An analysis, written by this project

The annual town reports

The sixteen annual town reports, FY2011–FY2025, every one downloadable with the sha256 of the bytes we hold — and what we read out of them, by category. Says plainly which pages tie to a total the page itself prints and which carry figures nothing has ever checked, because those are not the same thing.

lunenburgbudgetproject.org — written by the Lunenburg Budget Project, an independent tool for residents. Not affiliated with the Town of Lunenburg, the School Committee or the school district. The document this page renders: /docs/analyses/annual-reports.md

annual town reports, FY2011–FY2025

477

pages holding a financial table

208 of 477

pages tied to a total the page prints

23,639

rows read out of them

What this report counts

One row per page of an annual town report that holds a financial table, and the datasets read out of those pages. A page is counted once per report; a figure is counted once per row.

01477pages holding a financial table

All 477 pages holding a financial table have been read.

None is unread, so the work left is checking rather than reading.

Sixteen documents, opened page by page. A page counted here is one that holds a financial table, not every page of the book.

What it rests on `annual-report-pages.csv`, which is generated from what the extractors actually wrote and refuses to publish a count off a stale database.

What it does not show Whether a page holds a table we have not recognised as financial. The classifier reads headings, and the pages whose heading did not classify are filed as `unknown` rather than dropped.

02208pages tied to a printed total

208 of 477 pages tie to a total the page itself prints.

The other 269 carry real figures that no arithmetic has confirmed.

A page is PROVEN only where the table states an identity about itself -- parts summing to a printed total -- and the extract closes it. Most tables here print no total, so there is nothing to close.

What it rests on the `state` column of `annual-report-pages.csv`, set from the proof each extractor recorded rather than from whether it produced rows.

What it does not show That an UNPROVEN figure is wrong. It means nothing has checked it, which is a different claim and the reason the column exists.

0377pages with no check

Appropriations is the least checked: 77 of its 95 pages have no check at all.

It is also the category a resident is most likely to quote.

These pages print what Town Meeting voted, article by article. 17 of them do tie to a printed total; the rest are read and unconfirmed.

What it rests on `annual-report-pages.csv` grouped by subject and state.

What it does not show Which of those pages COULD be checked. A page printing no total can never reach PROVEN however good the extractor gets.

Those are the answers this report could reach from the documents behind it. If the one you came for is not among them, ask us — no name or email needed.

Sixteen documents, 477 pages of financial tables, 23,639 rows read out of them.

The sixteen reports

Every one is downloadable, and every one carries the sha256 of the bytes we hold, so a figure can be traced to the page it came from and the page to the file.

fiscal year	document	size	download	sha256
FY2011	`4117-fy-2011-annual-town-report.pdf`	30.2 MB	download	`551f9edd75d051ae…`
FY2012	`4118-fy-2012-annual-town-report.pdf`	54.0 MB	download	`75d645ee7055f68c…`
FY2013	`4119-fy-2013-annual-town-report.pdf`	36.1 MB	download	`00670845b7d32216…`
FY2014	`4120-fy-2014-annual-town-report.pdf`	5.5 MB	download	`f5ef6e9aec7866f7…`
FY2015	`4121-fy-2015-annual-town-report.pdf`	7.2 MB	download	`712afbb458b2a452…`
FY2016	`4123-fy-2016-annual-town-report.pdf`	5.2 MB	download	`f13af7a7c78d4a14…`
FY2016 addendum	`4122-fy-2016-annual-town-report-addendum.pdf`	2.1 MB	download	`b354a3941e2451f7…`
FY2017	`4124-fy-2017-annual-town-report.pdf`	35.4 MB	download	`4c6fbf70d0aff247…`
FY2018	`4125-fy-2018-annual-town-report.pdf`	7.7 MB	download	`364a1138df754668…`
FY2019	`4126-fy-2019-annual-town-report.pdf`	8.9 MB	download	`220f1c778148feb9…`
FY2020	`4127-fy-2020-annual-town-report.pdf`	19.6 MB	download	`c3bf67f1259a5c19…`
FY2021	`4128-fy-2021-annual-town-report.pdf`	78.8 MB	download	`7ec4740034a2b9bd…`
FY2022	`4129-fy-2022-annual-town-report.pdf`	16.1 MB	download	`d1673d634dba15c2…`
FY2023	`4131-fy-2023-annual-town-report.pdf`	17.4 MB	download	`d273ebb71360e30d…`
FY2024	`4132-fy-2024-annual-town-report.pdf`	49.8 MB	download	`46f14ff8459ec81f…`
FY2025	`4130-fy-2025-annual-town-report.pdf`	6.9 MB	download	`a9ee7b51154d94fd…`

Read the rest of the analysisThe full version4 min

The other findings

0117datasets read generically

17 of the 24 datasets are read by a generic extractor that cannot name its columns.

Read `column_meaning` before quoting a value from one of those.

The generic extractor records a column as `v1`, `v2`, `v3` -- an ORDINAL, the first column of THAT page that held figures. The ruler is rebuilt per page, so the same name means different columns on different pages.

What it rests on the dataset inventory in `scripts/build_archive_guide.py`, read here rather than restated, so the two cannot disagree.

What it does not show Which printed column a given `v1` is on any particular page, except where `column_meaning` names it from an identity the table states about itself.

What is in them, by category

proven means the rows tie to a total the page itself prints. unproven means the rows exist and nothing has checked them — not that they are wrong. refused means an extractor reached the page and wrote nothing.

category	what it answers	pages	proven	unproven	refused
appropriations	what Town Meeting voted to spend, article by article	95	17	77	1
special revenue	the funds that sit outside the budget — grants, gifts, revolving	68	53	15	0
payroll	what the town paid its people, name by name	65	5	59	1
debt	what the town owes and when it falls due	46	14	31	1
trust and stabilization	the reserves, and how much is in each	39	25	13	1
receivables	what is owed TO the town, account by account	28	18	7	3
unknown	a financial table whose heading we could not classify	27	12	15	0
tax collection	what was committed, collected and abated	21	9	11	1
regional school	the assessment from the regional school district	18	2	15	1
balance sheet	what the town held and owed at year end	17	16	1	0
valuation	what the town is worth, by class of property	16	8	8	0
treasurers cash	the cash the Treasurer held	15	14	0	1
capital	the capital projects and what they cost	13	10	3	0
enrollment	how many children are in the schools	6	4	2	0
elections	how the town voted	2	1	1	0
cultural council	the grants the cultural council made	1	0	1	0
total		477	208	259	10

By report

fiscal year	pages	proven	unproven	refused
FY2011	28	10	17	1
FY2012	27	10	16	1
FY2013	29	8	20	1
FY2014	27	14	13	0
FY2015	31	13	18	0
FY2016	41	12	29	0
FY2017	33	14	16	3
FY2018	35	17	18	0
FY2019	38	11	26	1
FY2020	39	16	22	1
FY2021	37	16	21	0
FY2022	33	18	15	0
FY2023	34	15	17	2
FY2024	22	11	11	0
FY2025	23	23	0	0

The datasets read out of them

A dedicated extractor knows what its table means and can check it against a total the page prints. The generic one cannot: its v1…v8 are ORDINALS — the first, second, third column of that page that held figures — and the ruler is rebuilt per page. Read column_meaning before quoting one.

dataset	rows	years	built by	checked	check failed	no check
`report-appropriations.csv`	5,345	2011–2025	generic	311	5034	0
`staff-roster-entries.csv`	3,751	2011–2025	dedicated	—	—	—
`report-gross-wages.csv`	3,547	2011–2025	generic	0	0	3547
`special-revenue-funds.csv`	2,422	2011–2025	dedicated	172	2098	151
`report-elections.csv`	2,012	2011–2025	generic	323	1688	1
`annual-report-receipts.csv`	1,137	2011–2023	dedicated	504	85	548
`staff-roster-counts.csv`	698	2011–2025	dedicated	—	—	—
`report-index.csv`	652	2011–2025	generic	—	—	—
`report-trust-funds.csv`	644	2011–2025	generic	12	418	214
`report-capital-projects.csv`	618	2011–2025	generic	48	544	26
`staff-position-map.csv`	525	—	dedicated	—	—	—
`report-dept-activity.csv`	515	2011–2025	generic	0	0	515
`report-debt.csv`	366	2011–2025	generic	75	99	192
`report-enrollment-mcas.csv`	338	2011–2025	generic	0	0	338
`report-valuation.csv`	321	2011–2019	generic	0	0	321
`report-officials.csv`	237	2011–2024	generic	0	0	237
`report-signatures.csv`	182	2013–2025	generic	—	—	—
`report-filing.csv`	104	—	generic	—	—	—
`report-vital-records.csv`	96	2011–2025	generic	0	0	96
`report-monty-tech.csv`	70	2017–2017	generic	0	0	70
`report-appropriations-supplement.csv`	32	—	generic	—	—	—
`placement-counts.csv`	15	2011–2025	dedicated	—	—	—
`ballot-questions.csv`	9	2012–2025	dedicated	—	—	—
`report-appropriations-supplement-refused.csv`	3	—	generic	—	—	—

What this does not establish

That an UNPROVEN figure is wrong. Nothing has checked it.
What a generic v1 column is on any page where column_meaning says not established.
Whether a page we filed as unknown holds a table worth reading.

Limits that cannot be closed by better code are registered in sources/data/money-gaps.csv and published at /what-we-cannot-answer.

Where these figures come from

what	where
The sixteen reports themselves	/docs/town-annual-reports/docs/, each with its sha256 in /data/archive-manifest.csv
What is on each page and what proved it	/data/annual-report-pages.csv
What a column means before you quote it	sources/data/PROVENANCE-report-tables.md and `column_meaning`

Where this came from

Nothing on this page is an official document. It was written here, from documents the town and district published and from records obtained by request, and it has not been reviewed or endorsed by the Town of Lunenburg, the School Committee, the Finance Committee or Lunenburg Public Schools. The report index says the same thing at more length, and lists every analysis alongside the data underneath it.

This page renders the document itself, which is the source of truth: there is one copy of every sentence and every figure here, not a transcription of one.

Markdown1,450 words8 KBlast changed 2026-09-28sha256 429480a2e6b26196…

Every other report

The budget feed — every board, everything about money →What the town is deciding now →The blog — one finding at a time, in two minutes →This week in town — meetings coming up, minutes and recordings just posted →The boards — each one, in one place →Meeting minutes — written from the recordings →Lunenburg by the numbers — who lives here →Lunenburg’s homes and the tax bill →Homes and students — the town builds, the schools do not grow →The boards, compared →Youth sports and the fields →The School Committee’s finances — every fund and line it owns →Parks & Recreation — the department, its fund, its sales, its grounds →Health insurance →Free cash — can it fill the gap? →Teacher contracts →School user and athletic fees →Extracurriculars — sports, music and clubs →Classroom positions and class size →Overrides →Commercial development and new growth →Special education — four reports →How many students one special education group may have →The circuit breaker — what the state reimburses for the costliest placements →What other districts spend, for each pupil →What the state requires us to spend — and where that puts us →Chapter 70 — the formula, and why it pays the floor →How Chapter 70 actually works, in eight steps →School staffing — did it go up, and over which years →Who works in each school →The paraprofessionals →What courses actually ran, subject by subject →AP exams — who sits them, in what, and how they score →The cut register — what was announced, and what shows →Funding that stopped →When a grant ends — who picks up the bill →Who is in the schools — enrolment, FY1994 to today →Which grades students leave in →Who leaves Lunenburg schools, and where they go →Monty Tech — the assessment, and what sets it →If students leave — what school choice would cost →What a family actually pays →What sports cost, and who pays →Health insurance — the cost outside the school budget →Budget against reported spending →

Every analysis this project has written, in one index, is at reports.