Special education, and the money outside the operating budget

Correction, 28 August 2026 — read this first

Two growth figures below were derived by comparing actual spending in FY23 to a budget in FY26. Those are different quantities — budgets for some lines run about 7% above what is actually spent — so part of what is described here as growth is really the step from spent to budgeted.

On a like-for-like basis, comparing budgets to budgets (FY25 adopted to FY27 level service), the picture is materially different:

claim below	as written	budget to budget
Special education growth	13.0%/yr	1.9%/yr
Everything else	3.4%/yr	2.6%/yr
Special education share of all growth	51%	19%

What still holds: special education is about 24.5% of the budget on every basis tested; out-of-district tuition is budgeted down 46% for FY27; the circuit breaker account and its balances come from the town's own balance sheet and are unaffected; and the paraprofessional-for-tuition swap is budget-to-budget throughout.

What does not: any statement here that special education is growing far faster than the rest of the budget. On appropriations it is not.

The underlying question — whether actual spending has been rising faster than the appropriations that fund it — is real and is being worked separately. It is not settled, and nothing here should be read as though it were.

Research notes, 27 Aug 2026. Companion to FINDINGS.md, which covers the FY27 appropriation and the override. This file covers two things that document does not: special education as a cost driver, and the school and town funds that sit outside the operating appropriation entirely.

Everything here is reproducible from files under sources/. Where a figure is an inference rather than a published number, it says so.

Part 1 — Special education

1.1 It is a quarter of the budget and it was half of the growth

Totalling every SPED-coded line in data/lps-budget-lines.csv — functions 2110, 2310 (SPED teachers), 2320 (therapeutic), 2325 (SPED subs), 2330 (SPED paras), 2800 (psych), 3300 (SPED transport), 9300 (private tuitions), 9400 (collaborative tuitions):

FY23 → FY26 growth: SPED 13.0%/yr, everything else 3.4%/yr. In dollars, SPED added $1,979,619 of the $3,860,666 total increase — 51.3% of all budget growth was special education, while enrollment fell.

DESE confirms the shape independently (budget-workbooks/dese-all-districts.xlsx). FY18 → FY24, Lunenburg's Other Teaching Services — where DESE books paras, therapists, tutors — grew +51% against +32% for Teachers, on enrollment −2.5%.

1.2 The model in the app escalates the wrong 2.6%

fy28/src/data/model.json carries expenseBase.sped_tuition = 700,142 at 8%. That is $56,011 of the $1,313,594 of year-one escalation the model generates — 4.3%. The other ~$5.75M of SPED is folded into salaries at 4% and transport at 6%.

Observed rates say otherwise. In-district SPED (SPED total less tuition less SPED transport) ran $3,844,889 → $4,592,473 FY23→FY26 = 6.1%/yr. SPED transport ran $320,244 → $649,953 = +103% in four years; the district itself budgeted it at +10% for FY27 while the model blends all transport at 6%.

The concept: the six buckets are modeled as cash-limited — you set an amount and that is what gets spent. SPED is demand-led: the entitlement sets the cost and the appropriation only records it. A demand-led line inside a cash-limited model always understates, because the model cannot represent caseload, which is the thing that actually sets the number.

Escalating SPED at its own observed rates adds roughly $130k–$140k/yr of gap on top of the $552,621 headline — about 25% more hole, compounding.

1.3 The out-of-district cliff

Out-of-district tuition quadrupled in three years, then FY27 budgets it 46% below FY26.

FY26 was itself short. As of the Feb/Mar 2026 workbook, actuals-to-date plus encumbrances were $1,530,182 against a $1,291,293 budget — collaborative tuitions alone had spent $607,587 against a $302,663 line, with $171,965 more encumbered.

So FY27 budgets $700,142 against a committed FY26 run-rate of ~$1.53M. An ~$830,000 exposure in one line — more than the entire annual structural gap — before any circuit breaker offset (see §2.1, which changes this materially).

The district knew. School Committee, 4 Feb 2026: "we currently have students on our radar that may require out of district placement and we have also had students move into the district that require out of district placements." (minutes/text/school-committee/2026-02-04-minutes-7634.txt)

This line also carries a risk shape nothing else in the budget has: tail risk. One new residential placement is $150k–$400k arriving mid-year with no vote and no warning.

1.4 The make-or-buy move — the one real structural bend

FY26 → FY27 level service, two entries that almost exactly offset:

• Out-of-district tuition −$591,151
• SPED paraprofessionals +$530,038 (+39% in one year — the largest single-year increase of any line group in the FY27 budget, larger than the school-side health insurance increase of ~$293k)

Tri-Board minutes, 27 Jan 2026, state the strategy outright: "The School Committee had voted mid-year to add paraprofessionals to respond to individual students with acute needs, partly to avoid far more expensive out-of-district placements." (minutes/text/finance-committee/2026-01-27-minutes-7619.txt)

This is a make-or-buy decision — produce the capability in-house rather than purchase it — and on the model's own logic it is correct: it converts a demand-led line inflating at 8%+ into headcount inflating at contract rates of ~4%. It is the only structural curve-bend anyone in Lunenburg has actually executed, and the app does not mention it.

It is fragile in one specific way: it only works if placements genuinely come back in-district and stay. If a parent rejects the in-district program at a Team meeting, the district pays the tuition and keeps the paras. Downside-only. Worth modeling as a scenario.

1.5 Compliance exposure is already on the public record

Tri-Board, 27 Jan 2026: a Turkey Hill teacher testified that after the FY26 cuts her role was split between SPED and MTSS interventionist, and that "seven of my IEP students with significant pull-out service needs were being seen by a paraprofessional rather than a certified special education teacher."

A para delivering specially designed instruction that an IEP assigns to a certified SPED teacher is a compliance finding waiting for a parent to file, and the remedy is compensatory services — retrospective, unbudgeted, uncapped.

The honest framing: SPED cuts do not reduce spending, they defer it into liability. The cost reappears later, larger, and outside the appropriation.

1.6 The peer comparison omits the fastest-growing SPED cost

budget-workbooks/dese-all-districts.xlsx is Total In-District Expenditures. DESE's in-district measure excludes out-of-district tuition — a separate reporting category. So the app's conclusion #3, comparing Lunenburg's growth to its neighbors, is computed on a basis that omits the line that quadrupled.

The claim survives — in-district Lunenburg grew +21% FY18→FY24 against Groton-Dunstable +34%, Littleton +32%, Leominster +42%, and is bottom-of-pack alongside Wachusett's +20%. But it is an unstated scope limitation on a load-bearing conclusion.

Part 2 — The money outside the operating budget

Sources: budget-workbooks/school-funds-fy26.xlsx (period 13, i.e. FY26 year-end) and munis-ledgers/fund-balances/special-revenue-fy2026-p09.xlsx (as of 3/31/2026). The two cross-validate: School Choice opens at $246,902.71 in both, Athletics at $110,247.89, Gift at $89,822.59.

Column mapping in the special revenue workbook, confirmed against its own totals rows: col8 = beginning balance + col10 = revenue + col13 = expenditure = col15 = ending balance. Balance-sheet sign convention: fund balances are shown negative (credit); figures below are flipped to positive-as-money-available.

2.1 Special Ed Circuit Breaker — Fund 2640

This is the single most important finding in this document.

The district holds $615,301 in a restricted special-education account — 81% of the entire $761,000 FY27 net budget reduction — and had drawn $4,005 from it in nine months. Six-tenths of one percent, in a year the general fund carried $1.29M of out-of-district tuition.

Caveat, and it is a real one. Many districts book the circuit breaker offset as a single year-end journal entry, so a low Q3 draw may not mean a low annual draw. The FY26 year-end figure would settle it; this file stops at March. Get it before quoting the $615,301 as idle money.

Effective reimbursement rate. FY26 receipts of $325,970 reimburse FY25 claims. FY25 out-of-district tuition was $732,298 — a ~45% effective offset, or ~28% if the claim also covered the $434,922 of FY25 SPED transport (eligible at 75% since FY25; see minutes/text/school-committee/2025-05-07-minutes-7207.txt, which records the rate rising from 44%). Applied forward to FY26's $1.29M–$1.53M of out-of-district cost, roughly $575k–$680k of circuit breaker revenue should arrive during FY27.

Every out-of-district figure in the app is gross of this. The model has no offset.

The open question. FY26 gross out-of-district was $1,291,293. Net of a ~45% offset that is ~$710k. The FY27 budget line is $700,142. Close enough to be worth asking the Business Manager directly: is the FY27 tuition line gross, or already net of expected circuit breaker? If it is net, the make-or-buy explanation in §1.4 is wrong and the real story is an accounting change.

Peer datapoint: Groton-Dunstable's FY27 budget book states "The District is planning to offset $2M of expenses with Circuit Breaker Funding ($500K higher than FY26)." (peers/groton-dunstable-fy27-budget-book.txt)

2.2 Athletics revolving fund — the app understates fee revenue by 45%

FY26 year-end, under the old $250 / $140 / $85 schedule with a $475 family cap:

Expenditures $146,911 — salaries (4 revolving-fund staff) $30,514; purchase of service (officials, uniforms, transportation) $113,602; general supplies $2,795. Beginning balance $110,248 → ending balance $152,281.

Three consequences:

1. The app's base is wrong. model.json carries estimatedFy26Revenue: 130129; actual was $188,944, 45% higher. Worse, estimatedAthleticRevenue: 187451 is the app's estimate of revenue after the increase to $400 — but the old schedule already collected more than that. feeIncreaseValue: 77849 and the $960 self-funding fee are both calibrated off a base that is too low.
2. Athletic transportation was already partly running through this fund. The $113,602 purchase-of-service category names transportation explicitly. "We cut all athletic transportation, $127,550" is therefore not the clean statement it appears to be.
3. The fund gained $42,033 in FY26 and ended holding $152,281 — in the same year the town said it could not afford $127,550 of athletic buses. Revolving-fund spending is capped by annual Town Meeting authorisation, so this is not automatically available, but it is a fair question and nobody appears to have asked it.

Also: MS fees brought in $26,074 in FY26, and FY27 eliminated middle school sports. That revenue goes with it.

This answers two of the three questions in model.json → feeAccounting.unresolved.

2.3 School choice revolving fund

FY26 year-end: bus fee $52,717 net + state/local choice aid $63,314 = $116,031 revenue; expenditures $36,653; beginning $246,903 → ending $326,281.

Only 32% of what came in was spent. The fund grew $79,378 in a year the district cut $761,000. School choice funds are restricted to school purposes but are broadly usable within that.

2.4 Everything else on the school side

This is not $1.73M of free money and must never be presented as such. Lunch must stay in food service under federal rules; gifts are donor-restricted; circuit breaker is SPED-only. But it is $1.73M that never enters the budget conversation, against a $761,000 cut — and the two largest pots are circuit breaker (restricted to the exact cost driver eating the budget) and school choice (broadly usable).

This is the mirror image of the app's conclusion #13. That one says one-time money is funding recurring costs. This says recurring restricted money is not being spent at all.

2.5 Town stabilization and trust funds

From munis-ledgers/fund-balances/trust-agency-fy2026-p09.xlsx, as of 3/31/2026:

Stabilization grew $562,740 in nine months of FY26.

Read this carefully and fairly. Stabilization requires a two-thirds Town Meeting vote, and spending reserves on recurring costs is the same error the app already flags at $453,722 — just larger. Strong reserves are also what protects the bond rating.

The defensible observation is narrower: general stabilization ($3.24M) plus certified free cash ($3.354M, from FINDINGS.md) is ~$6.6M against a $49.96M omnibus — about 13%, at the top of the 5–15% range DLS considers healthy. A town at the top of that range cutting $761,000 is making a policy choice, not bowing to arithmetic. That is a fair thing to say. "They have $9.6M and won't spend it" is not.

One genuine oddity: the health insurance stabilization fund holds $11,368 against a $4M health insurance line growing 9%/yr. The fund exists in name only.

Part 3 — Why none of this surfaced during the budget debate

town-budget/docs/fincom-memo-fy26-q3.docx, from the Finance Director, is dated 11 August 2026 and reports the quarter ending 31 March 2026 — a four-and-a-half month lag, and explicitly the first of a resumed series ("In moving forward, I hope to present these reports quarterly to the Finance Committee").

Her own account of why, quoted from the memo: the Town Accountant of 38 years retired at the end of FY24; the successor gave notice after 18 months; the payroll person retired and went to the school; the town lost the Assistant Town Accountant. The Assistant Town Accountant was hired December 2025, the Payroll Benefits Coordinator March 2026, and the Finance Director herself at the end of January 2026.

So the FY27 budget was built in February–March 2026 and the override went to ballot on 16 May 2026 without current quarterly financial reporting, during a near-total turnover of the finance office.

This is not an accusation and should never be written as one. It is the stated record, and it is the most plausible explanation for how $615,301 can sit in a special-education account while the district cuts $761,000 and eliminates middle school sports.

Also in the Q3 report: revenue the FY27 debate did not count

• Local receipts came in at 116% of budget — $3,961,722 against $3,415,624, an overage of $546,098, attributed to "Investment Income and MVE exceeding budget."
• $318,000 of Smart Growth funds had not yet been received as of 3/31/26.
• Property taxes 75% collected, state aid 73% — both on pace.
• General fund expenditures $36,845,129 against a revised $51,531,201 budget (71.5%), including ~$2.6M of encumbrances.
• School Department: revised budget $26,323,868; expended $15,736,641 plus $1,668,043 encumbered = 66.1%; $8,919,184 remaining.

What would change in the app

Recorded, not yet implemented — deliberately.

1. Add a circuit breaker offset to the SPED model. Biggest available correction, and it moves the gap in the favourable direction, which makes the rest harder to dismiss as advocacy.
2. Re-base the athletics fee model on $188,944 actual rather than $130,129 estimated. Several downstream figures move, including the $960 self-funding fee.
3. Escalate SPED at its own observed rates rather than folding it into salaries.
4. Add a "money outside the budget" section — $1.73M school-side, honestly annotated for what is restricted and what is not.
5. Record the make-or-buy move on the bend-the-curve page, with its fragility stated.
6. Note the reporting lag in Context or Structural.

Still missing

• FY26 year-end circuit breaker balance and draw. The one number that would settle §2.1. Everything else here is solid; this is the load-bearing caveat.
• Out-of-district placement counts by year, not just dollars. Dollars ÷ count gives the average tuition, which is what actually escalates.
• FY26 year-end actuals on 9300/9400 — did the $1.53M committed hold?
• In-district SPED enrollment and the district SPED rate against the Chapter 70 foundation budget's assumed rate. Where actual exceeds the assumption the town eats 100% of the difference, and a hold-harmless district never catches up.
• Whether the FY27 tuition line is gross or net of circuit breaker (§2.1).
• Sept 3, 2026 STM warrant and result; FY27 certified free cash post-STM.

Part 4 — What this actually changes in the app

Worked through 27 Aug 2026, before any code changed. Three of the fourteen conclusions move; roughly half the app is untouched.

4.1 The headline gap: a range, not a number, until one figure arrives

The app headlines $552,621/yr average shortfall FY28–FY30. Two corrections pull in opposite directions:

The gap is somewhere between ~$88,000 and ~$688,000 a year — a 7.8× spread — and which end depends entirely on the FY26 year-end circuit breaker figure. In Branch A the accumulated $615,301 is additionally available as a one-time bridge; in Branch B it is float and worth nothing.

This is the single highest-leverage unknown in the project. One question to the Business Manager settles it.

Caveat on Branch A: the recurring offset is the annual receipt, not the balance. Once the district draws it down, the fund runs at steady state and the $615,301 is spent once. The offset is also capped by actual eligible SPED costs and cannot be spent on anything else.

4.2 Conclusions that move

#5 — "Athletics cannot pay for itself once you put the buses back" ($960/season). This is the one that may flip. Actual FY26 collection was $279.85 per HS participation against the model's assumed $214 — a factor of 1.308. Scaling through:

• self-funding fee: $960 → ~$734
• peak revenue at the revenue-maximising fee: $358,380 → ~$468,652

The app's headline says restoring the full $451,830 program is "out of reach at any fee." At $468,652 it is no longer out of reach — just barely reachable, at a fee around $1,185 and a loss of roughly 30% of participants. The conclusion needs rewriting from "impossible" to "possible but self-defeating," which is a different argument. Requires proper recomputation in price.ts — the 1.308 scaling is indicative, not exact, because the demand-dropoff curve interacts with it.

#3 — "Grew 1.08% while neighbors grew 2.9–6.5%." Survives, needs a footnote: DESE in-district expenditure excludes out-of-district tuition by construction (§1.6).

#14 — "Nothing closes the gap without either an override or teachers" (68%). Holds in Branch B. In Branch A, with a $326k–$600k recurring offset plus higher realized fee revenue, it is materially overstated and may be wrong. Do not restate this conclusion either way until §4.1 resolves.

4.3 Conclusions that do not move

#6 through #12 — the entire tax-base and business-formation argument. New growth, commercial share, homes-per-pupil, the $42.6M break-even, business counts. None of it touches SPED or fees. About half the app is unaffected.

#1, #2, #4, #13 hold, and #1 and #2 get stronger: a larger SPED share means a smaller cuttable base, so "cutting every extra buys one year" and "only classroom positions are big enough" are more true, not less.

4.4 Rates and figures in model.json that are wrong

Note chargeableParticipations: 582 excludes middle school, but MS fees of $26,074 were actually collected in FY26. FY27 eliminated MS sports, so that revenue disappears — which the model does not currently show either way.

4.5 Charts and components affected

Deep — engine-level, everything downstream moves: model/rates.ts (the six-bucket escalator), model/engine.ts. SPED currently splits across salaries / transport / sped_tuition at three rates none of which match its behavior. Either a seventh bucket or a SPED-specific rate. Touching this moves YearChart, Magnitude, CutLine, Forever, LevelVsSlope, Walkthrough and the headline automatically.

Direct — fee model: FeeCurve, SportCutter, Athletics, Levers, PriceList, Packages, Recommendation, model/price.ts, model/answers.ts.

Copy only: Peers (add the in-district scope footnote), Conclusions (#3, #5, #14).

New, parallel to the athletics treatment: A SPED analysis section — composition, the make-or-buy move, the out-of-district cliff, the circuit breaker. Plus a "money outside the budget" section for §2.4/§2.5.

Untouched: TaxBase, TaxBaseMix, CommercialTrend, BusinessFormation, GrowthDial, Development, TaxpayerView.

4.6 Order of work, when the time comes

1. Ask the Business Manager the circuit breaker question. Nothing else should be built until §4.1 resolves — it determines whether the app's central claim stands.
2. Re-base the fee model on FY26 actuals. Self-contained, no dependency on step 1.
3. Give SPED its own escalation rate.
4. Add the SPED analysis section.
5. Rewrite conclusions #3, #5, #14 last, once 1–3 have settled the numbers.