
| Question | What does removing, re-basing, or phasing out the temporary transitional aid guarantee do to Ohio's districts — under each of the three things the uncodified section that backstops it could do — and what does raising base cost cost while it is in place? |
| Perturbations | |
guarantee as-enacted -> removed | rebase:<factor> | phase-out:<remaining> backstop as-enacted -> repealed | rebased (Sec. 265.225, against every rule) base_cost_scale 1.00 -> 1.02, 1.05, 1.10, 1.20 minimum_state_share 0.10 as enacted for FY2026-27, not the 0.05 the plan began with phase_in base cost and categoricals separately, both 1.00 at baselineEvery run is a simulation at observed enrollment against the department's own FY2027 model, which project::policy::current_law reproduces to the cent for all 609 districts. Only the FY2032 block projects anything.backstop is a second dial rather than part of the guarantee rule because the two are separate instruments in separate law: R.C. 3317.019 names its own years, Section 265.225 is uncodified law of H.B. 110 extended by H.B. 96, and a bill could plainly do either. Its third value, rebased, keeps the section and recomputes [L1] without the guarantee inside it; that is an amendment in its own right, so it is not free even with the guarantee held at as-enacted. | |
| Input projections | One leg only, and it is reported apart from every other figure. Enrolled ADM is carried to FY2032 by a damped trend (damping 0.30, fitted — see “The fitted damping”) from a rate that is three-tenths the district's own FY2024-FY2026 history and seven-tenths its fourteen-year rate in the Census F-33 panel (see “The shrunk rate”), and the whole formula is re-run at the low, central, and high end of each district's band rather than scaling a central answer — the guarantee is a max, so the aid curve has a kink that scaling cannot reproduce. [K] is a second kink, in the other direction, and is re-derived at each end of the band rather than carried across from the central run.Assessed valuation is held fixed. It cannot be projected: the corpus has one observation per district and Ohio valuation steps on a county reappraisal cycle rather than moving smoothly. Local capacity is 60% valuation, so the local side of every projection here is an assumption. |
| Horizon | FY2027, with one projected leg to FY2032 |
| Results in the repository | crates/project — cargo run -p project --bin edfund-project -- --guarantee <rule> [--backstop repealed|rebased]. Pinned by crates/project/tests/ — the savings table by what_retiring_the_guarantee_actually_saves.rs, and the split [L1] the third column needs by project::transition_base — and by the checkpoints in crates/bundle, which the web layer re-derives and must reproduce before it will render a scenario. The scenario page's own backstop control offers all three readings, and the three columns below are the three positions of that control. Two checkpoints hold the browser to the second and third, both paired with a guarantee removal because at current law the section does not pay out and all three readings agree at zero; the split [L1] the third needs reaches the browser as a per-district column of the feed rather than being re-derived there. |
| Results | |
total state support, 609 districts, FY2027 Sec. 265.225 stands repealed recomputed
removal -$79.8M -$942.5M -$253.6M
half phase-out -$70.8M -$503.1M -$210.4M
rebase to 90% -$66.2M -$311.8M -$148.5M
districts cut 167 311 261
districts unmoved 442 298 348
districts drawing foundation aid, the same three rules: -$879.0M, -$439.5M, -$248.2M. Districts cut on that measure: 294 under all three. The difference between the two blocks is [K].base cost foundation aid [K] clawback total state support on guarantee $/point +2% +$106.6M -$5.0M +$101.7M 269 $50.8M +5% +$288.1M -$20.9M +$267.1M 243 $55.1M +10% +$639.7M -$42.4M +$597.3M 204 $66.0M +20% +$1,443.3M -$51.4M +$1,391.8M 153 $79.5MFY2032 at projected enrollment, foundation aid — current law: $7,233M in a band of $6,958M to $7,582M. With the guarantee removed: $6,301M in a band of $5,831M to $6,809M. The same on total state support: $8,021M at +/-4.23%, and $7,943M at +/-4.16%. With Section 265.225 repealed as well, +/-7.77%. | |
| Confidence | |
This node reports three answers to one question, and the set is deliberate. [I] is a term in [K]'s own subtrahend, so what Section 265.225 does is not a detail of a guarantee retirement — it is most of its cost. Three answers are coherent, and all three are now reported:- (a)Reporting only one of the three re-creates the defect this node was corrected for: (a) alone says a guarantee retirement is nearly free, (b) alone says it saves $942.5M, (c) alone reads as a compromise that is really two amendments, and a reader given any one would be wrong about the others. inference The simulation figures are exact against the department's model and carry no interval. "Current law" at FY2032 is not a law, and this node says it because the band is drawn under that label. R.C. 3317.011, 3317.017 and 3317.0217 each apply only for FY2026 and FY2027 by their own terms, and forty divisions of five further sections hand their values to a General Assembly that has not met. verified The FY2032 figures are this formula continued unchanged — the only tractable assumption for a five-year question about a two-year statute, and not a forecast of what the law will say. Section 265.225 is on weaker ground still: it is uncodified, its newest committed text names FY2022 and FY2023, and it survives to FY2027 only because H.B. 96 extended both devices in one clause. See “The plan expires and the projection does not”. The FY2032 band is not this district's uncertainty. Three observations cannot estimate a variance, so the interval rests on the cross-sectional spread of district growth rates used as a floor; a district with genuinely volatile enrollment has a wider true interval than this reports. FY2026 is itself partly departmental estimate. verified It is a floor, and now a measured one. Backtested against the F-33 panel, realized forecast error is wider than the plus-or-minus one sigma band at every horizon from one year to five, under the shipped method and under the one it replaced — the interval never over-states. But the margin widens with the horizon, from 1.18x at one year to 1.40x at five, because dispersion grows as horizon^0.60 while the band widens as horizon^0.50. Coverage falls from 67.3%, very nearly the 68.3% such a band should hold, to 55.8%. Measured in crates/project/tests/the_floor_the_interval_rests_on.rs.verified And the widening rule was recalibrated, which was the open question here. The square root held 55.8% of forecasts at five years against the 68.3% it claimed; fitted to coverage the exponent is 0.65, which holds it within 2.9 points at every horizon the backtest reaches. Sigma does not move — it is a measured quantity with a stated source, and only the shape in the horizon was wrong. The bands above are the widened ones. Recorded at “The widening rule”. verified The exponent holds to thirteen years, and five was never the panel's limit. The fit's own six origins reach eleven years — FY2013 plus eleven is FY2024 — and the two the panel supports but the fit did not use, FY2011 and FY2012, take it to thirteen. Run there, coverage of the cross-district error stays within 3.0% of the 68.3% the band claims at every horizon from one to thirteen — 66.1% held at one year and 69.4% at thirteen, so the line crosses the target rather than drifting off it — and refitting to coverage over the longer range returns 0.65 again. This node's six-year leg and the feed's ten-year horizon are inside the measured range now rather than past it. Measured in crates/project/tests/the_horizons_the_backtest_stopped_short_of.rs.verified What fails past five years is the center, not the width. Pooled across origins the band's coverage does sink — 64.5% held at one year, 67.1% at five, 60.1% at thirteen, and at its worst 8.9% of forecasts short of the target at eleven — and removing each origin's own mean puts the whole of that back. At a fixed horizon every origin runs the same method over the same districts for the same span, so an origin effect is a year effect, and the years are the shutdown: at five years the four origins landing on FY2020 or earlier average +0.002 to +0.014 while the two landing on FY2023 and FY2024 average +0.035 and +0.038. Cut every target at FY2020 and the pooled band holds within 3.8 points to nine years with nothing removed. sigma is a cross-sectional spread, so it prices cross-district error and a statewide level break is not in it at any exponent.verified And horizon^0.60 was partly the window. The dispersion exponent refits to 0.573, 0.600 and 0.623 over one to five, seven and nine pre-closure years, and 0.650 over the whole reach — it climbs toward the coverage exponent as the range grows. The two targets do differ, because the errors are not normal, but the size of the gap was a property of the five-year fit rather than of the distribution.verified The bias is the binding limit on a long projection, and it is a year effect that no correction keyed on a district can reach. Mean log error reaches +0.056 at the feed's ten-year horizon, so the mean district's point sits about 5.8% high; the statewide total's sits 3.2% high there and was within 0.7% at every horizon to seven years before the closure, with a sign that changes from origin to origin. Sorted on anything a forecast could see at its origin every quarter of districts is forecast high by about the same amount, and a district's long-run rate persists into its next three to nine years at 0.27 to 0.30 — the three tenths the damping already carries. Decaying the rate toward a share of the long-run rate over-corrects at every share above five hundredths, a drift fitted before the closure removes half of what crosses it and puts the total a point low before it, and an asymmetric band holds less than shifting the center by the same amount. Neither the center nor the band is corrected. Measured in crates/project/tests/the_bias_that_belongs_to_the_years.rs.verified The bias is published now, beside the projection and as four numbers per horizon rather than as a correction. Over every scored forecast the mean district's log error runs −0.0071 at one year and +0.0734 at thirteen, while the state total's — the same forecasts summed by size rather than averaged over districts — runs −0.0052 at one year and +0.0569 at thirteen. Restricted to forecasts whose target is FY2020 or earlier the panel reaches nine years, and there the mean district ends at +0.0272 and the total at +0.0171. This node's own projection runs a six-year leg to FY2032, and at six years those figures are +0.0117 for the mean district and +0.0036 for the total — so the FY2032 band above is centered high on both counts, by roughly three times as much on the district average as on the state total, and nothing above is adjusted for it. The same six years across the closure read +0.0237 for the mean district — twice the pre-closure figure at the identical horizon, which is why the population a six-year statement rests on has to be named rather than assumed. The district route's own fan runs this same six-year leg over one district instead of 606, and it states both. Both quantities start under the truth and end over it, and they do not turn together: before the closure the mean district crosses zero at four years and the total not until five, so there is a horizon where the average district is already being over-forecast and the statewide figure is still low. Neither line on either population turns back — the furthest any of the four ever sits from zero is its own last point, 0.0272 and 0.0171 before the closure and 0.0734 and 0.0569 across it — so the drift is not a middle-distance artifact that washes out, and the deepest horizon is the one where a correction would matter most and is trusted least. A level correction sized on one of the two quantities is the wrong size for the other, which is why both are published and neither is applied. Computed by crates/project's backtest::bias_profile and checked against the independent table in crates/project/tests/the_bias_that_belongs_to_the_years.rs, carried on the feed's projection.bias table, drawn on /method beside the coverage profile, and decided at “The bias published beside the point”.138 districts sit at the minimum state share and their response to a base cost increase cannot be computed, because the floor censors the quantity it would be computed from. They are held at the floor, which makes the base cost table a lower bound. inference [K] reaches further than this panel does. H.B. 96 extends the supplement to community schools and STEM schools as well as districts, and every figure here is over the 609 city, local and exempted village districts the FY2027 model carries. A statewide cost of repealing Section 265.225 is larger than the $63.6m column (b) is priced against, by an amount this model cannot bound. verified | |
Where this appears on the site Contents
The pages outside the corpus that link here, and the section of each the link sits in.
Links Contents
| Instance of | Worked Scenario |
|---|---|
| Alters | Base Cost Per Pupil |
| Alters | FSFP Phase-In Percentage |
| Perturbs within | FSFP Base Cost Calculation |
| Perturbs within | Temporary Transitional Aid Guarantee |
| Perturbs within | FSFP Formula Transition Supplement |
| Measured against | Fair School Funding Plan |
| Measured against | Bridge Formula |
| Runs over | Fiscal Year 2027 |
| Runs over | FY2026-27 Biennium |
| Refines | FSFP Cost Input Refresh vs. Freeze |
| Bears on | Equity |
| Bears on | Adequacy |
| Sourced from | FY27 TRAD State Foundation Funding Calculator |
| Sourced from | FY2019 Final Traditional District Foundation Payment Report |
| Sourced from | Foundation Funding Bases — the two bases the phase-in runs from |
Pointed at by
| Fund the Plan and Retire the Guarantee | Priced by |
|---|---|
| FSFP Formula Transition Supplement | mentions |
| Temporary Transitional Aid Guarantee | mentions |
| Enrolled ADM | mentions |
| Base Cost Per Pupil | mentions |
| Minimum State Share | Altered by |
What this node used to say Contents
The corpus is not rewritten to have always been right. Each entry is a claim this node carried, what replaced it, and the thing that settled it.
Correction 1 of 6
It said
“(c) [K] is recomputed against an FY2021 base that excludes [I]. Not reported, and
not for want of a lever: [L1] is a single published column per district, and splitting it
into its guarantee and non-guarantee halves needs FY2021 per-district guarantee payments that
no committed source carries.” — the third of three coherent readings of Section 265.225,
declined on the availability of a source.
It says
Reported: -$253.6M on a removal, -$210.4M on a half phase-out, -$148.5M on a
rebase to 90%. [L1] is still one published column, but it does not have to be split — it
can be joined. The refusal named the wrong year: the FY2021 settlement report carries no
guarantee line at all, while [L1]’s first term is the FY2019 total, and the FY2019 final
settlement report publishes TRANSITIONAL GUARANTEE per district. Both files were openly
served the whole time, so this was never blocked on portal access.
Settled by
The department’s own Introduction sheet in the funding-bases workbook, which says the base
“default[s] to the funding base used for line A Base Cost” and that FY2019 stands in for
FY2020 because state foundation funding was flat between them. Extracted by
connect::fixtures::transition_base, joined and asserted by project::transition_base, and
priced by policy::Backstop::Rebased.
What else it touched
A recorded claim that a source cannot be had is a claim about the world, and this one was wrong in a way no gate could catch: it was true of the year it named and false of the question. The reading it declined is the middle answer on all three guarantee rules, and it carries a finding neither of the other two shows — that recomputing the base is itself an amendment to Section 265.225, worth $21.3M with the guarantee left alone.
Correction 2 of 6
It said
The savings table stated -$879.0M for removal, -$439.5M for a half phase-out and
-$87.9M for a rebase to 90% — with the results property giving -$248.2M for that
same third row — and put 294 districts cut, 315 untouched against every one of them.
It says
Those are falls in foundation aid, not savings, and only for the first two rows. On total state support, which is the measure a cost is quoted in, the three rules save $79.8M, $70.8M and $66.2M with Section 265.225 standing, and $942.5M, $503.1M and $311.8M with it repealed. 167 districts are cut, not 294, and 442 are untouched, not 315.
The -$87.9M was never computed: it is a tenth of $879.0M, reasoned from a tenth off the
floor. The guarantee is a max, so a tenth off the floor pushes 64 districts off it
altogether and takes their whole top-up — 28.2% of the guarantee. The results property’s
-$248.2M was the model’s own answer to that, and right about the fall in foundation aid.
Settled by
crates/project::hold_harmless, which inventoried four hold-harmless devices where the
corpus had recorded one, and PolicyEffect::cost widening to [R] Total State Support so
that [K] is inside the figure a lever is priced at. Pinned in
crates/project/tests/what_retiring_the_guarantee_actually_saves.rs.
What else it touched
The savings table carried no figures: entries, so nothing in this repository could see it go
stale: the crate moved twice and the prose did not move at all, while every gate stayed green.
That is the #120 shape with nothing on the crate side to disagree with. Every delta on this
node is bound now, and the manifest holds them as magnitudes because it cannot hold a sign.
Correction 3 of 6
It said
“No policy applied to the guarantee can reach the 315 districts the formula already pays.” verified — offered as the structural fact underneath every distributional argument about the guarantee, and as grounds that “a proposal framed as fixing the formula by reforming the guarantee changes nothing for half the state”.
It says
True of a guarantee-only lever and understated: the unreachable population is 442, because
[K] makes 127 of the 294 guaranteed districts whole as well. False of the hold-harmless
system — repealing Section 265.225 alongside reaches 311 districts, including 17 that draw
[K] and were never on the guarantee.
Settled by
repealing_the_backstop_beside_it_saves_what_both_cost in crates/project/src/report.rs, and
the_three_answers_reach_167_261_and_311_districts.
What else it touched
The scoped claim is sharper than the unscoped one was, and it changes what the node concludes: the guarantee question and the adequacy question are disjoint, and the hold-harmless question and the adequacy question are not.
Correction 4 of 6
It said
“The largest per-pupil loser is one of Ohio’s poorest districts” — East Cleveland City at $6,489 per pupil under a half phase-out, offered as the exception showing that dependence on the guarantee and wealth are different distributions.
It says
That is its fall in foundation aid. Its total state support does not move at all: [K] holds
it at [L1] to the cent. The largest per-pupil loss on a removal is West Geauga Local at
$797 per pupil, on valuation of $604,523 against the statewide median of $248,097, and the
median district the retirement cuts sits at $318,480 against $257,278 for the median
district the backstop makes whole.
Settled by
hold_harmless::retirement, and
the_largest_per_pupil_loser_is_a_wealthy_district_once_the_backstop_is_in_the_model.
What else it touched
The exception is withdrawn rather than refined, and the wealth finding it was an exception to
is stronger without it: the backstop filters a guarantee retirement up the wealth
distribution. temporary-transitional-aid-guarantee’s quintile table already pointed this
way — Q4 and Q5 losing most per pupil — and the two nodes were in tension until [K] was
modeled.
Correction 5 of 6
It said
“Removing the guarantee nearly doubles the state’s exposure to enrollment forecast error.” inference — from a foundation-aid band of ±4.3% under current law against ±7.8% with the guarantee removed.
It says
True of foundation aid and false of total state support, where the same two runs are 4.23%
and 4.16% — no wider. [K] is a floor against a fixed FY2021 total, so it inherits the
absorbing the guarantee gives up, reaching $926.8M at FY2032. The doubling is recovered
only when Section 265.225 is repealed as well: 7.77%.
Settled by
the_backstop_inherits_the_guarantees_budget_stability, after report::forecast was widened
to accumulate total state support beside realized aid. The six bound FY2032 figures did not
move, because they are foundation aid and [K] is outside it — which is how a node whose
savings table was in one measure and whose projection was in the other stayed green.
What else it touched
The budget-stability argument survives and changes owner. It is an argument for the hold-harmless system, and the device that delivers nearly all of it is the one paying $63.6m to 144 districts rather than the one paying $879.0m to 294.
Correction 6 of 6
It said
The base cost sweep read +$95.4M at two per cent against 272 districts still on the
guarantee, rising to +$1,294.6M at twenty against 156, at $47.7M a point for the
first two and $72.5M for the last ten. The restatement that put [K] into this node
replaced all eight cells and recorded the movement as the backstop’s.
It says
Only part of it was. The foundation column is +$106.6M at two per cent against 269,
and it has read that since 2026-08-09 — forty days before [K] entered the model.
base_cost_scale was given the base-cost-denominated categoricals that day: special
education, English learners and career-technical are each priced as a weight times the
statewide average base cost per pupil, so a lever that moves the average moves them too.
That is $812.5m of exposure, it delivered +$11.2M more at two per cent, and it lifted
three more districts off the guarantee — the same +3 that commit reported for its own
refresh scenario, 41 lifted rather than 38.
[K] then took -$5.0M back out of the same row, in the opposite direction. The $6.3M
between what this node published and what it now reports is the net of two unrelated
corrections, and the first restatement attributed all of it to the second one.
Settled by
git bisect over the 351 commits between the table’s own commit and its restatement, run on
the on guarantee count because that column is decided inside realized aid and so cannot be
reached by a device paid downstream of it. First bad commit 66e2312, whose own message
states the mechanism and whose decision is recorded at
“Scenario models Ohio”. Probing its tree
directly returns $106,645,239.28, to the cent what the crate returns today, so nothing in
the intervening forty days moved the column again.
What else it touched
The conclusions are unharmed — the gradient was monotone on the published figures and is monotone now, and the fiscal-note warning holds on both. What was wrong was the causation, and it was wrong in the entry written to correct exactly this class of error.
The lesson is not about [K]. A table left unbound does not go stale once and get caught;
it accumulates, and the first correction to reach it inherits every earlier drift as though
it were part of what the correction found. This node’s savings tables were unbound for the
same reason and for longer. The foundation column is bound now, separately from the total,
because a figure bound only on the net of two opposing corrections can absorb both and still
match.