A Field Guide in Ten Parts
Companion to You Cannot Eat Code: 250 Years of Proof That the Future Is Physical, Local, and Human — If You Move Now
Web edition — v1.0, June 2026
You Cannot Eat Code made an argument: that the patterns of organization, infrastructure, and local commitment which built the American republic are the same patterns needed to navigate the AI transition, and that the unit of account is the town, not the platform. That book ends on a question, and the question belongs to you. This book is what you pick up afterward — not to have the question answered, but to be armed for it.
Every part below runs the same circuit. The claim, stated plainly. The proof — named towns, named programs, named statutes, with the numbers at full precision, because numbers carry authority and rounded numbers carry none. The move, priced in hours and dollars. And first actions: what a reader can start within thirty days, with the office, the form, or the meeting where it begins.
The guide is written for two readers at once, and it does not apologize to either. The first lives in a town — five hundred to fifty thousand people — and can reach a selectboard, a clerk’s vault, a fire company. The second rents a floor in a city of millions, and her town is a neighborhood: a community board, a county recorder, a block. Every part ends with first actions for both, because a field guide that serves only one of them is a brochure.
Three honesty rules governed the research behind these pages, and they show. First, failure cases are load-bearing: for nearly every move this guide recommends, it also names a town or a program that tried it and failed, and why. Second, claims that could not be verified to a named, dated source were cut — what survived is the smaller, harder set. Third, no technology in this book is promised beyond what it delivers; where the evidence is thin or new, the prose says so in plain sight.
Web links appear throughout as named references; the printed edition gathers every one of them, current at press time, in the Live Links section at the back, and the web edition keeps them live and updated. Programs change faster than books — when a named program has moved or died by the time you read this, the pattern it proved still stands, and the web edition carries the replacement.
Ten parts. Assess your town. Harden the record. Watch the books. Build the stack. Decide together. Stand up response. Charter the commons. Grow the work. Know the law and the money. Climb the ladder. None of them requires permission to begin, and none of them will tell you which one is yours.
That part was never going to be in a book.
Part I
Ten questions to answer before you spend a dollar or an evening.
At 42C North Street in Goshen, Connecticut — population about 3,150 — there is a counter, and behind the counter a fire-rated vault, and inside the vault, the town. Past the heavy door the air goes cool and the smell is paper: bound record books carrying every birth, marriage, and death that has happened in Goshen in an unbroken run from the colonial period, and deed books running from 1736 — “deeds, mortgages, releases, liens and subdivision maps, which represent the history of Goshen’s land ownership since 1736,” as the town clerk’s office describes its own holdings. Two people keep it: town clerk Barbara L. Breor and her assistant, Michelle J. Pannullo. The office names its first duty without ornament — “maintaining and preserving the records of the town” — and runs machinery fine enough to include a $5 surcharge, under Connecticut General Statutes § 7-34(a), for a deed that omits the grantee’s mailing address.
That counter is the best-protected layer of your town’s existence, and standing at it you can see none of the things this book will ask you to fix. You cannot see that the index to those deed books lives with a private vendor in Liverpool, New York, at searchiqs.com/CTGOSH. You cannot see that Goshen has no charter — that it runs, like most of the Litchfield hills, on the state’s default operating system, the enumerated powers of CGS § 7-148. You cannot see when the backup of the born-digital records was last restore-tested, or how many residents attended the last selectboard meeting, or how many ballot lines carried one name or none last November, or the average age of the fire company that would answer for this building at two in the morning. The vault is the visible town. This part is the audit of the invisible one.
You Cannot Eat Code argued that the town is the unit of account. A unit of account can be audited — and should be, before any move in Parts II through X, because hardening a record you do not hold, building a network nobody will staff, or chartering a commons your state has already preempted is how civic energy dies young. The audit costs nothing but attention, and the asset under examination has earned it: in a Gallup poll fielded September 2025, 65% of Americans trusted their local government to handle local problems — about what Gallup measured in the 1970s, which makes local government the only level of American government whose trust line has held roughly flat across fifty years. Two readers run this audit. One lives in a town of 3,000, where the answers sit at the clerk’s counter and the firehouse. One rents in a city, where “town” means a community board, the clerk is a county recorder, and the firehouse is a borough Office of Emergency Management — her version rides under each question. Both follow one rule: write the answers down. An audit in your head is an opinion. On paper, dated, it is a baseline.
Start with the question almost nobody can answer about the place they live: is your town a charter municipality or a statutory one — and does your state treat its towns as rights-holders or as wards? The American baseline is Dillon’s Rule, named for Iowa Supreme Court Justice John F. Dillon’s 1868 opinions and validated by the U.S. Supreme Court in Atkin v. Kansas in 1903: a local government exercises only the powers expressly granted by the state, those necessarily implied, and those essential to its declared purposes. Home rule is the counter-theory — a zone of local initiative the state cannot casually invade. By the National League of Cities’ tally, 31 states run on Dillon’s Rule, 10 on home rule, 8 apply Dillon’s Rule only to certain municipalities, and Florida applies home rule to everything except taxation — call it 39 or 40 states applying some form of Dillon’s Rule to at least some of their towns.
The label matters less than the live preemption statutes underneath it. Texas is a strong home-rule state, and its HB 2127 — the “Death Star” law, left standing by a Texas appeals court on July 18, 2025 — bars cities and counties from regulating in any field occupied by eight state codes. Connecticut is the hybrid worked example: Dillon’s Rule baseline, softened by Article Tenth of the state constitution, adopted 1965, and the Home Rule Act. The default machinery is CGS § 7-148 — a long enumerated list, construed the Dillon way: Connecticut courts hold that municipalities “have only the powers expressly conferred upon them by general statutes or by special act and those which are fairly to be implied as necessary.” Of the state’s 169 municipalities, 113 have adopted charters and 56 have not — and Goshen is one of the 56, along with Cornwall, Litchfield, Salisbury, Sharon, Kent, Norfolk, Warren, Washington, Morris, Bethlehem, and Roxbury. Most of the Litchfield hills are statutory towns: selectmen, town meeting, and the list. A statutory town is not powerless — § 7-148 reaches property, roads, health, safety, much of daily life — but a move outside the list needs a charter, a special act, or a statute; a charter adds, under CGS § 7-194, general power over the town’s finances and property, its conveyances, and its contracts and debt.
The city renter inverts the question. Her city almost certainly has a charter; her binding constraint is preemption. The first question is never “may my city act?” but “has the state already occupied this field?” — trackable field by field in the LawAtlas state-preemption dataset. Part IX maps the whole constraint layer — statutes, money, liability. This question tells you which map you are standing on, and you can answer it this week with the town website or one call to the clerk.
The clerk is the legal custodian of the town’s records. The operational custodian is, increasingly, a short list of private companies. American recording runs through roughly 3,600 jurisdictions — county recorders, registers of deeds, and, in Connecticut and Rhode Island, the clerk of every municipality; Connecticut abolished county government effective October 1, 1960, leaving 169 separate recording jurisdictions. By July 31, 2019, e-recording had reached 2,007 of the nation’s recording offices — more than 55% of them, covering more than 86% of the population — and the dominant submission network, Simplifile, has belonged since 2019 to ICE Mortgage Technology, a subsidiary of Intercontinental Exchange, the company that owns the New York Stock Exchange.
Now the roster behind your town’s website, agenda system, ordinance code, and land index. Tyler Technologies: $2.14 billion in 2024 revenue, up 9.5%, more than 13,000 client locations in all 50 states. Kofile, which deacidifies, rebinds, and scans the old books, has belonged to Audax Private Equity since 2020 and bought ECS Conservation in February 2026. Granicus — agendas, minutes, meeting video — belongs to Vista Equity Partners and Harvest Partners. CivicPlus — host of Goshen’s website and owner of Municode, and with it many towns’ ordinance codes — took Insight Partners as majority owner in a $290 million deal. Goshen’s own deed index sits with Info Quick Solutions of Liverpool, New York. Add it up: four private-equity firms, one exchange operator, and one public company now stand between most American towns and their own records.
The audit question is not whether your town uses vendors — it does — but whether it could leave them. The vendor’s index is the finding aid, and a town that has never demanded a full export in an open format has never tested whether it can leave. So ask, in writing: who holds our land-record index and images, and do we have a current full export? Which company runs the website, the agendas, the code, the alert service — and who owns each of those companies this year?
The renter runs the same trace one office up: find your county recorder or city register, load its public search page, and read the footer — the vendor’s name is usually printed there. Then pull your own building’s deed and save the PDF, because the searchable index that found it is, in practice, private property. Part II gives the architecture that makes the record citizen-verifiable for under $5,000 a year; this question measures how far the record has drifted from the people it is about.
Rank your town’s records by durability and the order inverts everything the sales brochures imply: the paper in the fire-rated vault is now the most durable object in the system; everything that makes it findable, provable, and usable — index, backup, vendor, basement annex, recording — fails first. The threats the marketing aims at, insider alteration and forged deeds, are the rarest; what actually destroys town records is ransomware, water, and neglect.
The frequency is on the record. Emsisoft counted 77 confirmed ransomware attacks on U.S. local governments in 2021, 106 in 2022, 95 in 2023, 117 in 2024 — every figure an undercount. Even the vendor of record is a target: Tyler Technologies, the dominant municipal-software company, was itself ransomed on September 23, 2020 — North Carolina’s court system suspended all connections to Tyler until it received written assurances. Beneath the attacks runs the slow loss: professionally racked drives die at 1.36% per year — one in 74, in a climate-controlled data center — and no failure statistics exist at all for the town-hall closet, which is itself the finding: that loss is silent and unreported.
So inventory the exposure, which for a typical town is six items long: born-digital minutes and ordinances that exist only on a vendor’s cloud and a clerk’s PC; meeting video, lawfully destroyable six months after the minutes are approved; municipal email, destroyable at two years; the vendor-held land index; GIS layers, permit databases, and assessment systems; and the newest land-record images in towns that stopped security microfilm after going digital — nobody has counted how many. Then put three questions to the clerk in writing: where is our off-site security copy; when was our backup last restore-tested offline — restore-tested, because in Cockrell Hill, Texas the scheduled backup ran after the infection and overwrote the good copies with encrypted ones; and which of the six items above has any copy outside the town and its vendor.
The renter’s version: at the community board or school board, ask where the minutes and recordings are stored, who besides the keeper holds a copy, and how long the video survives — then mirror what matters to you, because the retention schedule will not. Part II answers this question with a six-layer vault costing under $1,200 a year in its minimal form, and Part VI covers what response looks like when loss arrives anyway. This question tells you what your town stands to lose first — which is never the deed book, and almost always the index that makes the deed book usable.
Two numbers define the watching layer in 2026, and they pull in opposite directions. Pew, surveying 5,146 adults in January 2024: 85% of Americans say local news outlets are at least somewhat important to their community’s well-being — and 15% have paid for local news in the past year. The trust and the funding have decoupled, and the census shows where that ends: the 2025 Medill State of Local News report counts 213 news-desert counties — counties with no local news source at all — up from 206 the year before and roughly 150 twenty years ago; 1,524 more counties are down to a single source; some 50 million Americans have limited or no access to local news. 136 newspapers closed in the year — better than two a week — and, in a departure from earlier years, most were the smaller, independently owned papers, the most trusted kind.
A dead newspaper is not a sentimental loss; it is a measurable tax. After a closure, municipal borrowing costs rise 5 to 11 basis points — an estimated $650,000 per bond issue (Journal of Financial Economics, 2020); violations at local facilities of public companies rise 1.1% and penalties rise 15.2% (JFE, 2022); mayoral fields thin, margins widen, incumbents stop drawing challengers.
The other half of watching is the law, and the law is only as good as its enforcer. Massachusetts routes complaints to the Attorney General’s Division of Open Government, which received 395 of them in 2025 and found a violation in roughly 56% of the complaints it reviewed. Connecticut runs the country’s oldest free-standing FOI tribunal: complaint within 30 days, no lawyer, no filing fee, civil penalties of $20 to $5,000 against officials who deny FOI rights without reasonable grounds. New Jersey leaves you to sue — and the only statewide compliance study of its kind found no public entity fully compliant, with municipalities averaging 54%. “Unfortunately, public bodies are aware of this and ignore the law with impunity,” said Walter Luers, the open-government attorney on the study foundation’s board.
So audit. Find your county on the Medill map. Ask when a reporter last sat through a full meeting of your selectboard. Then test the statute yourself — in Connecticut, minutes within seven days, each member’s votes in writing within 48 hours — and note the date your town misses. The renter checks whether her city is a Documenters Network city, where residents are trained and paid to cover public meetings, and reads her community board’s last posted minutes for the gap between meeting date and posting date. Part III builds the replacement watching layer — one volunteer, ten hours a month, roughly $0. This question establishes whether anyone is watching now.
Assume nothing about the cell network; it fails without weather. On February 22, 2024, a configuration error during a network change at 2:42 a.m. took AT&T Mobility down nationwide for at least 12 hours: more than 92 million phone calls blocked, more than 25,000 failed attempts to reach 911, over 125 million devices affected, in all 50 states. The FCC attributed it to failure to follow internal procedures. The platform layer fails on a clear day, by its own hand.
Add weather and the failure is general. Hurricane Helene, September 2024: 74.31% of cell sites in North Carolina’s disaster area out of service on September 28; Avery, Mitchell, and Yancey counties down to fewer than 10% of sites functioning at the trough. More than 1,700 miles of fiber destroyed; 19 of North Carolina’s 100 counties technologically isolated; 17 of the state’s 911 answering points unable to receive calls; an estimated 150,000-plus people in total technological isolation. What demonstrably carried traffic across the 2024–26 disasters, in rough order of reliability: AM/FM broadcast with backup power; licensed amateur radio; GMRS at street scale; satellite; and volunteer-run information services — Watch Duty, the nonprofit wildfire-tracking app, doubled its all-time user count between the Tuesday the January 2025 Los Angeles fires broke out and that Friday. What did not carry: commercial cell and broadband, consistently first down and slowest back — and LoRa mesh, which in Helene carried nothing, because no organized mesh existed before the storm. The stack has to exist before the day you need it.
And the binding failure is often institutional, not technical. Kerr County, Texas, July 4, 2025: more than 100 dead on the Guadalupe; no wireless alerts pushed through FEMA’s IPAWS as the flooding began; the opt-in CodeRED system not fully activated; no flood sirens, the proposal having been drafted and then cut after public pushback; the county’s top two emergency-management officials asleep through the critical pre-dawn hours. Every layer of that failure was a decision, made in budget meetings, years before the rain. An assessment is a budget meeting held early.
The audit is one page: which repeaters, licensed hams, GMRS users, Starlink kits, and backup-powered AM/FM stations exist in town — the communications-resilience inventory, put on the next selectboard or emergency-management agenda. Part IV cashes that page into hardware at three price points. The renter’s page is shorter: what in the building works in a blackout, who on the floor owns a battery radio, whether her city has a mesh to join — New York has one, at nycmesh.net — and what her borough OEM says will carry alerts when the towers don’t.
Whoever they are, there are fewer of them than at any time since counting began. Volunteers are 62% of America’s firefighters — 635,100 of 1,018,100 — and of the nation’s 29,452 fire departments, 18,873 are all-volunteer and another 5,335 mostly volunteer: 82%. Those departments protect roughly 30% of the U.S. population, concentrated in exactly the towns this book is for. The curve runs one way: 884,600 volunteers in 1983; 635,100 in 2023, the lowest count the national survey has recorded — per-capita strength down from 8.05 volunteers per 1,000 Americans at the 1987 peak to 5.66 by 2020, while call volume more than tripled in 40 years.
None of this is abstract on the guide’s home ground. Torringford Volunteer Fire Department — Torrington, Connecticut, Litchfield County — closed April 1, 2025, after 68 years, “forced to shut down due to a lack of firefighters” despite spending thousands of dollars on recruitment, as WFSB reported that March; the city’s Burrville volunteer company had already closed in 2017. And recruitment is only the first failure mode. Yantic Fire Engine Company No. 1 in Norwich — serving since 1847, 58 members, 772-plus calls the prior year — was shut down by its own city at 10:02 a.m. on a Friday in February 2026 in a contract dispute. Companies die of emptiness, and they die of politics.
The audit instrument is a single agenda item: put “fire company status report” on the next selectboard agenda and require the chief’s roster count, average age, and call-coverage gaps in writing. The chief already knows these numbers; the town meeting almost never does. Count the second tier while you are at it: does the town have a CERT team — more than 3,200 local programs operate nationally, more than 600,000 people trained since the program went national in 1993 — and has it ever been activated? In the January 2025 Los Angeles fires, only the vetted, drilled call-out team deployed — not the basic-course graduates; a town that stops at the 20-hour certificate owns a roster, not a capacity. Rosters and capacities look identical on paper. The difference is drills.
The renter’s two-in-the-morning call is answered by professionals, so her audit drops one layer down: which neighbors would she knock for; does her borough OEM run CERT Basic — New York, Los Angeles, Boston, and Chicago all do, 20 hours, free — and who in the building has taken it. Part VI gives the 24-month path from zero to standing capacity, and Part VIII takes up the deeper pipeline — how a town grows the people who do this work. This question establishes the zero.
Walk the inventory: town hall, school, firehouse, library, highway garage, transfer station — then the quieter holdings, the land and easements and water lines and poles and the cemetery. Most towns have never listed their own assets in one place for their own citizens, which makes the second question unanswerable: what could the town own?
More than tradition suggests. Roughly 2,000 American communities get their electricity from a public power utility; public power serves more than 54 million people — about 15% of all electric customers — and the median public power utility serves about 4,400 people. Four thousand four hundred: the size of a Litchfield County town. Municipal enterprise is not a metropolitan exotic; it is a small-town technology with a century of unremarkable operation behind it. The ceiling case is audited — Chattanooga’s EPB, which built America’s first community-wide gigabit network in 2010, has generated $5.3 billion in community benefit since launch, more than six times the original cost, by a November 2025 peer-reviewed study.
The floor cases are audited too, and an honest assessment holds both. iProvo put roughly $40 million of public money into citywide fiber and sold it to Google Fiber for $1. Burlington Telecom was a $33.5 million build inside which officials borrowed about $17 million from the city’s cash pool without authorization; Moody’s downgraded Burlington’s credit six notches, and the network was sold. The difference between Chattanooga and Burlington was not the fiber; it was the governance — the clause-by-clause accountability Part VII exists to write. Note also which state you are standing in: 16 states still preempt or restrict municipal broadband, and Connecticut is not among them — a Litchfield County town faces process friction, not prohibition.
Ownership is not only municipal. The largest community land trust in the country, Champlain Housing Trust, closed FY2025 with total assets of $260,039,187, 706 shared-equity homes, and 2,929 affordable apartments — auditors reporting no material weaknesses. A commons can carry nine figures with open books. The renter’s commons are nearer and smaller: a tool library she can join this week — Chicago’s runs pay-what-you-can memberships against a recommended $100; Baltimore’s Station North prices membership at $1 per $1,000 of annual income — or a community fridge whose entire constitution is one host with standing, one open coordination channel, several restock partners. Part VII writes the charters, the oath layer that makes shared ownership durable. This question drafts the asset list those charters will govern: list what the town owns, then list what it has merely assumed it cannot.
Every plan in this book eventually meets a budget, so learn the shape of yours now. In Connecticut the governing valve is CGS § 7-348: a mid-year appropriation that alone or cumulatively exceeds $10,000 per department per year must go to town meeting in towns with a grand list of $20,000,000 or less; above that grand list, the threshold is $20,000 — figures last raised in 1990. The board of finance may keep a contingent fund of up to 3% of estimated expenditures. Read it as machinery, not obstacle: any unbudgeted experiment above the threshold is, by law, a public vote — so the pilots in this book are designed either to fit under the number or to win the vote, and which number is yours is a fact to establish before you propose anything.
Borrowing is sized against you. Cost of issuance averages 3.096% of principal for municipal bond issues under $10 million versus 0.741% for issues over $75 million — small towns pay 4.2 times the big-city rate. The escape hatch is the bank-qualified bond: an issuer selling $10 million or less per calendar year can designate its paper bank-qualified and let local banks buy it directly — worth 25 to 40 basis points, roughly $232,000 to $370,000 over a 15-year $10 million bond. Slower and cheaper still is the capital reserve fund under CGS § 7-360: small annual deposits, no debt, no bond counsel. Know your state’s cage, too: 46 states and the District of Columbia restrict property taxation in some form — New York caps levy growth at 2% or inflation, Massachusetts at Proposition 2½’s 2.5% — while Connecticut has no statewide levy cap; the town meeting and the board of finance are the only brake. In cap states, assume the general-fund door is nearly shut and plan for reserves, districts, or outside money.
Then list the grant channels your town has ever actually tapped against the ones alive in June 2026: FEMA’s flagship pre-disaster mitigation grants, killed in April 2025, resurrected by court order, reopened March 25, 2026 with a $1 billion pool and an application deadline of July 23, 2026; USDA Rural Development’s Community Facilities program for towns of 20,000 or fewer, with grant shares up to 75% for qualifying communities of 5,000 or fewer; the Trust for Civic Life’s $5,000–$25,000 civic-experiment micro-grants; and, in Connecticut, the Historic Documents Preservation grants running $5,000, $7,000, or $10,000 by town size in FY 2026. The renter’s ledger is the district’s: New York’s participatory-budgeting council districts each put at least $1,000,000 of capital money to a resident vote, and her community board files budget priorities with the borough — ask the district manager where that input actually lands. Part IX runs the full decision tree — the legal question, who answers it, the matched funding source. This question locates your town on it before you need it.
Count three things: the empty seats, the uncontested lines, and the chairs at the back of the hall. In July 2021, Wellfleet, Massachusetts counted 46 openings on 47 committees — Part X opens inside that listing, because the empty bench is the guide’s spine. The national accounting behind it: in 2024, 70% of the races on American ballots were uncontested, an all-time high, and the gradient steepens toward home — 91% of regional races, 82% of county, 74% of local. Not one cycle’s accident: 67% in 2022, 70% in 2020 — three cycles, two independent databases, the same answer. Across roughly 10,000 mayoral elections, 48% were uncontested, and an uncontested race depresses turnout by about 9 percentage points: the vacancy is not neutral; it shrinks the electorate around it.
The bench behind the ballot is thinning differently — wider and shallower at once. In the year ending September 2023, 75.7 million Americans — 28.3% of the population sixteen and over — formally volunteered through an organization, what the Census Bureau and AmeriCorps called “the largest expansion of formal volunteering” since tracking began in 2002. But average hours per volunteer fell from 96.5 a year in 2017 to 70 in 2023, and the median fell from 40 to 24; behind that, the oldest conveyor into civic service — weekly religious attendance — dropped from 42% of American adults in the early 2000s to 30% in 2024. More people willing to show up once. Fewer carrying anything home.
So count your own town. Ask the first selectman’s office for the standing vacancy list — Simsbury publishes one outright, titled “Vacancies on Simsbury Boards and Commissions as of February 1, 2025,” and most Connecticut towns keep the same list at the town website or the clerk’s office. Pull the clerk’s returns from the last municipal election and count the lines that carried one name or none. Then attend one meeting and count the audience after subtracting officials, applicants, and their relatives — that number, usually expressible on one hand, is your town’s watching-and-deciding capacity on an ordinary night.
The renter counts the same three things under different names: New York’s 59 community boards seat up to 50 unsalaried members each, appointed by borough presidents, and the appointed tier runs documented double-digit vacancy rates — Maui County, the cleanest measured case, ran 18%, 57 of 304 seats unfilled, with boards “paralyzed,” in Honolulu Civil Beat’s word, for a year for lack of quorum. Part V is how a town fills the room and turns a filled room into binding decisions; Part X prices every seat on the ladder. This question counts the empty chairs — and in most American towns, the count is the argument.
The last question turns the instrument on its operator. You have audited the town; now audit the evenings you can actually give it — because every gap this part has measured, the untested export and the unwatched meeting and the thinning roster and the uncontested line, is filled by hours, and the hours are yours or nobody’s.
Price honestly, from the documentation that exists. The bottom of the ladder costs 2 to 20 hours a year and $0 — or pays you: a poll-working day in Bristol, Connecticut pays $275 to $340 depending on position, and a Documenters assignment pays by the hour at site-set rates. The middle costs 25 to 160 hours in year one and $0 to $500. The top costs 100 to 800 training hours, or a career change. An appointed city seat runs 6 to 8 hours a month by official Brooklyn guidance; a small-town selectboard seat runs about two formal evenings a month plus near-weekly low-grade attention — documented guidance from one Vermont town of 1,200, and an estimate beyond it, because no national survey has ever counted those hours. Above that the record goes quiet altogether: nobody prices the volunteer firefighter’s year or the selectman’s evenings — the hour-cost of America’s smallest offices is mostly unmeasured. And the ballot itself is cheaper than any rumor of it: a petitioning candidate in Connecticut needs signatures equal to 1% of the votes cast for the office last time, or 7,500, whichever is less — 15 signatures where 1,500 votes were cast, 8 where 800 were.
Against the cost, the audited return. Across 148 studies and 308,849 participants, stronger social relationships were associated with a 50% higher likelihood of survival over an average 7.5 years of follow-up. In a four-year prospective study of 12,998 adults over 50, volunteering at or above 100 hours a year was associated with a 44% lower mortality risk — and, read honestly, with no detectable effect on diabetes, chronic conditions, cancer, or hypertension incidence. This guide does not promise transformation. It prices returns.
Now the audit routes. What your town legally is, and how it pays: Part IX. The record: Part II. The watching: Part III. The wire: Part IV. The room and the decision: Part V. The two-in-the-morning call: Part VI. The commons: Part VII. The work, and who learns it: Part VIII. Your own seat: Part X. Ten answers, written down and dated, are a baseline almost no one else in your town holds — which makes you, as of this evening, the best-informed person in the room you just counted. The book that sent you here left its final question standing, and this volume will not answer it either. It will only make the answering cheap.
For the reader in a town:
For the renter in the city:
Either reader:
Part II
“Technologies come and go. Public records, especially archival records, must endure.”
— Vermont State Archives and Records Administration, January 2019
For two days in October 1871 — the 8th through the 10th — Chicago burned, and somewhere in the middle of the burning, the building that held Cook County’s deed records went up with everything inside it. The Chicago Times needed one sentence for the damage: “There has been absolute destruction of all legal evidence of titles to property in Cook County.” Read it again: not the property — the evidence. The lots still stood, or their ashes did; what burned was the proof of who owned them.
What saved Chicago’s titles was not the county, and not the law. It was three private abstract firms — Chase Brothers and Co.; Shortall and Hoard; Jones and Sellers — whose clerks had spent years copying the public record into their own tract books, and who hauled those duplicates out ahead of the flames. John Shortall commandeered a wagon at gunpoint to do it. The legislature then did the only sane thing left: the Burnt Records Act of 1872 — alive today as the Destroyed Public Records Act — let courts accept the private books as evidence of title. The unofficial copy, kept by someone else, somewhere else, became the record. It usually does.
You Cannot Eat Code argues that the town is the unit of account. A unit of account keeps books. This part is about the books — who was born, who married, who died, who owns what ground, what the town decided — and about making them tamper-evident, redundant, and verifiable by any citizen — for under $1,200 a year minimal, under $5,000 fully loaded. First, who holds the record now — because the honest answer has changed underneath you.
The United States has no national register of births, marriages, or deaths. Vital records are state law, kept two ways: most states make the health department legal custodian, fed by county registrars; New England creates and keeps the original certificate in the town where the event happened, a copy sent up to the state; the federal government holds no certificates at all.
In Goshen, Connecticut — population about 3,150 — the record lives at 42C North Street, in the office of town clerk Barbara L. Breor — registrar of vital statistics, election officer, “agent for the State of Connecticut” — and her office names its first duty plainly: “maintaining and preserving the records of the town.” That means the original certificate of every birth, marriage, civil union, death, and fetal death in the town’s history, bound into record books in the fire-rated vault — an unbroken run from the colonial period, with deed books starting in 1736. Connecticut triple-records: town of occurrence, town of residence, and the State Department of Public Health — three statutory copies, before computers existed (CGS §§ 7-42 and 7-44).
That is the record. Now follow the index — no longer in the vault. Goshen’s land-record index lives with a private vendor, Info Quick Solutions of Liverpool, New York, at searchiqs.com/CTGOSH. The deed pipe is privatized too. Recording in America spreads across roughly 3,600 jurisdictions — county recorders, registers of deeds, and in Connecticut and Rhode Island the clerks of every municipality; Connecticut abolished counties effective October 1, 1960, leaving 169 recording jurisdictions. By July 31, 2019, e-recording had reached 2,007 of them — over 55% of the country’s recording offices, more than 86% of the population. The dominant submission network, Simplifile, has belonged since 2019 to ICE Mortgage Technology — Intercontinental Exchange, owner of the New York Stock Exchange — and advertises “the nation’s largest eRecording network — over 2,600 counties covering 90% of U.S. residents.” The NYSE’s parent company sits in the deed pipeline.
The rest of the vendor layer follows the pattern. Tyler Technologies posted $2.14 billion in 2024 revenue, up 9.5%, claiming more than 13,000 client locations in all 50 states. Kofile of Dallas — which deacidifies, rebinds, and scans the old books — went to Audax Private Equity in 2020 and bought ECS Conservation in February 2026: even the rebinding of the old books is consolidating. Granicus (agendas, minutes, meeting video) belongs to Vista Equity Partners and Harvest Partners; CivicPlus — host of Goshen’s website, owner of Municode and many towns’ ordinance codes — took Insight Partners as majority owner in a $290 million deal.
Add it up: four private-equity firms, one exchange operator, and one public company now sit between most American towns and their own records. The clerk holds the statutory duty; the roll-up holds the index, the images, the website, the agenda system, and the alert service. The vendor’s index is the finding aid, and a town that has never demanded a full export in an open format has never tested whether it can leave.
On August 16, 2019, twenty-two Texas local governments were encrypted simultaneously — one compromised managed-service provider, one toolchain, one collective ransom demand of $2.5 million. In Borger, Texas — population about 12,500 — the city’s press release said it in municipal prose: “Vital Statistics (birth and death certificates) remains offline.” The town-clerk function itself was down — no certificate destroyed, none obtainable either. Texas refused to pay — the state reported itself “unaware of any ransom being paid in this event,” and all twenty-two were in remediation within a week — while that same summer Riviera Beach, Florida paid 65 bitcoin, roughly $600,000, after one clicked email attachment. The contrast is the policy lesson: backups plus coordinated response beat ransom; the cities that paid were paying for the backup they didn’t have.
This is not rare weather: Emsisoft counted 77 confirmed attacks on U.S. local governments in 2021, 106 in 2022, 95 in 2023, 117 in 2024 — every figure an undercount. The town record dies five ways, each with a named proving case:
One inversion should reorganize your town’s spending: the paper in the fire-rated vault is now the most durable object in the system; everything that makes it findable, provable, and usable — index, backup, vendor, basement annex, recording — fails first. In every case above, the paper survived. The ability to use it did not.
Now the honest frame — the next section is full of people who didn’t have one. Tamper-proof is a marketing claim; no deployed civic-records system delivers it. What mathematics actually delivers is tamper-evidence: if a record is altered after the fact, the alteration is detectable by anyone holding the right fingerprint. NIST’s own blockchain overview — NISTIR 8202, October 2018 — says “tamper evident and tamper resistant.” Not tamper-proof. The agency chose carefully; so should you.
The second distinction matters more: entry versus alteration. No integrity system prevents a false record from entering through the legitimate channel; it only prevents silent change afterward. A forged deed that is hashed, timestamped, and anchored is now a forged deed with an excellent audit trail. The marquee case: in May 2024 a fake lender, “Naussany Investments & Private Lending LLC,” moved to foreclose on Elvis Presley’s Graceland over a fabricated $3.8 million loan. Lisa Findley forged Lisa Marie Presley’s signature and a Florida notary’s and filed a fake deed of trust with the Shelby County Register’s Office — which accepted it, because recording offices authenticate format, not truth. Findley pleaded guilty to mail fraud and drew 57 months in federal prison on September 23, 2025. Every safeguard in this part would have recorded her fraud beautifully; the defense at the entry gate is human review plus rapid notice — the boring, free fraud alert: Maricopa County’s Title Alert, launched June 2023, passed 92,900 subscribers. The entry problem is not boutique: the FBI logged 9,359 real-estate fraud complaints with $173.6 million in losses in 2024, rising to 12,368 complaints and roughly $275 million in 2025.
Here is the thread that runs through the whole guide: the human gate is always the weak layer. The forger at the counter is one version. The keeper is the other — rarer and worse, because the clerk is the system of record and the most dangerous adversary is the one with legitimate write access. Tina Peters, clerk and recorder of Mesa County, Colorado, facilitated unauthorized access to her own county’s election systems in May 2021 — convicted August 12, 2024, sentenced to nine years, $1.4 million in costs to her county. (Commuted May 15, 2026; the politics are not the point — the write access is.) Only externally published commitments — fingerprints the keeper cannot quietly edit — constrain the keeper; the architecture below is built on that.
The honest list, printable in the town report: tamper-evidence cannot stop a false entry (durable, datable garbage), cannot stop deletion (it only makes the gap visible), cannot alone prove who entered, and cannot survive abandonment — if nobody checks the hashes, tamper-evidence is theater. One design rule rides with it: prefer keyless integrity — publish hashes rather than manage signing keys. DigiNotar went from breach to bankruptcy inside three months of 2011; Estonia blocked roughly 760,000 national ID certificates on November 3, 2017 over one vendor’s flawed key generation; a hash register has no key to lose in a desk drawer or retire with an employee. Verification has to be somebody’s job. We will make it somebody’s job.
The blockchain decade produced a clean experimental record — read it before a vendor reads it to you. Cook County, Illinois — the office the 1871 fire taught about copies — ran the most rigorous American pilot, October 2016 to May 2017. Its final report, by Deputy Recorder John Mirkovic, May 30, 2017, delivered the era’s central legal finding: “a blockchain transfer, to be afforded notice in Cook County, must ultimately produce a paper document that evidences a transaction.” The office adopted none of it and was later abolished into the County Clerk — which today runs a conventional free Property Fraud Alert instead.
South Burlington, Vermont ran six weeks with Propy in early 2018 and produced what the trade press called the first U.S. blockchain-recorded deed — Katherine Purcell’s condominium, transferred to her own LLC by a paper deed carrying a code for the blockchain entry, recorded in parallel with, never in place of, the city’s land records. The silence was the verdict: the state archives’ own 44-page report on blockchains for land records, January 2019, never mentions the pilot. Vermont had already concluded, in a January 15, 2016 state report, that “blockchain technology would be of limited value in conducting state business,” and answered with a statute instead of a system: 12 V.S.A. § 1913 (2016) made blockchain-registered records self-authenticating evidence — it cost nothing and built nothing.
The rest of the graveyard reads quickly: Honduras’s celebrated 2015 Factom land-title project never deployed, and Factom filed Chapter 11 on June 18, 2020, owing up to $7.5 million — a record whose proof depends on a startup inherits the startup’s balance sheet. Eight years on, no U.S. recording jurisdiction has moved its legal land record to a blockchain. The deed book — paper, microfilm, and vendor-hosted images — remains the record.
Then the survivor — the pattern to copy. Since April 2018 the Washoe County, Nevada recorder has emailed certified copies of marriage certificates and real-estate documents as plain PDFs whose hash is anchored by the vendor Titan Seal — roughly 950 marriage certificates by early 2019 — and the county’s explanation — still live in June 2026 — is honest: verification works “without contacting our office,” and “A single pixel of difference on a document would cause it to fail verification.” Copy it carefully: verification runs through the vendor’s website — if the vendor folds, verification breaks though the documents survive — so also publish raw digests somewhere vendor-independent. The decade’s verdict, in one line: what survives contact with a clerk’s office is hashes, timestamps, and notification services — the primitives, not the platform.
The law has cleared the ground; one misconception dies first. The Uniform Electronic Transactions Act (UETA, 1999) — enacted in 49 states, the District of Columbia, Puerto Rico, and the U.S. Virgin Islands, New York alone holding out under its own ESRA (2000) — and the federal E-SIGN Act (2000; 15 U.S.C. §§ 7001 et seq.) make electronic records lawful between consenting parties; they compel no government office to accept or produce anything. E-recording is separate: the Uniform Real Property Electronic Recording Act (URPERA, 2004), adopted in 37 states plus the District of Columbia as of August 2021, with adoptions continuing since; Connecticut adopted it as CGS Chapter 92a via Public Act 08-56, effective October 1, 2009 — each clerk chooses whether to e-record, and every clerk who does must keep accepting paper. Both famous pilots converged on the same design rule: the statutory record stays canonical; the cryptographic layer is an anchor, never a replacement. Which yields the governing principle: hardening is legal everywhere; substitution requires the records administrator’s blessing. A town may freely add durable layers — microfilm, acid-free paper, anchors, copies; the law regulates destruction and the official copy. In Connecticut nothing is destroyed without the Public Records Administrator’s signed Form RC-075, and “No original document dated prior to the year 1900 shall be destroyed… without the express written approval of the Public Records Administrator” (Conn. Gen. Stat. § 7-109). No statute anywhere mandates any of this — search for one and you will come back empty. The lane is legal, cheap, and empty.
The reference architecture, end to end — for a volunteer-run town of 500 to 50,000 that wants silent alteration of minutes, ordinances, and land indexes detectable by anyone. Every component is alive in June 2026. Nothing here requires a programmer; everything requires a calendar.
Layer 1 — the paper root. Minutes and ordinances are still adopted on paper, signed in ink, in a sewn, paginated minute book on permanent paper — ANSI/NISO Z39.48-1992 (R2009): pH 7.5–10.0, ≥2% alkaline reserve, under 1% lignin — engineered to last several hundred years. A sewn book is a physical append-only log: removing a page is detectable by binding, sequence, and ink. Cost: the paper premium, under $150 a year.
Layer 2 — the digital twin. Within a week of adoption, the clerk or a sworn volunteer exports each document to PDF/A in a year-folder; one command — sha256sum *.pdf > manifest-2026-Q2.txt — produces the quarterly fingerprint register. A SHA-256 hash is a 64-character fingerprint that changes completely if one comma changes; the tool ships free in every operating system — the hard part is discipline (hash at adoption, every time), not technology. Cost: $0 and about an hour a month.
Layer 3 — the anchor. The manifest is stamped two independent, free ways: ots stamp submits its hash to OpenTimestamps, which aggregates it into a tree committed to the Bitcoin ledger — the most heavily witnessed public ledger there is — plus one RFC 3161 token from a free public Time-Stamp Authority (FreeTSA; DigiCert’s public endpoint) for any statute that names a standard. Now no one — the town included — can backdate or rewrite a manifest without the forgery being mathematically demonstrable. Cost: $0.
Layer 4 — the witness. The quarterly digest — one 64-character line — goes where the town cannot quietly edit: a page in the printed annual town report, mailed to every household and archived by the state library; the town website; optionally a legal notice in the local newspaper. The trick is old: Surety Technologies — founded by the inventors of digital timestamping — began publishing a weekly hash of its entire customer database in a New York Times classified in 1995, still running when Vice profiled it in 2018 as “the world’s oldest blockchain”; to forge that history you would have to recall every printed copy. The pattern scales: the internet’s certificate system has logged more than 17 billion certificates to append-only logs Chrome has required since April 30, 2018. Now the open lane: no U.S. town or city — none that any search can surface — publishes a hash register of its minutes and ordinances in its printed annual report. The move is unclaimed; a town of 3,000 could — as far as anyone has yet found — be first in America, for approximately $0. Cost: $0–$800 a year, depending on newspaper use.
Layer 5 — the copies. Nightly, a free rclone job syncs the records share to Object Lock buckets at two unrelated vendors — Backblaze B2, $6/TB/month; Wasabi, $6.99/TB/month — and a town’s full corpus is almost always under 1 TB: call it $156 a year for both. In compliance mode, not even the account owner can delete inside the retention window — converting ransomware from “records destroyed” to “records inconvenienced”: the highest-value dollar a town spends on records integrity. Backblaze’s 2025 statistics across 344,196 racked drives show 1.36% dying per year — one in 74, in a climate-controlled data center; the town-hall closet is worse. Once a year, burn the year-folder to two archival Blu-ray discs ($5–15 each): one for the vault, one deposited with a neighboring town — a reciprocal-deposit compact that, as far as the record shows, no two American towns have yet signed, waiting on nothing but two clerks and a memorandum of understanding. Give it a funding line and a named successor: MetaArchive, which replicated member archives across seven caches for two decades, dissolved March 31, 2025 with no data lost — cooperatives die of administration, not entropy. Land records stay under the state microfilm mandate where one exists — Connecticut’s: “Each town clerk shall maintain a systematic program for microfilming the land records and indexes,” the film “stored at a secure off-site facility approved by the Public Records Administrator” (Public Records Policy 11). Microfilm to ISO 18901 is rated for 500 years, readable with no electricity, no format, no vendor — the only digital-age medium with a five-century rating from a standards body rather than a marketing department.
Layer 6 — the verifier. By named appointment — library trustees, the moderator, a high-school civics class — someone who is not the clerk runs sha256sum -c and ots verify against the published registers, and the one-sentence result (“all 47 files verified against the published register, June 2, 2026”) is read into the next minutes and printed in the report. Ten minutes, twice a year, witnessed. This converts cryptography into civic ritual — the only form in which it survives staff turnover. Tamper-evidence is a social practice wearing a cryptographic coat.
The ledger: under $1,200 a year minimal — cloud, media, paper, notices — and under $5,000 with professional scanning of the land-index backfile amortized, microfilm service fees, and a records-grade fireproof cabinet; labor: one volunteer-hour a month plus two ceremonies a year. Spend in order: first dollar on the immutable backup (ransomware), second on getting sole copies out of the basement (Montpelier), third on the free fraud alert — and the zeroth, free, on the published hash register, the only countermeasure that addresses the insider, disciplines the vendor, and gives the citizen something to check. Connecticut funds it: the Historic Documents Preservation grants — fed by a $6 fee on every land-record filing — have distributed over $19 million since 2000; Goshen took $5,500 in FY 2021, and FY 2026 awards run $5,000 / $7,000 / $10,000 by town size. The architecture costs roughly one grant cycle.
Print the promise and the disclaimer together: any resident, any title searcher, any future historian can prove the minutes they hold are the minutes that were adopted — without trusting the clerk, the vendor, the cloud, or the town; and none of it protects against false records entered through legitimate process, deletion of what was never copied, or a town that stops the ritual.
One layer will outlive all the others, and your town already owns it. In 1986 the BBC built a national digital record of Britain — the Domesday Project, a survey on laserdisc honoring its 900-year-old namesake — and by 2002, sixteen years later, it was effectively unreadable — players and formats dead, rescued only by university emulation. The original Domesday Book, parchment, 900 years old, remains readable at the UK National Archives. Sixteen years against nine hundred — which is why Layer 1 is paper and Layer 5 includes film: the digital layers are anchors and copies, never the root.
Do not get romantic about parchment, though — paper’s failure mode is total. On July 12–16, 1973, the National Personnel Records Center in St. Louis burned through approximately 16–18 million military personnel files — 80% of Army records for discharges between November 1, 1912 and January 1, 1960 — and, in the National Archives’ own words, “no duplicate copies of these records were ever maintained, nor were microfilm copies produced.” Paper is a superb persistence layer and a terrible sole layer — never paper or digital; both, copied, apart.
Your town has already run the experiment, uncoordinated: the largest backup of American town records was made by a church. The Granite Mountain Records Vault outside Salt Lake City, drilled into 700 feet of rock, holds 2.4 million rolls of microfilm — approximately 3.5 billion images from over 100 countries, all digitized as of September 2021. For Connecticut, “most records from the colonial period to about the year 1900 have been microfilmed by the Latter-day Saints” — the records of towns like Goshen sit inside a Utah mountain whether or not the first selectman knows it. Beside it, an antiquarian layer: Lucius Barnes Barbour, State Examiner of Public Records from 1911 to 1934, abstracted the vital records of 137 Connecticut towns onto more than one million index slips and 14,333 typed pages. So a Goshen birth from 1790 now exists in at least five places — the town vault, the Barbour slips, the Barbour bound volume, the microfilm in Granite Mountain, and the digitized image online. Two hundred fifty years of accumulated, mostly uncoordinated redundancy — the model this part generalizes on purpose.
But the accumulated redundancy stops around 1900. Outside the town and its vendor there is typically no copy of: born-digital minutes on a vendor cloud and a clerk’s PC; meeting video, lawfully destroyed at six months; municipal email, at two years; the vendor-held index; GIS, permits, assessments; and the newest land images in towns that quit security film after going digital — nobody has counted how many. The five-places redundancy your great-great-grandmother’s birth enjoys, this morning’s minutes do not have. The architecture above exists to give it to them.
One boundary keeps community redundancy honest: only the registrar issues certified copies — the model act behind most state codes threatens up to a $10,000 fine and 5 years for false statements or unauthorized issuance — but everything short of the seal is open: church registers, funeral-home logs, society indexes. The design rule: copy and index the official record; never substitute for it. Connecticut hands citizens a statutory key: under CGS § 7-51a, members of authorized genealogical societies — the State Library lists 14 — may examine all vital records in any registrar’s custody, “certificates, ledgers, record books, card files, indexes and database printouts.” Join one, and the vault opens legally.
Now the other reader — whose town is a rental floor in a city, whose registrar is a county office, whose deed books belong to a recorder she has never met. The pattern is identical; only the names shrink. You will not get Connecticut’s genealogical key everywhere — New York outside the five boroughs holds birth records 75 years; California issues non-family members informational copies only — so start with what is yours by right: order certified copies of your own birth and marriage now and store them outside your home. Enroll in the recorder’s fraud alert and watch your landlord’s building along with your own name. Pull your building’s deed and parcel record from the public search, save the PDFs, hash them, stamp the manifest: you now hold an independent fingerprint of your own housing record. And when your community board’s minutes post, ask on the record: where are these stored, and who besides the keeper can prove they haven’t changed?
Chicago’s titles came through the fire in a wagon Shortall had to take at gunpoint, because nobody had decided, in advance, whose job the copy was. Your town gets to decide in advance. The vault is bought. The statutes permit. The tools are free, and the lane — as far as anyone has found — is still empty. What remains is a calendar, one page in the town report, and a named person who checks. The role, not the software, is the scarce part.
For the clerk, the selectman, the trustee:
shasum -a 256 over the town’s latest approved minutes PDF, paste the 64-character digest into the next selectboard packet, and ask that it be read into the record — town clerk’s office, ten minutes, $0, and the precedent is created.For the renter in the city:
sha256sum -c verification when your local body adopts the register pattern — because the role, not the software, is the scarce part.Part III
“…shall always be and remain public property, and shall be subject to the inspection of all persons therein concerned.”
— the open-books clause of New Mexico’s acequia statute, 1895 lineage, on the account books of a community irrigation ditch (NMSA 1978 § 73-2-21(D))
The voice was familiar. That was the weapon. On January 17, 2024, Baltimore County police learned of an audio clip spreading through Pikesville, Maryland, in which Eric Eiswert, the principal of Pikesville High School, appeared to be making racist and antisemitic remarks about his own students and staff. The clip moved the way sound moves now — phone to phone, feed to feed, faster than any denial could travel. It went viral. The district removed Eiswert from his school while it investigated. Threats came to him and to his family.
It took three months — Baltimore County police working with the FBI — to establish that the voice had never spoken. The audio was an AI forgery, and the man investigators concluded had built it was not a foreign operation or an anonymous troll but the school’s own athletic director, Dazhon Darien, arrested that April and charged with disrupting school activities, theft, retaliating against a witness, and stalking. In January 2025 Eiswert sued the district, alleging it had kept him out even after his union put forensic findings on the table: the audio had been “compiled in a fraudulent and deceptive manner, and possibly entirely fabricated using AI tools.”
Hold the detail that matters. The institution that employed Eric Eiswert could not authenticate its own principal’s voice, and its default — under pressure, in public, in real time — was to believe the clip. Pikesville gets filed under artificial intelligence. File it under absence. The forgery was an entry attack on a community’s information record — a false thing walking through the front door of attention — and there was no standing layer whose job was to check it, because there was nothing agreed-upon to check it against. Detection did not fail. Watching had never been built.
You Cannot Eat Code argues that the town is the unit of account. Part II of this guide hardened the unit’s record against silent alteration. This part is about the other gate, the one the Pikesville file walked through: entry — what gets believed, what gets spent, what gets decided in rooms with twelve folding chairs and, increasingly, no reporter in the back. The watching layer that used to cover those rooms has been dying for twenty years, and its absence is not a mood. It is a line item. Start there.
Count it first. The Medill State of Local News Report at Northwestern University — successor to Penelope Muse Abernathy’s news-desert censuses — put the 2025 numbers down flat: 213 American counties with no local news source at all, up from 206 a year earlier, against roughly 150 twenty years ago. Another 1,524 counties are down to a single source; combined, some 50 million Americans now live with limited or no access to local news, up from about 37 million two decades back. In the year before the report, 136 newspapers closed — better than two a week. The industry has shed more than three-quarters of its jobs since 2005, 7% of them in that final year. Web traffic to the hundred largest papers fell more than 45% in four years.
The character of the dying changed, too. “Unlike in previous years, however, the majority of papers shutting down now are smaller, family-owned enterprises,” Zach Metzger, director of the Medill project, said with the October 2025 release. “These are often the most trusted active local news sources, and their loss creates new challenges for local news access in many communities.” The chains cut long ago; now the families are surrendering. What backfills is not neutral: more than 300 local news startups launched in five years, 80% of them digital-only and overwhelmingly metropolitan, alongside 849 sites run by 54 national network chains — some of them “pink slime,” partisan product wearing a hometown nameplate. In nine counties the sole remaining source of local reporting is a public radio station — one transmitter, exposed the same year federal funding for public broadcasting was cut — and 250 more counties sit on the predictive model’s high-risk list for the coming decade.
Now price it, because it has been priced. When a local newspaper closes, the bonds of the governments it covered get more expensive — municipal borrowing costs rise 5 to 11 basis points, roughly $650,000 per bond issue — and the effect survives every control for underlying economic decline; the finding (Gao, Lee, and Murphy, Journal of Financial Economics, 2020) drags higher government wages, larger deficits, and costlier negotiated sales behind it. Lenders are not sentimental. They read the absence of a watcher as risk, and they bill for it. The rest of the ledger fills in the same direction. California cities whose newsrooms were cut deepest drew fewer mayoral candidates, wider victory margins, more incumbents running unopposed. When the Cincinnati Post folded in 2007, candidate counts, turnout, and campaign spending sagged across the Kentucky suburbs where its readers lived. Citizens’ knowledge of their own local politics declined measurably as coverage thinned. And after a closure, violations at the local facilities of public companies rise 1.1% while penalties rise 15.2% — the paper had been functioning as a regulator that never billed the state. A dead newspaper is not a sentimental loss. It is a measurable tax — on the bond rating, on the ballot, on the factory fence-line.
Here is the strange part, and the part this whole guide leans on: the asset underneath did not collapse. Gallup, fielding September 2 through 16, 2025, found 65% of Americans trust their local government to handle local problems — a figure Gallup notes is “largely similar to what was measured in the 1970s,” across the same half-century in which trust in the federal government’s handling of problems fell to historic lows. Local government is the only level of American government whose trust line runs roughly flat across fifty years. The affection even extends to the dead watchman: 85% of U.S. adults told Pew in January 2024 that local news outlets matter to their community’s well-being — 44% said extremely or very important — while only 15% had paid for any local news in the past year, and a majority believed, wrongly, that the outlets were doing fine. The trust survived; the funding decoupled from it. Which means the lane is not poisoned — it is merely empty. An empty lane beside a flat fifty-year trust line is the cheapest civic real estate in America. The rest of this part is about occupying it.
Take inventory of what you hold before you build, because the legal architecture of watching is already on the books. Every one of the fifty states carries both an open-meetings statute and a public-records statute; the federal FOIA — 5 U.S.C. § 552, enacted 1966 — never touches your town, which answers only to its state law. The decisive variable is the one nobody reads for: who enforces. Three models exist.
Massachusetts runs enforcement through the attorney general — a Division of Open Government, created 2010, sole enforcer of the state’s Open Meeting Law. In 2025 it received 395 complaints, resolved 306 of them through 264 determination and declination letters, and found a violation in approximately 56% of the complaints it reviewed — up two points from the year before. The same office trained more than 1,220 officials and answered some 1,482 helpline inquiries that year, and the violations it finds are mostly entropy, not conspiracy: vague notices, thin minutes, deliberation drifting outside the posted meeting, executive sessions that should not have been. Connecticut built the second model in 1975: the Freedom of Information Commission, the oldest free-standing FOI tribunal in the country. A citizen denied a record or shut out of a meeting files within 30 days — no lawyer, no fee — and the Commission can order disclosure, void action taken at an illegal meeting, and fine an official $20 to $5,000 for denying FOI rights “without reasonable grounds,” while the most a frivolous citizen complaint can draw is $1,000. The asymmetry is deliberate, and it favors you. The deadlines have teeth: minutes posted within seven days, each member’s votes in writing within 48 hours.
The third model — the default in most states — is the courthouse, and where the only remedy is a lawsuit, compliance decays. The New Jersey Foundation for Open Government ran the first statewide compliance study of its kind (data gathered 2018, released March 2021) against four basic requirements of the state’s Open Public Meetings Act and found no public entity fully compliant: municipalities averaged 54%, local independent authorities 44%, boards of education 60%. Walter Luers, the open-government attorney on NJFOG’s board, named the mechanism at the study’s release: “Citizens can file complaints in Superior Court, but few of them have the knowledge to file a pro-se lawsuit without legal representation and they don’t have financial means to hire an attorney. Unfortunately, public bodies are aware of this and ignore the law with impunity.” No national census of municipal open-meetings compliance exists — but where someone audits, they find violations in roughly a third to a half of the bodies checked. The law is owned. The watching is not staffed.
So here is what a citizen concretely does when a board goes dark — four steps, each cheaper than it looks. One: ask on the record, in writing, citing the statute by name — agenda, minutes, recording; in Connecticut the citation is CGS § 1-225, seven days and 48 hours. Two: file the records request; the letter generators are free — the Reporters Committee’s iFOIA system, the Student Press Law Center’s, NJFOG’s — and MuckRock will file, track, and publish it for you. Three: file the enforcement complaint — with the attorney general’s division in Massachusetts, with the FOI Commission within 30 days in Connecticut, for an advisory opinion from the Committee on Open Government in New York, in Superior Court with fee-shifting in the court-only states. Four: make the record yourself. All fifty states allow recording of open public meetings, subject to reasonable rules; Connecticut’s § 1-226 expressly protects broadcasting and photographing open sessions. That fourth step is the legal floor under everything in the next section — because on a San Diego Documenters team’s very first assignment, in 2024, trained volunteers caught a county committee canceling a meeting without public notice. Not a lawyer. A resident with a note template.
The note template has a payroll now. City Bureau, a Chicago civic-media nonprofit, launched Documenters in 2018: train residents to cover public meetings, pay them, fact-check the notes, publish free under Creative Commons. Eight years in: more than 6,000 Documenters trained, more than 14,000 public meetings covered, more than $1,500,000 paid — typically $15 to $20 an hour, set site by site, through local partners like Outlier Media in Detroit, Signal Cleveland, MinnPost in the Twin Cities, The Beacon in Kansas City.
The outcomes carry names. Detroit Documenter Rukiya Colvin’s notes from a Housing Commission meeting kick-started a two-year Outlier Media investigation. Chicago Documenter Amethyst Davis founded the Harvey World Herald, a newsroom for a suburb that had none. A Kansas City Documenter’s notes on soil contamination sent a grandmother to test her garden — dangerous levels of lead. The roster shows the failure file too: the network counts 25 communities, and the listing carries four sites in hibernation — Atlantic County, Bismarck, Spokane, Fresno — on a philanthropic base, a $10 million grant in 2022 included.
Vermont rebuilt a different layer: not the reporter but the tavern and the post rider. Front Porch Forum, founded in 2000 by Michael and Valerie Wood-Lewis as a neighborhood listserv in Burlington, now reaches nearly 250,000 members in a state of about 647,000. The mechanics are the design spec, each one refusing a platform-era default: one forum per town; members live there and post under real names; a paid human moderator reads every post before it publishes; one digest email a day — no feed, no algorithm, no grants; local advertising and member donations carry it. In June 2026 the founders announced an employee-ownership-trust exploration — succession designed to refuse acquisition. “Our goal is to use our tool set to get people offline and connecting with neighbors in real life,” Michael Wood-Lewis said. Why it works there and resists export: the median Vermont town is a few thousand people — the forum maps onto a real civic unit; real names price misbehavior; premoderation is the cost nobody else will pay; the habit is twenty-five years deep. It is the closest functioning 2026 equivalent of the tavern-and-post layer.
The third replacement needs no founder and no network — it needs a discipline. citymeetings.nyc, built by software engineer Vikram Oberoi in 2024, publishes AI-drafted, human-reviewed “chapters” of every New York City Council meeting the same day, 10,000-plus monthly visitors; the division of labor is the design — AI transcribes and segments, a human verifies every chapter before publication. Chalkbeat watches about 80 school districts in 30 states through one commercial account, its in-house tool abandoned. “It’s not a replacement for coverage, and we’re not trusting AI to get these things right. It’s more like a news tip,” said Eric Gorski, its managing editor for local news. In Camden, Maine, Civic Sunlight emails timestamped briefs of livestreamed meetings to the Midcoast Villager — four legacy weeklies consolidated into one outlet covering 43 towns, paid reporters and volunteer town columnists above the AI. The Villager’s deputy editor, Alex Seitz-Wald, priced it: “I would much rather have anything than nothing.” The reported failure mode is Maine itself: “The Maine accent has been a challenge.”
The counter-case arrived on schedule. In May 2024 the Hoodline network was exposed publishing AI-written “local news” under fake bylines — “Nina Singh-Hudson,” “Sam Cavanaugh,” AI headshots — swapped for small “AI” badges only after another outlet asked. The line to hold: AI that summarizes a real meeting with a human checker is a transparency tool; AI that simulates a reporter is a trust solvent. The honest status of the class: promising, deployed, unevaluated — no peer-reviewed study yet shows machine summaries moving a civic outcome. The economics can close: The Charlotte Ledger, founded 2019 by ex-Charlotte Observer reporter Tony Mecia on a free newsletter platform, reached 5,072 paid subscribers by late 2025 — revenue up 38% in a year, 67% subscriptions, profitable, locally owned.
Meetings are half the books; money is the other half — and the epigraph on this part is a working specification, not an antique. The 1895-lineage acequia statute orders the mayordomo of a community ditch to “make full written reports of all money received, expended and how expended… semiannually,” the books public property, open to all persons concerned. Burlington, Vermont, is what happens without the clause: starting in 2009 the city improperly diverted $16.9 million to keep its citizen-owned fiber network, Burlington Telecom, alive — concealed from council and ratepayers — wrecking the city’s credit and forcing the network’s sale, $30.8 million, March 2019. Burlington did not fail at engineering; it failed at monitoring — the $16.9 million moved without the written, public, semiannual accounting an 1895 ditch statute demands. The oldest clause in this guide would have caught it. Across the same city, the Champlain Housing Trust — the country’s largest community land trust, $260,039,187 in audited assets — keeps its records and minutes “open to, or available for, inspection by any person upon reasonable request.” Two commons, one city, opposite endings. The difference is the clause.
Pricing the clause: the market consolidated in 2024 — Cox Enterprises took majority ownership of OpenGov, nearly 1,900 government customers, at a $1.8 billion valuation — and the transparency module now mostly lives inside a finance-suite subscription. ClearGov, the small-government specialist, serves 1,700-plus local governments and school districts; the one documented contract, county-scale, ran $7,200 in setup plus $15,000 a year — town-scale runs lower, no public price sheet exists, get a quote. Ohio built the public option: Ohio Checkbook, launched December 2014 by the state treasurer, posts line-item spending free for any local government that volunteers — Stow, population about 34,000, links its full checkbook from the city site. A state paying the fixed cost so towns ride free is the cheapest known path to open books. The floor costs nothing: export the monthly check register and general ledger as CSV, post both on the town site, mirror a copy to the Internet Archive — functionally open books, no vendor. No statute in any state prohibits it. The obstacle is custom, not cost.
Know what you are reading when the CSV lands. In a Connecticut town, no officer may spend past a department’s appropriation mid-year — the lone exception, necessary repairs and “the care of the town poor,” capped at $1,000 — and any additional appropriation past $10,000 per department per year ($20,000 above a $20,000,000 grand list) goes to a town meeting, thresholds untouched since 1990. A capital reserve fund under CGS § 7-360 is the patient-money instrument — small deposits, no debt, no bond counsel. Appropriation, exception, threshold, reserve — four columns where the story lives.
One more use of open books: handing residents a slice to spend. Participatory budgeting has run in at least 64 American cities and counties, 258 districts or wards, and 260-plus schools — over $360 million allocated; a floor, not a census, and an undated one. Cambridge drew a record 10,522 voters, age 12 and up, to allocate $1,000,000 — still under 9% of the city. The failure modes deserve equal precision. Turnout decays after the novelty: Vallejo, the first citywide program in the country, allocated more than $8.3 million across five cycles, then watched a month of voting produce 1,909 ballots against a 7,000-ballot goal. Administration can rival allocation: Cleveland’s Issue 38 needed $350,000 of first-year overhead and lost by roughly two points, the council president campaigning against it — programs imposed against a council trigger the institutional immune system; nearly every survivor is council-sponsored. The evidence, read sober: participants’ own trust and civic knowledge rise; community-wide effects remain weakly evidenced. Deliberation mechanics belong to Part V; here lives the precondition — nobody can deliberate over a checkbook they cannot open.
The flat Gallup line is the least examined fact in this part: the one level of American government whose credit held is the one almost nobody measures on purpose. The instruments exist. The National Community Survey — fielded by Polco with ICMA — is the de facto standard: statistically sampled, repeated every year or two, benchmarked, vendor-described at hundreds of municipalities across two decades. The Local Government Trust Index — Polco and Arizona State University researchers under a National League of Cities challenge, launched February 2022, piloted in New Orleans, Broadview, and Dublin — is free to any government that registers. A town of 1,500 needs neither: the questions are public; a clerk or library trustee can field the same six every year and trend the answers. The instrument’s value is the repeat, not the sophistication.
What moves the number? Honestly: little that is proven. Participatory budgeting moves its participants — their trust, their knowledge — and, per NYU Steinhardt’s New York work, council spending priorities; community-wide effects are not established. Polco showcases towns that survey repeatedly and act visibly on results — the vendor’s exhibit; City Bureau reports bodies behave differently with a Documenter in the room — its own survey. What exists nowhere is a peer-reviewed American case of a named intervention durably raising a town’s measured institutional trust across years. A hole in the field is an open lane: the town that fields the same six questions for five years, publishes the trend beside the budget, and writes down what it tried would be the first documented case in the country. The toll is a postcard.
Pikesville opened this part; now place it. Part II named one gate alteration — silent change inside the archive — and answered it with the hash register. The forged voice is the other gate, entry: a false thing arriving from outside, asking to be believed. The gate it walks through is not a server — a person deciding under pressure, in minutes, with no cheap way to check. Build the cheap way to check, and the gate holds.
New Hampshire ran the enforcement experiment. Two days before the 2024 primary, an AI clone of President Biden’s voice told an estimated 5,000 voters to “save your vote” and stay home — commissioned by consultant Steve Kramer for reportedly a few hundred dollars. The FCC finalized a $6,000,000 forfeiture against Kramer and took $1,000,000 from Lingo Telecom, the carrier that moved the calls; a New Hampshire jury acquitted Kramer at trial in mid-2025 on all thirteen felony and thirteen misdemeanor counts, and by late that year he was refusing to pay a related civil judgment. Federal communications law worked; state criminal law did not.
Arizona war-gamed the municipal version — Secretary of State Adrian Fontes’s December 2023 tabletop, CISA, DOJ, and the FBI observing, injected precisely the town-scale threat list: phishing for election-office credentials, a deepfaked official citing a nonexistent court order, AI-generated photos of an election worker committing a crime — and CISA’s companion analysis (January 2024) kept its nerve: generative AI “will likely not introduce new risks, but may amplify existing risks.” The next rung — the forged municipal record, fake minutes, fake ordinance, fake audit — has no verified American case of passing as official, 2023 through 2026. The threat is anticipatory. The countermeasure is not: publish minutes and recordings through one canonical channel — town site, a hash or signed PDF, a same-day mirror to the Internet Archive — so any future “leaked document” can be checked against a timestamped original. The town that establishes the authentic channel before the incident wins the incident.
Provenance tooling helps exactly half of this. C2PA Content Credentials are deployed — 5,000 members in the Content Authenticity Initiative, a conformance program with a public registry since mid-2025, signing cameras from Leica (first, October 2023) through Canon, Nikon, and Sony, Google’s Pixel 10, the provenance mark on LinkedIn. The other half: most platforms still strip the credentials in upload and transcoding — breaking the chain exactly where civic misinformation spreads — most signing cameras omit trusted timestamps, and absence proves nothing, because almost no legitimate content carries credentials either. The civic verdict: sign the outbound record — official recordings, election notices, emergency communications — so the town’s own channel is verifiable; expect no credential to adjudicate a viral clip of the first selectman. For that, the working defense stays boring: a known canonical archive, fast official response, pre-incident relationships with local press and the state election and attorney general’s offices.
Now assemble the part into a job: one volunteer, ten hours a month, a town of 500 to 50,000 — the city renter reads “community board,” “ward,” or “school board” for “town.” Every tool named; every dollar honest.
Component 1 — the watch list (one hour, once). List every public body that spends money in your town: select board or council, school board, planning and zoning, fire district, library trustees, water and sewer. Subscribe to agenda notices — where a town won’t email them, request delivery in writing, citing the open-meetings law. Cost: $0.
Component 2 — the seat (two to four hours a month). Attend or stream one meeting and take structured notes: who was present, what was voted, what numbers were said. The free Documenters Field Guide is the template; inside one of the 25 Documenters communities, the same seat pays $15 to $20 an hour, set per site. Cost: $0 — or negative.
Component 3 — the recording (no extra hours; it runs while you sit). A phone on a $20 tripod — lawful in all fifty states, in Connecticut expressly (§ 1-226). Upload to YouTube; mirror to archive.org. If the town already records, mirror the official file instead — that mirror is the synthetic-record defense.
Component 4 — the transcript and summary (one hour a month). OpenAI Whisper — free, open-source, local — or any free-tier transcription, then a machine-drafted summary a human verifies line by line, under three named rules: the citymeetings.nyc discipline (AI drafts, a human checks), the Chalkbeat rule (a tip, not a story), the anti-Hoodline rule (label the AI assistance; never a fake byline). Cost: $0.
Component 5 — the records reflex (one hour a month). One public-records request on whatever the meeting left unclear — the check register, the contract, the engineer’s report. Free generators: NJFOG’s, the Student Press Law Center’s, iFOIA.org; MuckRock files and tracks for modest fees.
Component 6 — the open-books ask (one hour, quarterly). A standing request to the treasurer: the monthly check register as CSV on the town site. In Ohio, ask the fiscal officer to enroll in Ohio Checkbook, free. Everywhere else, the CSV floor costs nothing but habit.
Component 7 — the publication (two hours a month). A free newsletter — Substack, Ghost’s hosted tier, Buttondown — five bullets of meeting summary, the recording link, one number that matters, the next meeting’s date. The Charlotte Ledger arc is the ceiling; the floor is 40 subscribers who never miss a budget hearing again. In Vermont, post it to the town’s Front Porch Forum.
Component 8 — the trust gauge (two hours, annually). Six benchmarked questions, the same every year, postcard plus QR code — or the free Local Government Trust Index if the town registers. Publish the trend line next to the budget.
Component 9 — the succession rule (ongoing; it outranks the rest). Recruit two co-volunteers by month three; rotate the bodies covered. Every named failure in this part — the hibernating Documenters sites, Vallejo’s 1,909 ballots, the dead family weeklies — is at root a succession failure. The stack survives only if it is never one person deep.
The ledger: $0 mandatory, $20 to $120 a year optional — tripod, domain name — about ten hours a month. Everything above is deployed, named, and free as of June 2026: nothing piloted, nothing announced-only. Both gates are covered — the mirrored recording and canonical channel guard entry; the records reflex and open checkbook guard what moves inside. Twenty years of watching went dark for want of a business model. Ten hours a month turns it back on. The lane is still empty, and the deed costs a calendar, two recruited names, and a note template.
For the selectman, the treasurer, the clerk:
For the renter in the city:
Part IV
“For over a week, ham radio was our only reliable link to the outside world.”
— Thomas Witherspoon, K4SWL, Swannanoa, North Carolina, September 27, 2025
The rain was still falling when the Swannanoa River came out of its banks east of Asheville, North Carolina, on the morning of Friday, September 27, 2024 — and the first thing the flood took from the people above the waterline was the means of saying anything to anyone. Helene’s remnants destroyed more than 1,700 miles of fiber-optic cable across western North Carolina; the towers that stood lost the lines behind them. By Saturday the Federal Communications Commission’s disaster reporting had counted it: 74.31% of cell sites in the North Carolina disaster area out of service on September 28, and two of every three — 66.4% — still out the day after that. In Avery, Mitchell, and Yancey counties, fewer than 10% of sites were functioning at the trough. Seventeen 911 call centers lost the ability to receive calls on September 28; by Sunday the carriers were reporting complete loss of cellular voice and data across 11 western counties, an estimated 150,000-plus people in total technological isolation — 19 of North Carolina’s 100 counties cut off outright. On October 7, 209 of 1,448 impacted North Carolina sites — 14.4% — were still down. For days the valley smelled of river mud and chainsaw exhaust, and iPhones in the coves showed a single word — SOS — where the bars had been.
In one Swannanoa community of eighteen to twenty households, the storm found exactly one or two licensed amateur radio operators. One of them was Thomas Witherspoon — call sign K4SWL — and for the next week his family ran the valley’s switchboard: passing traffic, performing wellness checks, pressing spare handhelds into unlicensed neighbors’ hands, all of it lawful because the federal rules — 47 CFR §§ 97.403 and 97.405 — permit any means of communication in a genuine emergency. “By definition, this was a communications emergency, so no license was required,” he wrote, one year to the day after the water came — and then the sentence printed at the head of this part, the one this part exists to generalize. The link ran through the N2GE repeater on Mount Mitchell — the highest peak east of the Mississippi, one of the few mountaintop machines with hardened backup power — into the Tar Heel Emergency Net, which North Carolina ARES had activated at the state’s request the night of landfall, and onward through high-frequency nets at 7,232 and 3,923 kHz and Winlink radio email, which the ARRL reported was “used significantly in the recovery.”
Count the layers that carried traffic that week, in the order they held. The state’s hardened public-safety radio network, VIPER, “remained online throughout the storm” — Highway Patrol helicopters refueled mountaintop towers, and the state pushed more than 900 cache radios to responders. Next-generation 911 rerouted around the wreckage: survivors with any signal at all were answered by call centers hundreds of miles away, and the Newton Communication Center in Catawba County fielded more than 6,000 calls between September 27 and October 1 for its stricken neighbors. AM/FM broadcast carried the rest — emergency managers went live on the air, and residents called in stranded-person reports and proof-of-life messages. Ham repeaters carried traffic out of the valleys; GMRS handhelds knitted the neighborhoods together. Twenty-four counties eventually received Starlink kits by National Guard and law-enforcement delivery, and iPhone satellite-SOS appeared organically in the field.
And one honest absence, which this part will neither hide nor excuse: no LoRa mesh network carried Helene traffic, because no organized mesh existed in western North Carolina before the storm. The volunteer groups now growing one there — MeshAVL among them — organized afterward. Witherspoon’s community drew its conclusion and institutionalized it, adopting a GMRS community radio network in the weeks after the water dropped.
Greg Hauser, North Carolina’s statewide interoperability coordinator, compressed the planning lesson into one sentence: “When a catastrophe destroys the emergency communications ecosystem, it is unrealistic to assume that 20 satellite phones or 100 low earth orbit satellite terminals are obtainable.” You cannot order the stack after the storm. You can only have built it.
The failure does not require weather. On February 22, 2024, at 2:42 in the morning, a configuration error during a routine network change took AT&T Mobility down nationwide for at least twelve hours: more than 92 million phone calls blocked, more than 25,000 failed attempts to reach 911, over 125 million devices affected, in all 50 states plus the District of Columbia, Puerto Rico, and the U.S. Virgin Islands. The FCC’s July 22, 2024 report blamed the company’s failure to follow its own internal procedures, and referred the matter for enforcement. No hurricane, no flood, no saboteur — the continental platform your town’s civic life runs on failed on a clear day, by its own hand.
Fourteen months later the control experiment ran at continental scale. On April 28, 2025, a cascading grid collapse left more than 55 million people in Spain and Portugal without power for more than half a day, and with the power went the mobile networks and the card readers — two countries discovering in a single afternoon that the phone and the payment terminal are wall-powered conveniences. Amateur operators had nets running on repeaters and the HF emergency frequencies within hours; battery radios and a neighbor with a software-defined receiver became the neighborhood information sources. The half-day continental blackout is the cheapest realistic stress test there is: anything that needs wall power or a cell tower is not a resilience layer.
When the Eaton and Palisades fires killed 31 people and destroyed 16,251 structures in January 2025, Los Angeles County’s own commissioned after-action review found its alerting practices “outdated, unclear and contradictory,” with responders unable to share information consistently across failing cell connections and unconnected platforms. The layer that demonstrably carried public situational awareness was a nonprofit app, Watch Duty — roughly 15 staff and some 200 volunteers — which onboarded twice as many users between the Tuesday outbreak and that Friday as it had in its entire prior existence. Civic infrastructure can be an app with a governance model as surely as it can be a repeater.
Across the four major stress events of 2024–2026, the empirical ranking of what worked runs: broadcast AM/FM with backup power; licensed amateur radio — hardened repeaters, HF, Winlink; GMRS and FRS at street scale; satellite where deliverable; volunteer-run information services. Commercial cellular and fixed broadband were consistently first to fail and slowest to return. LoRa mesh remained promise rather than performance. And in two of the four cases the binding failure was institutional — alerts not sent, sirens not built — not technological.
Kerr County, Texas is the proof of that last clause, and the reason this part belongs on a selectboard agenda rather than a hobbyist’s bench. On July 4, 2025, the Guadalupe River flood killed more than 100 people. Kerr County sent no wireless alerts through FEMA’s national system as the flooding began; its opt-in CodeRED system was never fully activated; it had no flood sirens — a warning-system proposal had been drafted, then the sirens were cut after public pushback, the state denied grant funding, and a Texas bill to fund outdoor warning systems failed over cost objections weeks before the flood. The county’s new Motorola emergency radio system failed responders in the aftermath, and its top two emergency-management officials were asleep through the critical pre-dawn hours. Every layer of that failure was a decision, made in a budget meeting, years before the rain.
You Cannot Eat Code mapped the seven communication layers of 1776 — pamphlet, letter, sermon, tavern, militia, post, oath — onto their modern descendants and argued that a town must own them again; this part cashes the argument out in hardware, licenses, wire, and software, each priced. Four layers, then three builds, scaled to both readers — the town actor on the water tower, the renter on the roof of the building. Your town is answering the Kerr County questions right now, mostly by not answering them.
Start with the cheapest and newest layer, because it is the one a town can field this season. Meshtastic is open-source firmware that turns cheap LoRa radios into an encrypted, license-free text-messaging mesh on the unlicensed 915 MHz band — every node rebroadcasts every message, no tower, no subscription, no internet. Its physics are honest and narrow: 256-byte packets, roughly 200 bytes of usable payload — short text and GPS positions, not voice, not images. The official ground-to-ground range record stands at 331 kilometers — 205 miles, mountaintop to mountaintop, in Europe — and it is misleading for planning: practical town coverage comes from elevation — water towers, ridgelines, grain elevators — while handheld-to-handheld in wooded or built terrain is blocks to a few miles. Congestion is the second hard limit: a real mesh past 150 active nodes has hit channel-utilization peaks over 65% on the default preset and dropped messages, so past about 100 nodes somebody has to govern channels and presets on purpose.
The hardware is almost embarrassingly cheap. The consensus first node, the Heltec V3, costs about $25 and flashes from a web browser. Seeed’s SenseCAP T1000-E — $39.90, credit-card-sized, weather-rated, GPS aboard — ships with Meshtastic pre-loaded, and reviewers call the $46.99 Wio Tracker L1 Pro the easiest open-the-box handheld. Solar ridge and rooftop repeaters are built on the RAK4631 family — modular boards that sleep at microamps.
The institutional proof-of-concept is in Florida. In March 2025, Bob Jones, program manager of the Jefferson County CERT, went before the Monticello city council with a plan to mesh the county: small solar-powered nodes at about $100 each — device, panel, battery, enclosure — placed on top of water towers by drone, grant-funded through Volunteer Florida and others. “The idea with Meshtastic is they are small devices that we can deploy easily to reestablish a network within hours after a storm if we had lost all other methods of communication,” Jones told WTXL’s Lentheus Chaney. A county communications layer for the price of a used pickup — run by the volunteer emergency corps, not the carrier.
Now the honesty requirement, stated plainly: as of June 2026, no United States after-action report shows a LoRa mesh carrying primary disaster traffic. Not one. The 2024–26 record is post-hoc adoption — communities building meshes after the storm taught them the gap — and Jefferson County is the clearest pre-disaster institutional build in the country: grant-funded, CERT-run, and not yet tested under load. That does not argue against the layer; it defines it. A flood-routed mesh is its pre-positioned nodes — there is nothing to improvise after landfall — so pre-deployment is the entire game. Helene’s mesh would have had to exist on September 26.
Two governance notes belong ahead of the first purchase order. First, the protocol split: since 2025 a competing open firmware, MeshCore, has routed through designated repeaters instead of flooding, pitched at dense urban backbones — and the two are mutually incompatible. Meshtastic keeps the far larger community — 40,000-plus GitHub stars, 80,000-plus subreddit members, against MeshCore deployments clustered in a few cities as of early 2026 — so a town picking its mesh in 2026 picks one protocol and writes the choice down; fragmentation is now a real failure mode. The volunteer metro meshes — Austin Mesh, BuffaLoRa in Buffalo, NodakMesh in North Dakota, NH Mesh in New Hampshire — publish the guides and maps worth copying. Second, the law: LoRa devices run unlicensed under FCC Part 15, and a mesh carrying no commercial broadband service generally sits below the threshold that even the restrictive states regulate — but where a statute exists, ask first; the state broadband office, the municipal league, and town counsel are who answer.
The licensed layer is the one with the 2024 battle record, and its on-ramp costs about a tank of gas. The structure is two-tier. Amateur radio requires an exam and returns the heavy machinery — repeaters, HF long-haul out of a cut-off valley, Winlink email-over-radio. GMRS requires no exam, costs $35 for ten years, covers the licensee’s immediate family on one license, and supports local repeaters. The proven post-Helene pattern: licensed hams carry traffic out of the valley; GMRS handhelds knit the neighborhood together.
The prices, precisely. A ham Technician license: a 35-question multiple-choice exam, $15 at a volunteer-examiner session — some teams charge nothing — plus the FCC’s $35 application fee; call it $50 all-in, good for ten years, free study at HamStudy.org. A GMRS license: $35 to the FCC, no exam, ten years, the whole household — the fee was cut from $70 in 2022. Entry hardware: the Baofeng UV-5R runs $17.20 to $22.99, with the honest caveat that its transmit quality varies and clubs steer reliability buyers toward $100–$200 handhelds. A repeater now arrives in a box: the Retevis RT97 at roughly $360–$400 — $500.99 for the higher-power model — puts a functioning neighborhood repeater, rooftop antenna and feedline included, well under $1,000; the vendor claims 20-plus miles of range with proper antenna placement, and terrain will decide what you actually get. A Winlink-capable HF station — radio, modem, antenna — runs roughly $1,000 to $2,500.
But the hardware is the cheap half, and the Helene record carries three warnings about the people and the paper. The licensing funnel: Witherspoon’s community had one or two licensed operators across eighteen to twenty households before the storm, and it was multiple licensed operators per family that made continuous net coverage possible — a plan that assumes “the ham will do it” fails when the ham is clearing trees. The repeater problem: western North Carolina’s traffic concentrated on the few mountaintop machines with generator or solar backup, and repeaters without hardened power went silent. And the paper: the instrument that turns a radio club into a town capability is a signed memorandum. ARES is the amateur league’s volunteer field organization; RACES — 47 CFR § 97.407 — is amateurs operating under government emergency-management control; the bridge between them and your town is an MOU. The reusable template is Gaston County, North Carolina, where ARES executed a Memorandum of Agreement with the county Emergency Management Department on November 23, 2016, formally creating the county RACES — signed by the Assistant County Manager, recorded by the Clerk of the Board. It had to be created at all because the structure had lapsed — RACES is dormant in many counties, and a town that has never signed has no legal hook to task its volunteers when it matters. Fifty dollars licenses the operator. A signature deputizes the club.
The city runs the same spectrum at block scale. California fire country’s Community Emergency Radio Association operates Neighborhood Radio Watch programs with weekly on-air check-in nets — training residents on the calm days — and Butte County runs a community GMRS network on the same pattern. A renter or a block association needs three things: the $35 license, a $25–$50 GMRS handheld, and the local net schedule from the myGMRS repeater directory. A building is a neighborhood stood on end; a weekly net is how it learns its own names.
Mesh and radio are the storm layers. The everyday layer is wire — and the wire, it turns out, is political.
For the city reader the standing proof is NYC Mesh: more than 2,000 active member nodes across the five boroughs, organized under a Network Commons License rather than customer contracts — a volunteer rooftop install at a suggested donation of $290, $160 subsidized, or $40 a month for six months; service afterward is donation-based, $20 to $60 a month suggested, free if you cannot pay; members keep their rooftop routers powered so neighbors connect through them, and the network reaches the internet through fiber-fed supernodes at the exchange points, no commercial ISP in the path. On June 3, 2026, Public Advocate Jumaane Williams released a 53-page report calling for citywide low-cost municipal internet “akin to a public utility” — an NYC Mesh representative standing at the press conference. The pattern scales: guifi.net in Catalonia, the largest community network on Earth, counted roughly 37,000 active nodes and 71,000 kilometers of links as of December 2021.
For the town reader the wire is municipal, and more than 500 American communities are already served by some form of municipal network. The flagship is Chattanooga: EPB built the country’s first citywide gigabit fiber-to-the-home network in 2010 for approximately $220 million, and an independent study by Bento Lobo at the University of Tennessee at Chattanooga found $2.69 billion in community benefit in the first decade — 9,516 jobs created or retained, roughly 40% of the county’s job creation in the period; power-outage minutes cut 40–55% a year, worth about $26.6 million annually; $59.9 million paid to local governments from 2011 to 2020 — with the follow-up study putting the cumulative figure at $5.3 billion. The wave has reached town scale: Traverse City, Michigan — population 15,424 — is finishing a $14 million citywide build; Vienna, Maine — 600 residents — launched its own municipal fiber network in January 2026; Utah’s UTOPIA open-access consortium passed 67,000 subscribers.
Now the two ways the wire fails. The first is financial, and mostly historical: iProvo put roughly $40 million of public money into citywide fiber and sold it to Google Fiber for $1 in 2013, and the defensible reading of the wreck ledger is that the failures cluster in early-2000s business models. The second is political, and current. Sixteen states preempt or severely restrict municipal broadband: Alabama, Florida, Louisiana, Michigan, Missouri, Montana, Nebraska, Nevada, North Carolina, Pennsylvania, South Carolina, Tennessee, Texas, Utah, Virginia, Wisconsin. The mechanics deserve a town counsel’s attention: Pennsylvania requires a town to first ask the incumbent telephone company’s permission; Nevada bans municipalities above 25,000 population from offering telecom at all; Nebraska bans it outright for non-power entities; North Carolina’s 2011 requirement stack has produced zero new municipal networks since. The direction of travel is loosening — Arkansas and Washington rolled back barriers in 2021, Colorado repealed its referendum requirement in 2023 after more than 120 communities had voted to opt out, Minnesota repealed in May 2024 — and Connecticut is not on the list: a Litchfield County town faces process friction over pole access, not prohibition.
What the fight looks like from inside a town: Wilson, North Carolina built Greenlight through its electric utility, was confined to its home county by the 2011 state law, was briefly freed when the FCC preempted the North Carolina and Tennessee restrictions in February 2015, and extended gigabit service to the town of Pinetops and a nearly-300-employee family farm — until the Sixth Circuit ruled, on August 10, 2016, that the FCC had no such authority. Facing an October deadline to cut Pinetops off, Wilson’s council voted to give the service away free for six months — the statute restricted selling, not giving. A 2017 special law let Wilson serve Pinetops only until a private provider arrived; when Suddenlink built in, Greenlight was forced out. The same ruling re-caged Chattanooga, whose gigabit utility remains confined by Tennessee statute to its electric footprint, barred from the unserved counties next door.
The freshest failure is the one to study before your own town meeting, because it was political end to end. Longmeadow, Massachusetts, May 12, 2026: voters rejected, 374 to 270 — short of the required two-thirds — an $8.6 million loan for phase one of a $27 million town fiber utility: a central hub plus a pilot serving about 1,600 homes and 50 businesses, funded by a $97-a-year property-tax increase. The town finance committee had declined to endorse it over debt risk. And a Minnesota-based 501(c)(4) dark-money group, “Mass Priorities” — an extension of the Domestic Policy Caucus — ran calls, texts, and mailers against the project while town officials were legally barred from campaigning for it. Dark money discovered the town meeting before most towns discovered the dark money; in Arrowsic, Maine, the incumbent began undercutting the town network it had refused for years to build. Gigi Sohn of the American Association for Public Broadband drew the line plainly: “Communities like Longmeadow deserve an honest debate about costs, governance, and accountability, not recycled myths designed to spread misinformation, protect the status quo, and discourage local choice.”
Do not plan around federal rescue. The $2.75 billion Digital Equity Act was ended by social-media post on May 8, 2025, termination letters out the next day. BEAD was restructured in June 2025 and finally moves — first construction expected in the summer of 2026 — but Connecticut’s arithmetic shows the shape of the help: a $144 million allocation, of which the state’s final proposal awards $7.2 million, to two incumbent ISPs — Southern New England Telephone $6.5 million, Comcast $643,299 — for 724 eligible locations. The wire your town wants is the wire your town will vote for, fund, and defend — against opponents who do not live there.
The fourth layer is software — specifically, the software a committee can keep running when the vendor, the platform, or the network goes away. The term of art is “local-first,” coined in an April 2019 essay from the research lab Ink & Switch: software that works offline, syncs when it can, survives the vendor, and keeps its data in files the user controls — built on CRDTs, conflict-free replicated data types, the mathematics that merges concurrent edits with no central server to arbitrate. This is the book’s correspondence layer rebuilt: letters that can arrive in any order and still assemble the same record. The math is production-grade now — Automerge 3.0, released in 2025, cut memory more than tenfold; a document the size of Moby-Dick that took 700 megabytes of RAM in version 2.0 takes 1.3 megabytes in 3.0, and a pathological document that took 17 hours to load opens in 9 seconds.
A town committee does not need the math; it needs the finished tools, and the working shelf is short. Obsidian keeps minutes, procedures, and rosters as plain markdown files on every member’s machine — the app free, official sync $4 a month — a vault each member holds in full, offline, forever. Syncthing replicates a document folder peer-to-peer with no server at all, free and actively maintained. CryptPad is the closest thing to Google Docs without the platform — an end-to-end-encrypted office suite, free on public instances or self-hosted, a fully managed instance from €1,500 a year with a 50% nonprofit discount on the 50-user tier — with its caveat said out loud: it survives the platform because you hold the server, but it is not offline-first, and it does not survive a dead network. The shared-database layer — Automerge and Yjs themselves — remains a developer toolkit rather than a product; a town without a developer volunteer chooses from the finished apps. Carry the cautionary tale too: Syncthing’s official Android app was discontinued in December 2024, its maintainer citing “a combination of Google making Play publishing something between hard and impossible and no active maintenance” — even local-first software has a platform dependency at the distribution layer. And no US town government has yet been found running CryptPad or an equivalent self-hosted suite as its committee workspace — an absence after search, and an open lane.
Then Nostr, soberly, because some reader’s nephew will propose it. Nostr is the censorship-resistant broadcast protocol — the pamphlet layer — and its real numbers deserve daylight: against some 33.5 million keypairs ever created, the spam-filtered count in October 2025 was 21,281 trusted users worldwide — 3,675 active on a given day, 14,777 in a month. Relay infrastructure shrank by roughly a third in the year to April 2024, to about 639 online. It is not where your residents are; it is not a moderated forum; and it requires working internet, so it is not an emergency channel either. What it is: publisher survivability — a keypair no platform can delete, content replicated across willing relays — which makes it an honest mirror for the town’s newsletter, minutes feed, and notices, never a primary. A town relay runs on strfry on a $5–$20-a-month server — under $5 on a Raspberry Pi — plus a $12-a-year domain; one Linux-comfortable volunteer, one afternoon. Run it whitelisted, accepting posts only from the town’s published keys — clerk, selectboard, fire company, library — because the protocol has no moderation layer at all: an open relay hosts whatever arrives until somebody notices, which is unacceptable exposure for a town, and the network’s main working spam defense is, of all things, a paywall. The whitelist is the only configuration a town attorney should accept. No US municipality has yet been found operating an official relay — by now you know what that means. The mirror is cheap, legal, and unclaimed.
Now the bills of materials — three reference builds, scaled to a household, a fire company, and a selectboard. The named hardware exists and is purchasable as of June 2026; line items marked (est.) were not individually price-verified, and the two largest want two local quotes before they enter a budget.
Build A — the $470 neighborhood node. Renter-compatible: one household plus its immediate neighbors. The bill: GMRS license, $35; two GMRS handhelds, $80 (est.); a battery-and-crank AM/FM/NOAA weather radio, $50 (est.); two Meshtastic nodes — the $25 Heltec V3 and the $39.90 SenseCAP T1000-E, both price-verified — $65; a 100-watt-hour power bank with a 20-watt folding solar panel, $120 (est.); a window or roof GMRS antenna with cable, $60 (est.); one ham Technician license, $50; a laminated channel card, neighbor roster, and the local net schedule, $10. Total ≈ $470. Skills: smartphone literacy and browser flashing — zero soldering; one afternoon to assemble, one evening to walk three neighbors through channel 1. Maintenance: charge the batteries quarterly and check into the local GMRS net. This build is the empirical stack ranking in a shoebox — broadcast receive, licensed voice, street-scale voice, and a mesh seed.
Build B — the $4,665 town emergency mesh. CERT or volunteer-fire scale, town of 500 to 5,000. The bill: RT97 GMRS repeater, $400; base antenna ~$300, fifty feet of LMR-400 feedline $120, mast and mount $150 — $570 (est.); repeater power — 100-amp-hour LiFePO4 battery, charger, 100-watt solar panel, weatherproof box — $600 (est.); ten solar Meshtastic relay nodes at the Jefferson County benchmark of about $100 deployed, $1,000; twelve GMRS handhelds, $480 (est.); two dual-band ham base stations — emergency operations center and firehouse — $800 (est.); a VHF Winlink packet station, $400 (est.); licensing and training — six Technician licenses $300, the town’s GMRS license $35, written procedures and a first drill $80 — $415. Total ≈ $4,665. Skills: two to three licensed Technicians — the bottleneck, so start the license class before the purchase orders; one browser-flashing volunteer; a firefighter or electrician for the mounts; a monthly net coordinator. And the governance line sits in the build instructions on purpose: sign the ARES/RACES-style MOU with the town before buying hardware — the Gaston County model.
Build C — the $50,000 town-wide layer. A selectboard budget line, town of 5,000 to 50,000. The bill: hardened repeater site, $15,000 (est. — quotes vary widely; get two); emergency-operations-center radio room, $8,000 (est.); a LoRa backbone of six high-site solar router nodes at about $250 a site plus forty subsidized citizen T1000-E nodes, $3,100; neighborhood caches — forty GMRS handhelds and four portable RT97 repeaters — $3,200; satellite failover — three Starlink kits at the verified $499 plus twelve months of service at $85 a month each — $4,557; a durable-records layer — CryptPad’s managed nonprofit tier at €1,500 ≈ $1,600, a Nostr relay server at $240 a year, two mini-PCs as Syncthing/Obsidian replicas $600 — $2,440; a broadcast bridge — an MOU with the local AM/FM station, a backup studio link or low-power-FM feasibility study — $2,500 (est.); training and exercises — twenty Technician licenses $1,000, CERT integration, two tabletops and a field drill a year — $3,000; resilient power — two 2-kilowatt-hour battery stations with panels — $4,000 (est.); spares, lightning protection, and contingency at roughly 8%, $3,700. The line items sum to $49,497 — call it $50,000 one-time — plus about $5,500 a year recurring. Skills: a standing MOU with the county ARES or radio club; one IT volunteer or a budgeted MSP hour; an electrician; an emergency-management director who owns a drill calendar.
And the money. A capital reserve fund started this year buys Build C in three — small annual deposits, no debt, no bond counsel. Civic-experiment micro-grants of $5,000 to $25,000 are Build-B-sized to the dollar; resilience-framed versions fit FEMA’s reopened pre-disaster mitigation pool — $1 billion for FY2024–25, applications due July 23, 2026 — and EMPG pass-through money at 50% match through the state emergency-management agency; communications equipment for a town hall or firehouse fits USDA Community Facilities, up to a 75% grant for low-income towns of 5,000 or fewer. Jefferson County’s water-tower nodes were grant-funded through Volunteer Florida — state service-commission money — and fire-department auxiliary funds round out the list.
Stand the two towns side by side one last time. Kerr County had drafted the warning-system proposal; the sirens were cut after public pushback, the grant was denied, the funding bill died over cost objections — and on the night the river rose, the official layers had nothing to say. Swannanoa had a repeater with hardened power on the highest peak in the East, a statewide net activated the night of landfall, one family with licenses — and handhelds to spare for the neighbors. The difference between those two valleys was not weather, wealth, or luck. It was inventory pre-positioned and paper pre-signed — decisions made in advance, in rooms like the one where your selectboard meets next month. The stack in this part costs $470 to seed, $4,665 to stand up, $50,000 to finish — against a county radio system that failed when it was finally needed, and a $27 million wire that died at a town meeting it never saw coming. Buy the layers in the order the storms rank them. Sign the MOU before the hardware. And put the first node where Jefferson County put theirs — up on the water tower, where the whole valley can reach it.
For the selectman, the fire chief, the emergency-management director:
For the renter in the city:
For either reader:
Part V
From talk to a decision that binds — five formats, priced, and the clause that keeps them honest.
On the night of November 4, 2025, Fort Collins, Colorado counted ballots for a piece of ground where Hughes Stadium used to stand, and the count was not close: 25,422 to 12,235 — 67.5 percent — for ballot Question 2H. What the voters ratified was not a developer’s plan and not a council compromise. It was the recommendation of twenty residents chosen by lottery.
Lay the chain out link by link, because every link is ordinary and only the welds are rare. The city had acquired the stadium site in 2023, under a 2021 ballot measure requiring the ground rezoned for open lands. Two community surveys, 2022 and 2023, failed to produce consensus. On August 20, 2024, City Council voted 3–2 to convene a civic assembly. Twenty delegates — drawn by civic lottery, the selection posted in three public livestreams — met for two weekends, April 12–13 and May 3–4, 2025, run by Healthy Democracy with Colorado State University’s Center for Public Deliberation, trained “Community Guides” having gathered public input through February. Twenty-two recommendations cleared the assembly’s 75 percent threshold, four unanimous, support for a multi-use vision running near 90 percent. Council referred the plan to the ballot as Question 2H — up to 60 acres of natural area, 30 acres for environmental education and wildlife rehabilitation, 35 acres of city park. On November 4 the voters made it law, better than two to one.
Two surveys failed; a lottery, a deliberation, and a ballot succeeded. That is the cleanest deliberation-to-law chain yet run in the United States, and the only exotic thing in it is the last link. American towns are drowning in input — surveys, listening sessions, comment periods — and starved of decisions. The book this volume accompanies argued that the town is the unit of account; a unit of account must be able to decide. So this part is about deliberation that ends in something binding — a bylaw, a budget line, a ballot question — and about the documented way deliberation dies instead: the report received with thanks and shelved. It runs for both readers: the reader whose town meets on a floor with a warrant, and the city renter whose nearest instruments are the community-board hearing and the participatory-budgeting ballot. The mechanics differ. The design problem — how does talk become a decision? — is the same.
Start with the original American instrument, because we have better data on it than on anything built to replace it. Over three decades, 1970 through 1998, the political scientist Frank Bryan and his University of Vermont students attended and coded about 1,500 Vermont town meetings — 238,603 individual acts of participation by 63,140 citizens in 210 towns, published as Real Democracy in 2004. Average attendance: 20.5 percent of registered voters. The range ran from 94 of 130 registered voters — 72 percent — in Newark, Vermont in 1974, down to 36 of 3,600 — 1 percent — in Jericho, a Burlington bedroom town. The variable that predicted the spread was not the one anyone assumes: aggregate wealth and education did not predict attendance. Size did — smaller towns turn out dramatically higher shares — and the law replicates next door, where a 2012 New Hampshire study found a correlation of −.62 between voting-population size and attendance percentage. Deliberation is a function of scale. And nobody is gauging it: that study covers just 27 towns “due to a widespread lack of attendance figures,” and no statute anywhere in New England requires a town to report meeting attendance to anyone. The oldest deliberative institution in the country runs without instruments.
What the instruments would show is conversion. Between 2019 and 2025, the share of Vermont municipalities deciding budgets by Australian ballot — all-day secret ballot, no floor debate — rose from 28 percent to 37; officer elections by ballot went from 66 to 68 percent; public questions from 40 to 45. The structural story is not abolition. It is a deliberating legislature converting itself into a ratifying electorate. The trade is priced: Séan Sheehan, Vermont’s director of elections, reported the seven municipalities that mailed ballots to all active voters for the 2025 annual meetings averaged 36 percent weighted turnout — Burlington 34, Strafford 50 — roughly double Bryan’s floor average. Ballots raise headcount. Floors produce deliberation. Susan Clark, Middlesex’s longtime town moderator and co-author of All Those in Favor, names the tension: “One of them is democratic quantity, like, how many people are involved… The other one is democratic quality, like, what is the democracy that we are inviting people into?” And names what the ballot-only version forfeits: “When the only thing we ask people to do in a democracy is cast a ballot, that’s an impoverished view of what democracy is… Town meeting is a legislature. We weigh the pros and cons together. We can change each other’s minds.”
Towns are running the argument live, in both directions. Bedford, New Hampshire converted in 2024, after a school-board member watched roughly 4,000 residents vote in the 2023 ballot elections while “50 or 60 people would show up” the next night to pass the budget on the floor; the ballot question carried 74 percent. Enfield went the other way — adopted the ballot system in 1996, then returned to the floor on March 13, 2001 after deliberative attendance collapsed, the minutes recording that “the ‘old style’ town meeting allowed for a greater number of voters to be educated on the articles.” Bow has rejected the switch six times, most recently in 2021, 426 of 537 attendees voting to keep the floor; a third of New Hampshire’s 221 towns have adopted ballot voting since the 1995 law. The instrument is not dying so much as standing for election, town by town.
Where it fails, the failure is specific — and recoverable. West Tisbury, Massachusetts, population about 3,500, quorum set at 5 percent of registered voters — about 130 people — failed to make quorum at an October 2024 special town meeting called partly to buy a wildfire-capable fire truck, the town’s second such failure. The select board proposed cutting the quorum to a flat 100. At the April 2025 annual meeting the voters rejected the cut from the floor — and over 300 were in the room to do it, nearly triple quorum, after a warrant with real stakes (a leaf-blower phase-out, school spending) and a new outreach push: text messages, email blasts, online announcements. Both edges teach. The quorum failure was real, and it was cured with content plus contact — not a lower bar.
The contact half of the cure carries a quarter century of randomized evidence, beginning in Connecticut: Gerber and Green’s 1998 New Haven field experiment found in-person canvassing substantially raised turnout while commercial phone banks did nothing. The accumulated experiments price the door at roughly one additional vote per 14 people contacted, about $19 a vote — and price the relationship far higher: a voting nudge texted by a friend raised turnout 8.3 percentage points where a mass text from a stranger raised it 0.29, a factor of about 28, replicated at 8.6 points in Florida’s 2024 cycle. Being asked, by a person who knows your name, has been a first-order predictor of participation since Voice and Equality in 1995. A town that wants its floor back needs no campaign budget — it needs a contact list, a warrant with power on it, food, childcare, and askers. The personal ask is the hinge rung of Part X’s ladder; this is the room it fills.
Which brings the moderator’s rule, this part’s first law. Clark: “People will come to meetings that have power. They won’t bother going to meetings that don’t.” Hold that sentence against every advisory committee you have watched die — and against the hybrids too. Jericho, Bryan’s 1 percent town, now runs January budget hearings where voters shape amendments, then a March ballot; its own clerk reports the design contested — “There’s no information at the polls. People need to know their information ahead of time.” Clark endorses the inverse hybrid: officers by ballot, budget and public questions on the floor — headcount where headcount belongs, power where deliberation lives.
For the question a floor meeting handles worst — the thorny, values-laden one that deadlocks the elected body — the last decade built a second instrument: the civic lottery. The OECD counted 733 representative deliberative processes worldwide as of December 2023. Three lineages carry the weight.
Ireland proved a lottery could move a constitution. The Constitutional Convention of 2012–2014 — 66 randomly selected citizens seated with 33 politicians — recommended marriage equality, and the referendum of May 22, 2015 passed with 62.07 percent, the first country to adopt same-sex marriage by popular vote. The Citizens’ Assembly that followed — Ms. Justice Mary Laffoy in the chair, 99 citizens stratified by gender, age, location, and social class — deliberated the Eighth Amendment over five weekend sessions; its recommendation passed at referendum on May 25, 2018 with 66.4 percent. That assembly’s official budget, to the cent: €1,505,960.90; the program’s first decade ran over €6.8 million. Expensive — and cheap against the stalemates it broke.
Ostbelgien made it permanent. On February 25, 2019, the parliament of Belgium’s German-speaking community — about 78,000 people, a polity the size of a county — unanimously created the first standing, legally mandated citizen deliberation attached to a legislature: a 24-member Citizens’ Council picking topics and monitoring implementation, per-topic assemblies of 25 to 50 residents drawn by lot, participants paid €155 a day. The teeth are the point: ministers and parliamentarians must publicly respond to each recommendation and meet the citizens again within one year to account for implementation. Five years in, the monitors’ findings read like shop notes: early recommendations failed because they were “too vague… to make into a law or decree” — the fix was teaching citizens to write justifications a drafter could legislate from — and the calendar fell to about one assembly a year because staff could not follow up faster. Deliberation is bottlenecked by follow-through, not enthusiasm.
Paris proved a permanent assembly could write law in a major city. Its Assemblée citoyenne — 100 residents by civic lottery, non-citizens eligible — worked homelessness from September 2023 to April 2024 through 4 plenaries, 16 half-day workshops, 7 site visits, and 3 drafting sessions, reached unanimity, and on July 10, 2024 the Conseil de Paris passed the citizens’ bill directly into law: 20 measures, including extending youth accompaniment from age 18 to 25 — the first time a major city council enacted legislation drafted by a permanent lottery-selected assembly.
The American versions are younger, smaller, and instructive in their modesty. Petaluma, California ran the state’s first municipal civic assembly in 2022: $450,000 budgeted, 36 delegates stratified down to housing status and disability, about 90 hours of deliberation on the fairgrounds’ future. The adoption trail is the honest part — guiding principles that October, a lease keeping the annual fair alive in December 2023, the city managing the grounds from January 1, 2024. Partial, slow, real. Akron, Ohio drew 65 housing-assembly delegates from 584 lottery entrants in February 2026 — $1,000 stipend plus transportation, childcare, meals, and language support, outcome pending as this volume goes to print. Mark the arithmetic, renter-reader: 65 seats from 584 respondents. The lottery letter that looks like junk mail carries roughly one-in-nine odds of a seat at the table.
And now the instrument arrives on this book’s home ground. Next month — first session July 11–12, 2026, on Yale’s campus in New Haven — the Connecticut Citizens’ Assembly on Property Taxes convenes: 100 residents drawn by lottery from invitation mailings, paid $1,200 for full participation, six sessions running through September 26 in New Haven and Hartford with two virtual sessions between. It is led by State Comptroller Sean Scanlon and the Connecticut Conference of Municipalities, with Yale’s Josh Kalla and Hélène Landemore and UConn’s Michael Morrell on the design, and it asks one question: “Should Connecticut change how we pay for and deliver local services like education, road maintenance, or public safety?” The recommendations go to a State Legislature hearing, presented by the Comptroller — billed in advance as “the primary evidence base for municipal funding reform” in a state that, by the organizers’ own accounting, hasn’t meaningfully reformed how it funds local services since 1972. There is no outcome to report; the first session has not yet been gaveled in. So watch it with the checklist this part builds: whether the recommendations come out specific enough to legislate from — the skill Ostbelgien had to teach — whether the hearing keeps its date and the findings their named sponsor, which is the Comptroller’s role by design — and whether anything with a bill number emerges the following session. A lottery assembly sits upstream of the law. Downstream is where these things die, and the next section is the autopsy.
On March 8, 2024, Ireland — the country that proved the lottery — recorded the largest “No” in the history of its constitution. The Assembly on Gender Equality had recommended replacing the constitution’s “women in the home” language and broadening the definition of family; the government put diluted wordings to referendum and campaigned half-heartedly; the family amendment lost with 67.7 percent against, the care amendment with 73.9. An assembly confers no immunity at the ballot box — least of all when the government redrafts its language and half-defends the result. The later Irish assemblies stalled short of the ballot altogether: Biodiversity Loss delivered 159 recommendations that sit in committee follow-up; Dublin’s directly elected mayor remains unimplemented.
France quantified the failure mode. The Convention Citoyenne pour le Climat — 150 citizens, 149 proposals, President Macron pledging to pass 146 “without filter” — got its independent accounting in April 2025: 20 percent of recommendations fully implemented, 51 percent partially or through alternative measures, 33 recommendations — 22 percent — abandoned, rejected, or never acted on. Germany’s first Bundestag-mandated citizens’ assembly, 160 members on food policy, adopted nine recommendations in January 2024 — free school lunches at the top — handed over its report in February, received one plenary debate in March, and went to committee, where it sat while the governing coalition fell. Different countries, one shape: recommend and shelve. The room did its work. Nothing obligated anyone to answer.
Set the dead ends beside the live chains and the difference resolves into five antidotes: a statutory response clause — Ostbelgien’s ministers must answer and reconvene within a year; council referral to a binding ballot — Question 2H; a named elected sponsor with a hearing date — Connecticut’s Comptroller; money attached — participatory budgeting, next section; and the warrant article, where deliberation and decision happen in the same room on the same night. Every documented success has one of these; every documented dead end lacks all five. If you carry one design rule out of this part, carry that one — before anyone deliberates anything, get the response mechanism in writing.
The warrant article is the oldest antidote and the cheapest, so here are its mechanics. In Massachusetts, 10 registered voters by petition put an article on the annual town meeting warrant; 100 — or 10 percent of registered voters, whichever is less — insert one into a special town meeting; 200, or 20 percent, compel the select board to call a special meeting within 45 days. In Connecticut, the selectmen must warn a special town meeting on the application of 20 qualified voters, held within 21 days, on a petition form approved by the town clerk; case law lets the board refuse only if the object is unlawful, improper, or frivolous. Twenty neighbors and a clerk-approved form are the entire activation energy.
And the floor’s product is not advisory. In November 2020, a special town meeting in Brookline, Massachusetts passed the nation’s first tobacco-free-generation bylaw — no sales of tobacco or nicotine products, ever, to anyone born after January 1, 2000. Retailers sued; the Superior Court dismissed in 2022; on March 8, 2024 the Massachusetts Supreme Judicial Court upheld the bylaw, and other municipalities began copying it within weeks. A floor vote of townspeople made national policy and survived the state’s highest court — the same day Ireland’s diluted amendments died. The binding power is real, and it exists only when bodies are in the room; West Tisbury’s empty October is the converse case.
One more antidote, with its own failure inside it. Oregon’s Citizens’ Initiative Review — piloted 2010, statutory by 2011 — convenes 18 to 24 randomly selected, demographically matched voters for 3 to 5 days to interrogate one ballot measure, advocates and opponents testifying under questioning, then writes a one-page citizens’ statement printed in the official Voters’ Pamphlet mailed to every voter in the state; voters who read it measurably knew more about the measure and were more confident in what they knew. But the statute made the commission responsible for its own costs, the legislature has never appropriated a dollar, and no statewide review ran in 2018 for lack of funding. Institutionalization without appropriation is a slow-motion sunset. The binding clause has to bind the money too.
The fourth antidote puts dollars directly under the deliberation, and it is the city reader’s nearest lever. Participatory budgeting in the United States stands at no fewer than 64 cities and counties, 258 council districts or wards, and over 260 schools, allocating more than $360 million cumulatively — an undated floor of a count, since nobody keeps a current census — descended from Chicago’s 49th Ward, where Alderman Joe Moore put his roughly $1.3 million in “menu money” to a public vote in 2009. New York City Council’s program, launched 2011, now has a majority of the 51 districts putting at least $1,000,000 of capital money apiece to a resident vote — more than $30 million a year, over 93,000 residents voting on the FY2026 projects. The citywide “People’s Money” adds $4 million decided by every New Yorker aged 11 and up, regardless of immigration or incarceration status. Cambridge’s tenth cycle drew a record 10,522 voters aged 12 and up to decide $1,000,000 — still under 9 percent of residents; Boston put $2,000,000 citywide through an office a 2021 charter measure created; and Paris, the ceiling case, has run €768 million of participatory budget since 2014 — 1,345 funded projects, €263 million reserved for low-income districts.
The failure cases are better teachers. Vallejo, California ran the first citywide American program — adopted 2012 after municipal bankruptcy, seeded with 30 percent of a sales-tax increment, $8.3 million allocated to 47 projects across five cycles — then decayed: by the 2020 cycle, a month of voting produced 1,909 ballots against a 7,000-vote goal in a city of about 125,000. Participatory budgeting without continuous organizing decays into a small-sample poll with a budget line. Cleveland’s Issue 38, on the ballot November 7, 2023, would have been the country’s most ambitious program — ramping to 2 percent of the general fund — but it was designed as a referendum against the council: it needed $350,000 of first-year administration, drew a council-aligned PAC that raised more than $95,000 for the “no” side and a council president warning of a “devastating impact on the city,” and lost by roughly two points. Nearly every surviving American program is council-sponsored; deliberative money lives when the body that controls the budget owns the process, and bypass triggers the institutional immune response. Read soberly, the evidence says participants’ own trust and civic knowledge rise while community-wide effects remain weakly evidenced — which argues for honesty, not retreat. A participatory-budgeting vote is a recruiting engine attached to a binding decision. Sell it as that, not as a cure for civic trust.
Everything so far happens in rooms. The funnel above the rooms went digital a decade ago, and the record is good enough to copy — provided you copy the whole pattern, including the part where humans decide.
Taiwan wrote the reference case. The vTaiwan process — launched 2015 by the g0v civic-tech community with then–digital minister Audrey Tang — ran on Polis, open-source software that clusters participants by their votes on short statements and surfaces the statements that bridge clusters; it has no reply button, so there is structurally nothing to flame. In the defining case, 31,115 votes were cast on candidate UberX regulations; “the dispatch system must have a two-way rating system” reached 95 percent consensus; on May 23, 2016 the government agreed to ratify the consensus points, and the transport ministry amended its regulations. The epilogue teaches as much: vTaiwan declined after 2018 — COVID killed the meetups, the volunteer load was heavy, and no statute ever required ministries to respond — before reviving as a volunteer laboratory pairing Polis with LLM-assisted summarization. A consultation platform without a mandated response clause decays when the founding personalities move on. The funnel obeys the floor’s law: power, or attrition.
Barcelona built the other reference. Decidim — “we decide,” in Catalan — is open-source participation infrastructure born for the city’s 2016 action plan, which drew between 9,000 and 10,000 citizen proposals through roughly 400 face-to-face meetings, about 70 percent of citizen proposals accepted into the final plan. Adoption counts conflict by methodology — call it dozens of serious municipal deployments, hundreds of registered instances — but the deployment that matters to the renter is participate.nyc.gov, where The People’s Money runs on Decidim now.
The newest American case put AI in the loop and measured it. Bowling Green, Kentucky — city population about 75,000 — ran “What Could BG Be?” on Polis from February 14 to March 17, 2025, feeding its 2050 comprehensive plan: nearly 8,000 residents, over 4,000 distinct proposals, more than 1,000,000 votes, with Google Jigsaw’s Gemini-based Sensemaker categorizing and summarizing the mass — the first major US municipal use of LLM sensemaking on a civic conversation. 83 percent of participants believed the process would produce a better plan; 96 percent of local leaders felt better equipped to represent their community. Participation near a tenth of the city is extraordinary. The caveat is this part’s thesis: the output feeds a plan, not a binding vote. The warrant-article stage still has to happen.
Know the frontier experiment at its true size too. In Science, October 17, 2024, Google DeepMind’s “Habermas Machine” — two fine-tuned language models acting as caucus mediator — ran 5,734 UK participants through deliberations on Brexit, immigration, wages, climate, and childcare; participants preferred the AI’s common-ground statements over a human mediator’s 56 percent of the time, and groups measured less divided afterward. The lead researcher’s caveat travels with the result: the model “doesn’t have the mediation-relevant capacities of fact-checking, staying on topic, or moderating the discourse.” DeepMind has no plans to release it, and the Knight First Amendment Institute’s warning stands — optimizing for agreement is not the same as improving deliberation.
So the doctrine, from every working case on record: digital tools collect and structure breadth — thousands of voices, honestly mapped; a small accountable human body converts the structured input into a decision; no documented case lets the platform decide. The machine widens the funnel and drafts the summary. Named humans, in a room, with a response clause over their heads, decide. The architecture is copyable with free software — Polis and Decidim are both open-source; the real cost is staff time and outreach.
Which leaves the room — and the rectangle of glass in every pocket in it.
Be honest about the evidence, because the honest version is still strong enough to act on. The famous laboratory findings — that a phone merely sitting on the table degrades conversation and thought — are fragile: a pre-registered 2021 replication of the foundational study failed, a 2022 replication of the working-memory result found no difference by phone location, and a 2024 meta-analysis puts the pooled mere-presence effect at small to null. What is not contested is that phone use displaces attention and conversation. The strong case for device-free formats comes from institutions, not dyad labs: Norway’s staggered middle-school smartphone bans, run against national administrative data — a working paper, and the strongest single quantitative result in this literature — found girls’ grade averages rose, their visits to psychological specialists fell by almost 60 percent, bullying fell for both sexes, and girls became 4 to 7 percentage points likelier to enter an academic high-school track. Venues built the practice at scale — Yondr’s lockable pouches, the phone on your person but unusable until tapped on an unlocking base, cover more than 2.5 million students at $25 to $30 a head per year — and since 2017, Philadelphia’s criminal courthouse has pouched the phones of jurors and visitors rather than confiscate them.
Now the gap, at full width: no documented town meeting or citizens’ assembly anywhere runs a formal phone-pouch rule, and no controlled study of device-free civic deliberation exists. The professional assemblies get the result by furniture instead — printed briefing books, small tables, paper worksheets, timed spoken turns; the format crowds out the phone without confiscating it. Stanford’s “America in One Room” put 523 delegates — the most representative sample ever assembled for a deliberation — through four days of printed briefings and moderated small groups in September 2019, and measured opinion movement toward the center persisting at least a year. So run the proposal as exactly what it is: a phones-in-the-box rule for your town meeting is an evidence-adjacent design, backed by the school natural experiments and the courthouse practice, untested in the civic setting itself — and anyone who claims otherwise is selling something. A labeled box at the door, printed one-pagers on the chairs, the Philadelphia and Norway citations ready for the skeptic: total cost, a shoebox — or $25 to $30 a head for the hard version.
The room’s other infrastructure is trust between the people in it, and that has a randomized trial. Braver Angels’ Red/Blue workshops — founded December 2016, first workshop South Lebanon, Ohio — were tested on undergraduates at four universities, and the 2025 published result is plain: “The workshops substantially reduced polarization according to explicit and implicit measures,” and “These effects are consistent across partisan groups, though some dissipate over time.” The comparison megastudy — 32,059 participants, 25 interventions — found 23 of 25 reduced partisan animosity; durable and behavioral change remain unproven. Depolarization is cheap to start and unproven to last; buy it for what it costs. The dialogue-to-action version has a record in city policy: Portsmouth Listens has run study circles of 8 to 12 residents into New Hampshire city decisions since 1998, including the two-year program that built Portsmouth’s 2003–05 Master Plan — “it is fair to say that people consider it an important and driving policy for the city,” its co-chair Jim Noucas concluded. The adjacent formats — World Café, National Issues Forums — run on practitioner reports, not controlled outcomes; use them to build relationships and surface agendas, not to decide.
Five formats survive the record. Each gets a name, a size, a clock, a price, and the question it decides best. Pin them up where your select board — or your community board — can see them.
Pattern 1 — THE WARRANT-ARTICLE MEETING. Size: quorum to about 500; West Tisbury ran past 300 when the warrant had power. Duration: one evening to several, 2 to 5 hours each. Facilitation: an elected town moderator, Robert’s Rules or local custom. Cost: near-zero marginal — hall, clerk time, legal notices; the institution already exists in upward of a thousand New England towns. Decides best: binding bylaws, budgets, and land acquisitions in towns up to roughly 20,000. Evidence: Bryan’s 20.5 percent and the size law; Brookline’s court-proof bylaw. Failure mode: quorum. Revival levers: power on the warrant, direct text-and-email contact, food, childcare, officers by ballot.
Pattern 2 — THE CIVIC LOTTERY ASSEMBLY. Size: 20 to 100 — Fort Collins 20, Deschutes County 30, Petaluma 36, Akron 65, Connecticut and Paris and Ireland 100. Duration: 4 to 6 full days across one to three months; Petaluma’s roughly 90 hours is the high end. Facilitation: professional — Healthy Democracy, DemocracyNext, or equivalent — plus an oversight committee and a public, livestreamed lottery. Cost: $450,000 budgeted at Petaluma; stipends from $15 an hour to $1,000–1,200 flat; €1.5 million at national scale. Decides best: one thorny, values-laden question on which the elected body is deadlocked — land use, facility siting, tax structure — and only with a pre-committed response mechanism. Evidence: Fort Collins’s 67.5 percent ratification; Ireland’s 66.4; Ostbelgien’s one-year loop. Failure mode: recommend-and-shelve.
Pattern 3 — THE CITIZENS’ INITIATIVE REVIEW. Size: 18 to 24 demographically matched panelists. Duration: 3 to 5 consecutive days, finished before ballots print. Facilitation: professional moderation; advocates and opponents testify under questioning. Cost: privately funded in Oregon since 2011 — per-review cost unpublished, and the funding gap killed the 2018 statewide round. Decides best: nothing itself — it arms a binding vote, giving voters a trusted one-page brief on a contested measure instead of two campaigns. Evidence: ORS 250.137; measured knowledge gains; the statement mailed in the official Voters’ Pamphlet. Failure mode: no appropriation.
Pattern 4 — THE MASS DIGITAL CONVERSATION. Size: 1,000 to 10,000-plus — Bowling Green’s nearly 8,000; vTaiwan’s 31,115 votes; Barcelona’s 9,000–10,000 proposals. Duration: 2 to 6 weeks open, then synthesis. Facilitation: a small civic team seeding statements, plus a real communications push; LLM summarization compresses the output; no reply buttons, no flame war. Cost: the software is free and open-source — Polis, Decidim — so the true bill is staff time and outreach. Decides best: agenda-setting, option-surfacing, and consensus-mapping before a binding stage — never as the binding stage. Evidence: Taiwan’s ratified consensus; Barcelona’s 70 percent. Failure mode: no mandated response.
Pattern 5 — THE STUDY CIRCLE / RED-BLUE WORKSHOP. Size: 8 to 15 per circle, many circles in parallel. Duration: one 2.5-to-3-hour workshop, or 3 to 5 fortnightly 2-hour rounds. Facilitation: trained volunteers — Braver Angels, Everyday Democracy, NH Listens curricula. Cost: room, coffee, photocopies; facilitator training is the only real line item. Decides best: nothing, by design — it rebuilds the capacity to deliberate where trust is broken, and pre-frames a divisive question before it reaches a warrant or a ballot. Evidence: the 2025 randomized trial; Portsmouth’s Master Plan; 23 of 25 in the megastudy — the cheap part is real, the durable part not yet proven. Failure mode: mistaking it for a decision.
The sequencing rule compresses every working case into two sentences: Pattern 5 builds the room; Pattern 4 maps the options; Pattern 2 or 3 deliberates the hard call; Pattern 1, or a referred ballot, makes it binding. Towns that skip the last step produce reports; towns that skip the first steps produce 11 p.m. shouting matches.
Fort Collins ran the sequence without naming it — surveys, guides, lottery, council referral, ballot. Connecticut is about to run Pattern 2 at state scale with the named-sponsor antidote bolted on; watch it with the checklist in hand, then bring its central question home to your own selectboard or community board: when recommendations arrive here, who is obligated to answer? The formats exist, priced and proven. The response clause is the part only your town can write. And the first law holds at every scale, so end where the moderator ends: people will come to meetings that have power. Put some power on the next warrant.
If you live in a town:
If you rent in a city:
Either reader:
Part VI
Fire, CERT, mutual aid — the town’s standing capacity, and the 24-month path from zero.
On the first day of April 2025, the engines of the Torringford Volunteer Fire Department rolled out of their bays on the east side of Torrington, Connecticut, for the last time — not to a fire, but to a transfer of title. After 68 years protecting the city’s growing east side, the company was, in the words of the local coverage, “forced to shut down due to a lack of firefighters,” and every piece of apparatus passed to the city department. The company had spent thousands of dollars on recruitment, and recruitment lost. Tim O’Donnell, president of Torrington’s career firefighters’ union, told WFSB in March 2025 what the closure would mean, in the only unit that matters: “The closest staffed fire engine right now is eight to 10 minutes away, minimum. And that’s very problematic for anyone who is experiencing a structure fire, life threatening medical emergency, serious car accident, anything of that nature.” Torringford was not even Torrington’s first loss — the Burrville volunteer company closed in 2017, leaving Drakeville the city’s last volunteer holdout. This is Litchfield County, the home ground of the book this volume accompanies — the book whose seven-layer table put the volunteer fire company and CERT in the militia layer, the organized-response stratum of 1776 carried forward. The layer is failing on its own home ground.
Behind the local closure runs a forty-year national curve. In 1983, the first year the National Fire Protection Association ran its survey, the United States counted 884,600 volunteer firefighters. By 2000: 777,350. By 2017: 682,600. By 2023: 635,100 — the lowest count NFPA has ever recorded, a fact the National Volunteer Fire Council’s February 2026 fact sheet states without decoration: “The number of volunteer firefighters in the U.S. reached a low in 2023.” Hold the second number beside the first: those 635,100 are still 62% of the estimated 1,018,100 firefighters in this country. Of 29,452 American fire departments, 18,873 are all-volunteer and another 5,335 mostly volunteer — 82% of the institutions — protecting roughly 30% of the population, concentrated in exactly the rural and suburban towns this guide is addressed to. Per capita the fall is steeper than the headline: 8.05 volunteers per 1,000 Americans at the 1987 peak, 5.66 per 1,000 by 2020 — a 30% per-capita decline across four decades in which emergency call volume more than tripled, most of the growth medical. The remaining ranks are aging — in communities under 2,500 people, 53% of firefighters are over 40 and 32% are over 50 — and the work has not gotten safer to give away: of the 72 American firefighters who died in the line of duty in 2024, 32 were volunteers.
Recruitment is the first failure mode. The second is institutional, and Connecticut supplied the national case study ten months after Torringford. Yantic Fire Engine Company No. 1 — serving Norwich since 1847, 58 members, 772-plus calls the previous year — was shut down by its own city at 10:02 a.m. on a Friday in February 2026, after declining to sign a nine-page operating contract presented after business hours with less than an hour allowed for attorney review; the city froze the company’s budget, cut its credit-card access, and removed it from dispatch, and four of Norwich’s five volunteer companies are now suing their own city. The pattern repeats at every scale: Wallace, North Carolina temporarily disbanded its volunteer department after a wave of retirements left too few hands to operate; in Surry County, North Carolina, commissioners declined in a closed-door meeting to renew the Franklin Community Volunteer Fire Department’s contract, setting a June 30, 2026 closure that costs the county roughly 60 firefighters across two communities. A volunteer company can die of an empty roster or of a broken relationship with its government. Both deaths end in the same place — eight to ten minutes away, minimum.
This part is about standing capacity: the people, gear, paper, and drills a town holds ready between emergencies. For the reader in a town, the assets are the fire company and a CERT team that probably does not exist yet. For the reader renting in a city — where the engine company is career and the tax bill already carries it — the gap is different: the deployable-civilian tier, the building-scale roster, the mutual-aid network that outlived 2020. Both readers get the same set-piece at the end: twenty-four months, zero to drilled and insured capacity, priced at every stage.
Start with the national number, dated honestly because no newer one exists: donated volunteer-firefighter time saves American localities an estimated $46.9 billion a year — a figure derived by NFPA’s Fire Protection Research Foundation in October 2017 and still the one carried in the NVFC’s February 2026 fact sheet. The arithmetic underneath it has only steepened: training and equipping a single firefighter “can exceed $35,000,” with a self-contained breathing apparatus alone running about $10,000 at January 2026 prices, a pumper $450,000 to $675,000, an aerial $1.5 million to $1.8 million, an ambulance $275,000 to $350,000.
National estimates persuade nobody at a town meeting. The town of Thompson, Connecticut — population about 9,450, five all-volunteer companies — produced the document that does: it paid VFIS, the fire-service insurer and consultancy, for an independent assessment from June 2017 to February 2018, then published the numbers on June 20, 2019. Total fire and EMS protection: $509,578 a year — $52.00 per resident, $136.62 per household, with $68.41 per resident later proposed as the honest correction. The lines a selectboard can actually read: liability and vehicle insurance plus workers’ compensation for all five companies, $87,698; hydrant rental, $80,318; 911 dispatch, $22,937; paramedic intercept, $19,000; the volunteer tax abatement, $40,000; physicals, $750. And one interior firefighter, outfitted and trained, itemized to the cent: turnout coat and pants $1,923; boots $387; helmet $270 and shield $50; gloves $79; hood $60; SCBA pack $6,856; mask $308; 45-minute bottle $1,244; the Firefighter I course $950 — 170 hours, in Connecticut; flashlight $80.63; Halligan bar $215.99; ax $135; portable radio $780. Total: $13,338.62, at 2019 prices — against which the NVFC’s 2026 answer to the same question is “can exceed $35,000.”
Read Thompson’s fine print and the fragility shows through the bargain. Each company received a $48,000 annual town grant — $12,000 in 1990 — and fundraises the rest, and capital is the company’s problem, not a town line item: a $975,000 pumper-tanker and a $950,000 station sit on the books of what is, legally, a supper-fundraiser nonprofit. The VFIS assessors flagged it in consultant’s deadpan: “…it appears the budget is slightly less than appropriate…” The replacement clocks tick regardless — turnout gear on a 10-year cycle at roughly $3,000 a set, SCBA on a 15-year cycle at $8,000 to $10,000 a set, apparatus on 20-to-30-year cycles. A town that has never run Thompson’s exercise does not know what its fire protection costs; it knows only what it pays — and the difference between those numbers is the subsidy its neighbors donate in hours, gear, and 2:00 a.m. tone-outs. The city reader’s version of this section is one sentence: you pay full freight in taxes, the career department is a bargain you already bought, and your unpriced gap is the one the next sections cover.
Honesty first, because the recruitment-incentive literature runs on advocacy: not one lever on the list that follows — pensions, tax breaks, marketing, housing — has a published randomized or rigorous quasi-experimental evaluation behind it. What exists is mechanism plus named department counts, and the obligation to say which is which. The clearest mechanism finding is the NVFC’s own 2015 formative research: most volunteers join through personal connection — “connection, culture, and proximity” — not advertisements. The recruiting tool is a member who asks a neighbor, with an institution behind the ask; Part X measured the same mechanism at the ballot box and found the personal ask beating broadcast by a factor of about 28. The levers below do not replace the ask. They are what makes the ask answerable.
The pension lever: Length of Service Award Programs — LOSAP — pay a volunteer a small defined benefit or contribution after vesting, typically five to ten years of point-qualified service. Roughly 600 New York fire districts and municipalities sponsor one; recent New York legislation raised the defined-contribution annual maximum from $700 to $1,200 per volunteer; federal tax law caps annual awards near $6,000, and a “No Tax on LOSAP Act” has been introduced in Congress. Adoption is documented; the retention effect is advocacy-grade, and this guide will not pretend otherwise. The mechanism is the vesting itself: a volunteer five years into a ten-year vest has a reason to stay that no t-shirt supplies.
The tax lever — and here Connecticut wrote the statute a Goshen or Cornwall town meeting would invoke: CGS § 12-81w lets a municipality abate up to $2,000 in property taxes for volunteer firefighters, EMTs, fire police, civil-preparedness staff, and retired volunteers with 25-plus years. Newtown, Wilton, Monroe, Old Saybrook, Simsbury, and Canton have adopted it by ordinance, and Canton’s version carries the detail that matters: the $2,000 is conditional on attending 60% of drills. The respect is priced, and so is the showing up. New York’s 2022 uniform law, RPTL § 466-a, lets any local government grant volunteers a 10% exemption on the assessed value of a primary residence, typically after two to five years of service. Pennsylvania supplies the category’s failure case, which belongs in this part as much as any success: Act 172 of 2016 authorized municipalities, school districts, and counties to credit volunteers’ earned-income and property taxes — and eight years on, the State Fire Commissioner’s October 2024 report counted 60 active programs among 2,500-plus municipalities, crediting 1,093 responders across three reporting years, with the typical credit running $200 to $500. Set $300 against a volunteer’s 200-plus annual hours and the incentive prices the work under $1.50 an hour. An incentive statute without local adoption moves nothing. The one structural novelty in the Pennsylvania data points somewhere genuinely new: the Exeter Township School District grants volunteers a $599-to-$600 school-tax rebate — a school district, not a town, paying for fire protection.
The federal lever survived the 2025 reorganization of FEMA: SAFER — Staffing for Adequate Fire and Emergency Response — announced $360 million in FY2024 awards on May 20, 2025, and the FY2025 round makes $324 million available through a window that closes June 22, 2026, days after this edition ships, with the next round expected on similar spring timing. The detail built for the volunteer reader: Recruitment & Retention grants carry no local cost share at all — an all-volunteer company can apply with zero town dollars. What SAFER money buys at statewide scale is visible in New York, where FASNY runs a SAFER-funded apparatus — the “Fire In You” media campaign, the RecruitNY open-house weekend each April across hundreds of departments, a tuition program paying up to $1,500 a semester since 2011. By the program’s own count it has added more than 20,000 to the state’s volunteer fire service over the years; over roughly the same two decades, New York’s volunteer ranks still fell from about 120,000 to about 80,000. Both numbers can be true. Marketing slowed the bleed; it did not reverse it.
The lever with the strongest mechanical claim solves a different problem from the roster count: coverage. The College Park Volunteer Fire Department, on the University of Maryland campus, houses 24 full-time student volunteers a semester in the John L. Bryan Fire Service Dormitory — the “Sackroom,” established 1956, with student volunteers since the 1940s: twelve furnished two-person rooms wired to the alerting system, in a station “mostly staffed by live-in volunteers.” The model repeats at Hyattsville, Berwyn Heights, and Rockville in Maryland, and at Aetna Hose, Hook & Ladder in Newark, Delaware, where University of Delaware students get free housing in exchange for roughly three 10 p.m.-to-6 a.m. on-call shifts a week. Cost to the department: dorm buildout and utilities. Return: a station staffed at 3:00 a.m. — the one thing no tax abatement has ever produced.
Three smaller dials, stated at their true size. The junior pipeline: the NVFC’s national program dates to 2007, New York and Pennsylvania admit juniors at 16 — some Pennsylvania departments at 14, restricted — but no national study measures how many juniors convert to interior firefighters, so the pipeline is offered here as practice, not proof. Small money: the NVFC and State Farm’s Good Neighbor program has moved $2.5 million to 250 departments in $10,000 awards since 2024 — 150 departments funded in 2026 alone. And the dial most towns never notice: state training minimums for volunteers range from a few dozen hours of exterior-only certification to several hundred hours where volunteers meet the identical standards as career firefighters — Connecticut’s Firefighter I runs 170 hours — and every added hour prices volunteering upward in the only currency a volunteer actually spends.
Whether any of this can rebuild a dead company is not theoretical, because one town just did it, with counts. Utica Township, Clark County, Indiana: after the district shed a troubled contractor carrying nearly $1 million in apparatus debt, residents stood up Utica Fire & Rescue as a new all-volunteer department in the summer of 2024. By December 2025 the roster held 37 firefighters with a waiting list of recruits; 20 members were sitting Firefighter I and II exams; close to 20 advanced EMTs and paramedics were building a basic-life-support medical program; two of three debt-laden vehicles and an unused firehouse had been sold; new bunker gear, three sets of rescue tools, and a flood boat had been bought; a mutual-aid agreement was in negotiation with a Kentucky department across the East End Bridge; and the board had committed to reserving the full payoff of a $900,000 bond during 2026. The chief, Joe Jarles, had been the board’s president — he stepped down in rank to step up in role — and he named the rebuild’s hidden subsidy: “I can’t even name all the departments that have helped us.” With January 1, 2026 set as the fully-operational date, board president Kelly Khuri compressed this entire part into five words: “We’ll be waiting for the call.” Eighteen months, zero to operational, waiting list attached. What saved Utica was none of the levers alone — a clean break from a failed institution, a leader willing to descend, visible quick wins, borrowed training capacity. The levers funded it. The neighbors did it.
CERT — the Community Emergency Response Team — was invented by the Los Angeles City Fire Department in 1985 and went national under FEMA in 1993; it now stands at more than 3,200 local programs across all 50 states, tribal nations, and territories, with more than 600,000 people trained since the program went national. It survived the 2025 shrinking of its federal parent, and the evidence is named and dated: FEMA’s Emergency Management Institute kept CERT Program Manager and Train-the-Trainer courses on the calendar into 2026, a National CERT Conference ran in 2024, and the FY2025 grant cycles proceeded on schedule. The deeper reason is structural. A CERT program is sponsored, trained, and insured locally — FEMA supplies curriculum, a registry, and instructor courses, not operating money — so federal turbulence thins the support services without switching off a town’s team. The exposure runs the other way: teams built on federal pass-through dollars are the fragile ones.
The product is a twenty-hour neighbor. CERT Basic runs approximately 20 classroom hours — commonly seven to ten weekly evenings, or two to three weekend days — through disaster preparedness, fire safety and utility control, two units of disaster medical operations, light search and rescue, team organization on incident-command lines, disaster psychology, terrorism awareness, and a final disaster simulation; a hybrid format pairs 12 online hours with 16 in person, and Teen CERT mirrors the adult curriculum. The price rounds toward zero: FEMA’s own startup guidance says flatly that “successful CERT Programs do not necessarily require significant budgets” given a donated room, loaned equipment, and volunteer instructors — the required steps are local-government recognition through a sponsoring agency, usually the fire department or county emergency management, plus registry listing, a training plan, and waiver forms. A per-member safety kit — hardhat, vest, goggles, gloves, ID, basic medical — runs about $35, and a realistic startup for a 20-person class with donated space and county instructors lands near $700 to $1,500. That figure is an assembled estimate; no official total exists, and this guide flags its estimates.
What does the twenty-hour neighbor do when the weather arrives? North Carolina answered after Hurricane Helene: in the first three weeks of recovery, 80 CERT volunteers logged more than 3,500 hours — staffing the State Emergency Operations Center’s front desk 12 hours a day and working United Way 211 calls, including missing-person reports and wellness checks — donated time the state valued at over $120,000, at $33.80 an hour. Behind the surge stood standing infrastructure: 72 teams, 2,557 active volunteers, five university campus programs, more than 100 high schools teaching the curriculum — 807 students completed it in 2024. Dawn Truskowski, a 20-year CERT volunteer from Cary, gave the state’s blog the volunteer’s whole theory of the case: “I just think everyone deserves a little bit of kindness. I want to give back and do my part.” Notice what the work was — phones, front desk, logistics. Not rubble. That is the honest profile of modern CERT deployment, and a town that recruits on rubble fantasies will lose the volunteers it gets.
Los Angeles, January 2025, taught the harder lesson. LAFD’s CERT call-out team worked the Palisades, Hurst, and Sunset fires — support operations at incident command, meals trucked to crews, donation inventory, members working about seven consecutive days, daily resupply of shelters near the Eaton fire. But only the call-out team deployed: members with additional training, defensive-driving certification, and background checks. Basic-course graduates were not deployed at all — in the most destructive urban fire event in California history. The two-tier rule falls straight out of those facts: the 20-hour course makes a prepared neighbor; a deployable team requires a second tier of vetting and drill. A town that stops at the certificate owns a roster, not a capacity. One more datum, for timing: NPR documented a wave of ordinary citizens seeking CERT certification in the months after Helene — the recruitment window opens when the sky closes. For the city reader this section is the whole game: New York, Los Angeles, Boston, and Chicago all run basic classes, and the building you rent in is the natural team boundary — the floor list is the roster, and the course costs twenty hours and nothing else.
More than 800 local mutual-aid networks formed in the United States during the COVID emergency, and no census records how many survived — an absence that is itself the finding: nobody is even counting the die-off. The case record tells the story instead. Bed-Stuy Strong, founded in Brooklyn in March 2020, delivered a week’s groceries to 28,000 people in Central Brooklyn and moved $1.2 million in crowdsourced donations in fifteen months — then closed its intake line on June 2, 2021, calling it “a gratitude-filled end to our COVID-19 food provision model.” The network survives as a membership community — street cleanups, reading groups, a community fund — but the emergency operation is gone. Allston-Brighton Mutual Aid in Boston wrote the franker obituary, quoted in The Christian Science Monitor in January 2023: “The pool of volunteers that we had in 2020 has since dwindled, and after months of struggling to keep ABMA afloat, we have decided to shut our operations down.” And Mutual Aid Medford and Somerville shows what survival looks like: in 2024 it reimbursed its volunteer grocery deliverers $89,993 and supported 60 families weekly — having narrowed from everything-aid, pet care to translation, down to three durable programs: a gardening collective, grocery delivery, neighbor-pairing.
The survivors share four design features, and the peer-reviewed literature confirms the decay they engineered against — post-disaster “solidarity behaviors tend to decline over time, even when needs remain high,” as Bender and colleagues put it in the American Journal of Community Psychology in 2024. Survivors narrowed to one or two concrete recurring services; converted donor surges into recurring small-dollar funding; kept a membership structure rather than a volunteer-dispatch model; and either pre-dated the emergency or outlived its identity. Membership, not dispatch — hold that clause, because the institutions that have run mutual aid for generations are membership all the way down.
The Amish do not carry commercial insurance. After a fire or flood, “the Amish community rallies quickly to clean up the debris and construct a new building,” with “hundreds of people converg[ing] to erect a new structure in a single day” — barn raisings led by two experienced Amish “engineers” who plan the structure and stage the materials in advance, as the Young Center for Anabaptist and Pietist Studies documents. For hospital bills, “many Amish communities have an informal aid plan to help members with large hospital bills or other exceptional needs,” funded by special offerings “collected by the deacon in door-to-door visits with each family in the district,” plus benefit auctions; the formalized version layers community-wide assessments over congregational alms. The economic design is risk pooling without premiums — labor and assessments on demand, enforced by membership rather than contract — and its precondition is the one thing a modern town lacks: a closed membership with high exit costs and weekly face-to-face assembly. The Latter-day Saints run the most fully institutionalized lay disaster machine in the country, mechanics published: every ward and stake maintains a written emergency plan “prepared under the direction of the bishop or stake president”; ministering assignments double as a welfare-check roster; the church pre-positions relief supplies at bishops’ storehouses sited for disaster geography — Orlando, Slidell, Tucker, Raleigh. After Helene, by the church’s own accounting, 6,172 Helping Hands volunteers from 400 congregations in six states worked more than 3,000 homes through 11 command centers — 77,000-plus hours of muck-outs, debris clearing, tree removal, and roof tarping, in small teams under designated leaders, with issued tools and yellow vests. A secular town cannot copy the covenant, and the covenant is what does the retaining. It can copy the mechanics: a standing roster with pre-assigned check-on-your-neighbor pairs, a named local commander, pre-positioned tools, and a uniform that makes the institution visible — in a city building, that is a floor-captain list with vests in the super’s closet.
The Cajun Navy shows improvisation institutionalizing itself in real time. Born in volunteer boat rescues after Katrina in 2005, it scaled in the August 2016 Baton Rouge floods on two phone apps — Zello push-to-talk for dispatch, Glympse for location — vetting Facebook rescue requests and routing nearby boats. Then, formalization in three steps. Incorporation: Cajun Navy Relief took 501(c)(3) status effective September 6, 2016, two sister organizations following by 2018. A failed mandate: that same August, a Louisiana state senator floated requiring training, certificates, and permits for citizen rescuers — and was buried in backlash, including threats. And a successful opt-in: in May 2018 the legislature passed HB 388 on a 26-5 Senate vote, directing the state emergency office to build a voluntary credentialing program that gives registered volunteer groups access through emergency perimeters. Jon Bridgers, founder of Cajun Navy 2016, testified for it in words The Advocate preserved: “It would be a good thing to fall under your umbrella as far the liability part.” Mandates provoke revolt; opt-in credentials plus liability cover get volunteers to formalize themselves. File that sentence — it is the design rule for everything in the next section.
One more durable form belongs in this inventory, because standing capacity includes the slow emergencies. The village model — a member-funded nonprofit whose volunteers supply rides, groceries, light maintenance, and a social calendar so older residents stay in their homes — has run since Beacon Hill Village organized in Boston in 2001, and several hundred villages now operate nationally. Boston’s dues run $750 for an individual and $1,100 for a household, with reduced-fee tiers at $110 and $160 — two-tier pricing that names the model’s known weakness, dues-dependence in lower-income neighborhoods — and it is the rare institution in this part built for dense urban blocks first. Time banks still run at modest scale — hOurworld’s cooperative platform carries 30,000-plus members across 200-plus banks, and Kent, Ohio’s has exchanged more than 101,000 hours over 15 years — but a time bank is a ledger for the care layer, not the care layer itself, and it works where an anchor institution — library, senior center, church — hosts the coordinator.
In October 2017, after the Unite the Right rally, the City of Charlottesville and local businesses sued the private armed groups that had deployed on their streets — under Virginia’s constitution, its anti-paramilitary statutes, and common law. Within eight months, nearly all the defendants — the Pennsylvania Light Foot Militia, the New York Light Foot Militia, the III% People’s Militia of Maryland — had signed court-enforceable consent decrees barring their return as armed, coordinated units. What survives every such challenge, by contrast, is the chartered, subordinated volunteer: the CERT team enrolled under a state emergency-management statute and activated by a municipal director, and the volunteer fire company — the oldest continuously legal community-defense institution in America. The rule deserves stating plainly: community capability organized under civil authority is two centuries of settled law; community capability organized beside it is a consent decree waiting to be signed. Every team this part proposes is built under.
Beneath the rule sit three protective layers, each with a catch. The federal floor is the Volunteer Protection Act of 1997: an uncompensated volunteer — the compensation cap is $500 a year — of a nonprofit or governmental entity is not personally liable for harm caused by ordinary negligence within the scope of their responsibilities. Not covered: “willful or criminal misconduct, gross negligence, reckless misconduct, or a conscious, flagrant indifference to the rights or safety of the individual harmed” — nor harm done while driving, nor several categories no statute anywhere forgives. And the clause every selectboard should read twice: “Nothing in this section shall be construed to affect the liability of any nonprofit organization or governmental entity.” The individual is shielded; the organization is not. The state layer adds Good Samaritan statutes in every state and volunteer-firefighter immunity and benefit laws in most — New York’s Volunteer Firefighters’ Benefit Law is a parallel workers’-compensation system in itself. The emergency layer turns on a switch: in Connecticut, CGS § 28-13 immunizes the state, its subdivisions, and “members of the civil preparedness forces” complying with civil-preparedness orders — willful misconduct excepted — with workers’ compensation attaching for members activated in declared emergencies.
The switch is enrollment, and it is the most commonly missed step in the country. FEMA’s CERT Liability Guide walks the three exposure directions — volunteer injured, volunteer injures a third party, sponsor liability — and the standard mitigations: sponsoring-agency recognition, scope-of-activation orders, waivers, training records. California wires the switch directly into state law: CERT members registered as Disaster Service Worker Volunteers carry state workers’-compensation coverage when activated, and a “train and retain” city-of-record arrangement preserves coverage across city lines. The general rule for the town actor: a CERT member activated by the sponsoring agency is covered; a freelancer in a vest is not.
Two more pieces of law belong in the wrapper. First, a drafting trap: Connecticut’s municipal-liability statute — and its analogues elsewhere — leaves a town immune for discretionary judgment but liable for negligence in ministerial duties, the ones written so precisely that no judgment remains. The consequence runs against every instinct of a careful drafter: the more rigidly a CERT charter or mutual-aid MOU commands — “shall, within 24 hours” — the more it converts immune discretion into suable duty. Write “may” and “as resources permit” unless the town intends to be bound. Second, the risks no volunteer immunity reaches, because they belong to the entity: negligent maintenance of what you build, premises liability for program spaces, workers’ compensation, federal civil-rights claims, contract disputes. The mitigation is not heroism but paperwork: enroll volunteers formally, activate them in writing, inspect on a schedule the town can actually keep, and insure.
Insurance is where the wrapper gets priced, and the honest answer is that no public rate card exists. The specialty market’s largest firm is VFIS of York, Pennsylvania — founded 1969, more than 15,000 fire, ambulance, and rescue clients, selling accident-and-sickness, general liability, management liability, auto, and equipment coverage — but its rates are quoted, not posted, so this guide anchors on the two verified numbers. One: Thompson’s combined fire-EMS liability line of $87,698 — about $17,500 per company per year across its five companies. Two: Texas’s 2026 price signal, a new state program reimbursing rural volunteer departments 100% of member-benefit coverage up to $190 per member per year — roughly $5,700 for a 30-member department, with liability and apparatus extra. Most small towns buy none of this retail; they join member-owned intergovernmental risk pools — 215-plus pools representing roughly 46,000 public entities nationally, and in Connecticut, CIRMA, formed 1980, now covering nearly 90% of the state’s public-sector market. Which yields the test that should precede every launch in this part: call the pool underwriter and ask what enrollment, inspection, and waiver paperwork would make the program insurable — because a program the pool will not insure is a program the town cannot afford to run. The checklist, complete: incorporate or attach to a municipal sponsor; write activation procedures so the immunity statutes attach; carry general liability, accident-and-sickness or workers’ compensation, auto, and directors-and-officers coverage; keep waivers and training logs; and if your volunteers improvise on boats, copy Louisiana — liability cover in exchange for registration, never mandate-first.
Everything above compresses into a sequence one organizer can run. It assumes a town with a selectboard, a fire company in some condition between Thompson’s and Torringford’s, and no CERT team. Hours and dollars are stated at every stage; figures marked as estimates are estimates.
Months 0–3 — convene and sponsor. One resident — the knowledge-worker reader can be this person — asks the selectboard to designate a sponsoring agency for a CERT program: the fire department or the town’s emergency-management director; FEMA requires local-government recognition, and the resolution costs nothing and takes one meeting. The same season, put a “state of the fire company” report on a town-meeting agenda — roster, ages, call-coverage gaps, in writing; Thompson’s June 2019 presentation is the template, and it began as a paid assessment. Ask the emergency-management director the enrollment question now: are our volunteers enrolled as civil-preparedness forces, so immunity and workers’ compensation attach? Organizer’s cost: about 20 hours. Town’s cost: $0.
Months 3–6 — train the first class. Host CERT Basic: 20 hours per participant, instructors typically free from the state or county, the town donating the room; equip each graduate with the $35 kit — $700 for a class of 20; register the team on the state registry; adopt FEMA’s waiver and training-log forms on day one. Organizer’s running total: about 60 hours. Dollars to here: under $1,500.
Months 6–12 — wrap it legally and pay for it. Insurance: the town adds CERT to its liability policy and verifies workers’-compensation treatment — California’s Disaster Service Worker registration is the model; in most states this is a rider conversation, $500 to $2,000 a year as an estimate — while the fire company prices accident-and-sickness coverage against the $190-per-member Texas benchmark. Money: file the FEMA SAFER Recruitment & Retention application — zero match for volunteer departments, spring windows — and the NVFC/State Farm $10,000 grant. Respect: adopt the tax incentive — in Connecticut the § 12-81w abatement up to $2,000, the line Thompson budgets at $40,000 a year town-wide; in New York the 10% exemption. One public hearing each. Organizer’s running total: about 80 hours.
Months 12–18 — build tier two. Stand up the call-out subset Los Angeles proved necessary: background checks, a defensive-driving course, ICS-100 and IS-700 free online through FEMA, a monthly two-hour drill. Sign mutual-aid MOUs with the two adjacent towns — Utica’s bridge agreement with a Kentucky department is the current example, Louisiana’s HB 388 the state-level template — drafted in “may,” not “shall.” Add the communications layer: licensed GMRS and ham operators plus a LoRa mesh fallback — the hardware Part IV already priced — at $50 to $100 per member for radios, as an estimate.
Months 18–24 — drill, integrate, perpetuate. Run a full-scale exercise with the fire department and county emergency management — the CERT final simulation satisfies the program’s annual requirements — then staff one real event: the storm shelter, traffic for the fair, wellness checks; the North Carolina profile, not the rubble fantasy. Seed the care layer: a village chapter, or a time bank hosted at the library or senior center — the software is free. Then make it permanent the only way towns make anything permanent: a warrant article making organized response a recurring budget line — Thompson’s all-in benchmark is $52.00 per resident per year for five companies and EMS support; a CERT line alone runs $1,500 to $3,000 a year, as an estimate.
The totals, priced the way Part X prices its rungs: a participating resident invests about 20 hours of training and roughly 24 hours a year of drills; the organizer, 150 to 250 hours across the 24 months; the town, under $2,000 in cash the first year and an estimated $3,000 to $10,000 a year at steady state — all of it before, and dwarfed by, the fire-company capital problem that SAFER and the grant channels in Part IX exist to offset. Now set the sequence against the opening of this part. Torringford spent thousands of dollars on recruitment and closed; Utica spent eighteen months on organization and opened with a waiting list. The difference was never the money. It was a sponsor, a roster, a drill calendar, an insurance rider, and a member who asked a neighbor — capability organized under civil authority, built before it was needed. A town that runs the twenty-four months is waiting for the call. A town that does not is eight to ten minutes away, minimum.
If you live in a town:
If you rent in a city:
Either reader:
Part VII
“…shall always be and remain public property, and shall be subject to the inspection of all persons therein concerned.”
— NMSA 1978 § 73-2-21(D), lineage Laws of 1895, chapter 1
The ditch is dry when the village walks it. Early spring in northern New Mexico, cottonwoods not yet leafed, and the shovels go in — a line of neighbors working down the channel behind the mayordomo, the ditch boss, cutting out a winter’s worth of silt and willow root until the first water comes down carrying the smell of wet earth ahead of itself. The cleaning is the limpia, and it is owed, not volunteered: the commissioners may “assess fatigue work or tasks of all parties owning water rights” — labor pegged to water, the ditch priced in sweat — “together with the power to receive money in lieu of such fatigue or task work at a price to be fixed by them” (NMSA 1978 § 73-2-21(A)(1), (A)(5)). The statute traces to the Laws of 1895, chapter 1. The practice predates the statute by two centuries.
The same statute makes the mayordomo “the executive officer of the ditch,” holding “the superintendence of all work thereon, the distribution of the waters thereof and the collection of fines” (§ 73-2-21(B)(1)). Then the clause this part exists to teach. The mayordomo must “make full written reports of all money received, expended and how expended… semiannually, on the first Monday in June and the last Monday in September” (§ 73-2-21(B)(3)); the commissioners must approve or reject those reports before their own terms expire; and the books “shall always be and remain public property, and shall be subject to the inspection of all persons therein concerned” (§ 73-2-21(D)). Twice a year, in writing, on named Mondays, open to anyone who carries water. A nineteenth-century legislature wrote it down because two centuries of ditch practice had shown what happens when nobody does.
Burlington, Vermont built a citizen-owned fiber network in the 2000s — and lost it. Starting in 2009, under Mayor Bob Kiss’s administration, the city improperly diverted $16.9 million in city funds to keep Burlington Telecom operating, concealed from the city council and the ratepayers; the concealment wrecked the city’s credit, triggered litigation with Citibank, and forced a sale — approved by the council in December 2017 and by Vermont regulators on February 19, 2019, closed by Schurz Communications’ Champlain Broadband for $30.8 million in March 2019. The negotiated covenants softened the loss — rates level for five years, net-neutrality adherence, $250,000 a year for ten years to a tech incubator, a repurchase option, a veto over monopoly-creating sales. That list is the rights of a former owner.
Burlington did not fail at engineering — a buyer paid $30.8 million for the network. It failed at monitoring, Elinor Ostrom’s principle 4. The $16.9 million moved without the written, public, semiannual accounting that an 1895-lineage statute demands of every ditch boss in New Mexico; the oldest clause in this part would have caught it. When a commons dies, the cause is rarely the people and almost never the asset. It is the missing clause — and the missing clause costs millions. Survivors write the accountability clause down.
You Cannot Eat Code maps the seven layers of 1776 onto their 2026 equivalents and lands the last one — the oath — on the Ostrom-compliant commons charter. An oath needs words. This part is the words — where they were already written, and which vessels will hold them.
Ostrom published the eight design principles in Governing the Commons (1990), distilled from commons that had survived for centuries; the work earned the 2009 Nobel Memorial Prize in Economic Sciences. Compressed: (1) clearly defined boundaries — who is in, and what exactly is held; (2) rules congruent with local conditions; (3) collective choice — the governed can change the rules; (4) monitoring, by monitors accountable to the users; (5) graduated sanctions; (6) cheap, fast conflict resolution; (7) recognition of the right to organize; (8) nested enterprises. Do not read them as a theory of virtue. Read them as a checklist of clauses — every one is visible, verbatim, in the governing documents of commons that work.
New Mexico has more than 700 acequias — community irrigation ditches recognized as political subdivisions of the state — and one statute runs nearly the whole checklist. Collective choice: commissioners must, “immediately upon taking office, provide bylaws, rules and regulations not in conflict with the laws of the state for the government of the ditch or acequia, a printed copy of which shall be furnished to each owner of a water right in such ditch” (§ 73-2-21(A)(6)). Boundaries: the water-right roll itself — “Acreage of land with water rights is controlling factor in apportioning water” (1974 N.M. Op. Att’y Gen. 74-23). Conflict resolution: commissioners who deny a water-right transfer owe “a written decision explaining the reasons,” appealable to district court within 30 days and reversible only if they “acted fraudulently, arbitrarily or capriciously” (§ 73-2-21(E)) — upheld in Pena Blanca Partnership v. San Jose Community Ditch (2009). Recognition: New Mexico’s courts have called acequias “involuntary public quasi-corporations” since Candelaria v. Vallejos (1905) — public bodies run by their own members.
None of it starts from a blank page. The New Mexico Acequia Association maintains the Acequia Governance Handbook and has supported the adoption or update of more than 400 sets of acequia bylaws and more than 100 capital improvement plans — a charter press for its member commons, America’s closest thing to a template library for commons governance.
Maine wrote its version in 1995, for traps instead of water: the lobster co-management law (PL 1995, c. 468) created trap limits stepping to 1,200 per license by 2000, trap tags, an apprenticeship program, and seven zones, A through G, each governed by an elected council. The boundary clause is a declaration: only license holders who “declared at the time of obtaining a Class I, Class II or Class III license the zone in which the person fishes a majority of that person’s lobster traps” may vote in that zone, and “a person may declare only one zone” (12 M.R.S. § 6447(7)(B)). The collective-choice clause is a priced referendum: “A lobster management policy council must submit proposed rules to referendum in the zone in which the rules would apply before submitting those proposed rules to the commissioner. A lobster management policy council may submit proposed rules to the commissioner if the proposed rules are approved by 2/3 of those voting in the referendum” (§ 6447(6)). It runs in the wild: Zone A’s trawl-limit rule “was supported by over two-thirds of the Zone A license holders voting in a referendum conducted in June 2018.” Sanctions feed eligibility: “An individual who has been convicted or adjudicated of a lobster violation within the previous 7 years is not eligible for election as a council member” absent a commissioner’s waiver (§ 6447(3)). And the nesting is explicit: councils propose, the state commissioner adopts, a nonvoting legislator sits on each council.
Now the failure mode, because boundary clauses curdle. Maine’s limited-entry zones set exit ratios — licenses retired per new license issued. By 2019 the ratios had hardened into decade-long queues: 55 of 248 people on the waitlists had waited ten years or more, and a 2012 Gulf of Maine Research Institute study had already concluded the barriers were protecting privilege, not stock. The lesson belongs in the charter: write the boundary clause, then write its review date.
The Swiss tested theirs across half a millennium. On February 1, 1483, the villagers of Törbel, in canton Valais, signed articles of association regulating communal use of the alp, the forests, and the waste lands — including the Winterungsregel, the winter rule: no citizen may send more cows to the alp than he can feed over the winter. One sentence does the work of a grazing model, enforced by every man’s own barn. Törbel is the first case study in Governing the Commons; Robert McC. Netting’s Balancing on an Alp (1981) drew on village records running back roughly 350 years. Two cantons east, the Oberallmeindkorporation Schwyz — a public-law corporation first documented in a charter of Emperor Henry V in 1114, older than the Swiss Confederation — counted 19,895 corporation citizens as of March 31, 2026 and manages 24,000 hectares — roughly 59,300 acres — summering about 13,000 animals a year under statutes dated October 19, 2014.
Nine hundred years of solvency, one warning bolted on: OAK membership is hereditary — by descent, never by marriage — and women waited nearly 900 years for full participation. Closed boundaries are durable — at the price of everyone they exclude, for as long as the exclusion runs. A drafting committee has three boundary models on the table — the declaration (Maine), the dollar (Champlain, below), the bloodline (Schwyz) — and only one carries a nine-century apology.
The cheapest full Ostrom stack in American local government is the town forest. New Hampshire’s enabling clause has one moving part: “The legislative body of any city or town may vote to establish by purchase, lease, grant, tax collector’s deed, transfer, bequest or other devise, a city or town forest” (NH RSA 31:110, enacted 1975) — managed by a three-to-five-member forestry committee or the conservation commission, proceeds bankable in a separate forest fund. Mark the mechanism — title plus a vote: absent the vote, it is just town property. Interviews in 220 of New Hampshire’s 244 cities and towns found 4% of the state’s forest town-owned — about 1,700 parcels, 180,439 acres — 69% permanently protected, 54% under a management plan, 71% in professional care.
The set-piece is the Town of Randolph, New Hampshire — year-round population about 310 — owner since 2001 of the 10,000-acre Randolph Community Forest: precisely 9,797.4 acres in Randolph, 1,381.81 in Jefferson, 26.62 in Gorham. Call it 32 acres of commons per resident. Under the town’s forest ordinance the Planning Board supervises — approving the Forest Commission’s annual budget, every timber harvest, and major proposals; the Commission executes; named “activity managers” run specific uses — the Waumbek Methna snowmobile club for snowmobiles, the Randolph Mountain Club for hiking; a second ten-year stewardship plan landed in 2014. Principle 8 — nested enterprises — running in a town smaller than most graduating classes, on acreage that funds its own stewardship.
A vote, an ordinance, a committee, a plan, two named clubs — short documents, all public. The wrapper they sit inside is the next choice.
A commons needs a body the law can see. An American town or neighborhood in 2026 has six working vessels, in rising order of novelty: cooperative corporation, community land trust, nonprofit mutual-benefit corporation, special district, perpetual purpose trust, Exit-to-Community conversion.
Cooperative statutes exist in every state — the agricultural acts date to the Capper-Volstead era of the 1920s. The newer limited cooperative association admits outside investor-members with capped control; the caution travels with it — an LCA can drift from one-member-one-vote, the objection traditional co-op advocates filed. For a commons serving its own members — the tool library, the shared workshop, the community network — the default wrapper is the nonprofit mutual-benefit corporation.
The community land trust is the vessel with the audited track record: the trust owns the land forever; residents own the improvements; a 99-year ground lease plus a limited-appreciation resale formula keeps each home affordable to the next buyer, in perpetuity. The full exhibit is the Champlain Housing Trust — founded 1984 as the Burlington Community Land Trust under then-Mayor Bernie Sanders’ administration, today the largest CLT in the United States. Its FY2025 audited numbers: 706 shared-equity homes — 45 sold that year at an average price of $169,000 against a market exceeding $500,000 — 2,929 affordable apartments, 7 shelters with 220 beds, total assets of $260,039,187, auditors reporting “no material weaknesses.” The governance that holds it: a board of “one-third public members, one-third general members and one-third resident members,” elected annually by a membership anyone may join for $1.00 a year. One dollar buys a vote in a $260,039,187 institution. Now read the city name — Burlington. The town that lost its telecom to a concealed $16.9 million is the town whose land trust opens audited books to dollar-a-year members. The variable is not civic culture. It is the clause set.
Heavier still is the special district. The 2022 Census of Governments counted 39,555 of them — “independent, special purpose governmental units that exist as separate entities with substantial administrative and fiscal independence from general purpose local governments” — among roughly 90,000 American governments: fire, water, library, hospital districts, and New Mexico’s acequias. It is the heaviest vessel — petition, hearing, election — and the only one whose assessment power survives member turnover automatically. Connecticut keeps one such door on its books: under CGS § 7-325, fifteen or more voters may petition specifying a proposed district’s limits; the selectmen must call a meeting of those voters within 30 days; a two-thirds vote establishes a municipal corporation levying its own tax with the same lien power as town taxes — fire, sewer, lighting, sidewalks, recreation; hundreds already operate across the state. Whether to open it belongs to a district’s would-be members and the counsel who advises them — this part inventories vessels; it prescribes none. The big-city analogues are the business improvement district and the special assessment district. The wrapper question — district, nonprofit, or municipal board? — is any commons’s first legal question, and it has named answerers: town counsel for the choice, the Secretary of the State for incorporation, in Connecticut the Office of Policy and Management for districts.
The newest vessels hold purposes rather than members. A perpetual purpose trust is a non-charitable trust with no human beneficiaries — it enforces a stated purpose over an asset, forever. Organically Grown Company became the first US company owned by one in 2018, restructured never to be sold. Patagonia made the pattern famous on September 14, 2022: the Chouinard family moved the company — press-reported at roughly $3 billion — into a two-entity split, the Patagonia Purpose Trust holding all voting stock (2% of equity) and the Holdfast Collective, a 501(c)(4), holding all non-voting stock (98%). “Earth is now our only shareholder.” Strip the billions away and the drafting pattern remains: a small voting trust that enforces purpose, a large economic interest routed to mission. The same split works for a $400,000 building.
The longest-lived large digital commons — Wikipedia, OpenStreetMap, Debian — share one architecture: a thin legal vessel holding the assets, a thick community constitution holding the rules. The corporation owns the servers; the commons owns the rules. The Wikimedia Foundation’s FY2024–25 audit shows $208.6 million in revenue — $189.5 million of it donations — supporting more than 280,000 active volunteers: a nine-figure corporation whose binding rules are written and enforced by unpaid members. OpenStreetMap’s boundary rule lives in its license and contributor terms, its monitoring in changeset review run through its US chapter, which crossed 300 members in June 2025.
Debian wrote the constitution this part can quote — the plain text sits on a public mirror: “Version 1.0 ratified on December 2nd, 1998,” adopted by developer vote — 86 valid votes; the current text, “Version 1.7 ratified on August 14th, 2016,” governs a commons of roughly 1,000 developers. Its first working clause is the theory of volunteer commons in two sentences: “Nothing in this constitution imposes an obligation on anyone to do work for the Project. A person who does not want to do a task which has been delegated or assigned to them does not need to do it. However, they must not actively work against these rules and decisions properly made under them” (§ 2.1.1).
Then the priced supermajority. The developers may “Amend this constitution, provided they agree with a 3:1 majority”; may “Make or override any decision authorised by the powers of the Technical Committee, provided they agree with a 2:1 majority”; and “A Foundation Document requires a 3:1 majority for its supersession” — the Foundation Documents being “the works entitled ‘Debian Social Contract’ and ‘Debian Free Software Guidelines’” (§ 4.1). Ordinary business passes at 1:1. Values cost 3:1. The charter prices its own sanctity. Conflict resolution is last-resort by design — the Technical Committee “does not make a technical decision until efforts to resolve it via consensus have been tried and failed” (§ 6.3(6)). And the treasury is out of reach. Debian holds no money: “Property has to be owned by any of a number of organisations” — historically Software in the Public Interest — and “An organisation holding assets for Debian has no authority regarding Debian’s technical or nontechnical decisions” (§ 9, § 9.2.1). That is the Patagonia split — asset holder severed from rule maker — drafted by volunteers in 1998.
Platform cooperatives carry the same DNA into markets; the honest report is mixed. Stocksy United (founded 2013) pays artist-members royalties of 50% on standard and 75% on extended licenses and distributes 90% of surplus profit as patronage “in proportion to their contribution in total sales”; revenue doubled to $7.9 million by 2016. Up & Go, the New York City booking platform owned by the worker cooperatives on it — majority immigrant- and women-owned cleaning co-ops — leaves workers 95% of the revenue they generate, at $4–5 an hour above area rates. And the accounting no brochure gives: no rigorous failure-rate figure for the platform-co-op wave exists. The directory launched in 2016 to count them stopped being maintained in 2020; the best peer-reviewed evidence, a 2021 multiple-case study, names the dominant failure drivers as “complex governance and above all… lack of funding.” The survivors are countable on two hands, the directory of the dead was never finished, and any precise percentage you encounter is unsupported.
Exit-to-Community — converting an investor-owned platform to ownership by its users, named and theorized at Nathan Schneider’s University of Colorado Boulder lab around 2019–2020 — has a record to read sober. At scale it failed: #BuyTwitter drew 4.9% of Twitter’s shareholder vote on May 22, 2017 — enough to resubmit, nowhere near passage. At small scale it burned out: Ampled, the artist- and worker-owned co-op, shut down at the end of 2023 — its board wrote, “For the last two years, we’ve faced a combination of challenges — primarily from burnout and a lack of resources to hire full-time workers” — and open-sourced its code on the way out. The partial success is Open Collective: in October 2024 an independent, community-governed nonprofit created by its fiscal hosts took over the platform — all fees now flow to it; more than 3,500 collectives had raised over $100 million by 2024 — though the brand and most cash reserves stayed with the company. The users formed the institution; the exit was negotiated, not donated.
If a chartered commons running infrastructure sounds exotic, look at the electric bill. Roughly 2,000 American communities — Los Angeles, Austin, Nashville, Seattle, down to small towns and the Navajo Nation — get electricity from a public power utility — more than 54 million people, about 15% of electric customers. The median public power utility serves about 4,400 people — the size of a Litchfield County town. Municipal enterprise is not a metropolitan exotic. It is a small-town technology.
The audited ceiling is Chattanooga. EPB built America’s first community-wide gigabit fiber network in 2010 and launched community-wide 25-gigabit service in 2022; a November 2025 peer-reviewed study found the fiber network and automated grid “generated $5.3 billion in community benefit since their launch in 2010” — more than six times the original cost — counting 10,420 jobs supported from 2011 to 2024, 417.7 million minutes of outages prevented, $334,437,064 saved on customer bills, and free fiber for more than 28,000 students and family members. Underneath it: an independent city board, audited books every year, a $70 million fiber refresh paid from utility revenues — a commons cross-subsidizing itself, in public.
The stress test is RS Fiber. Sibley County, Minnesota spent a decade on broadband; when a purely municipal bond approach stalled, organizers reconstituted the project as a cooperative — ten cities and 17 townships formed a Joint Powers Agreement and in 2015 issued a $13.7 million general-obligation tax-abatement bond, lending the proceeds to the member-owned cooperative; phase one — fiber to all ten cities plus rural fixed wireless — was complete by early 2018. Then the squeeze: by December 2018 revenues were not covering bond payments, and member cities were asked to cover a shortfall from property taxes — “Taxpayers asked to pony up for RS Fiber shortfall,” ran the New Ulm Journal headline of December 2, 2018. The cooperative did not fold: its board “successfully negotiated improved loan terms extending through 2026,” and as of early 2026 it is operating, selling gigabit, advertising five-year price locks. The lesson deserves its discomfort: the general-obligation backstop meant the towns bore the construction risk — precisely what made the build possible where two incumbent providers had declined to invest. Charter the risk explicitly, or the op-eds will charter it for you.
Set them beside Burlington one last time. EPB submits its books to audit and its claims to peer review; RS Fiber’s shortfall hit the newspaper and the renegotiation table the same season — pain in public, on the record, survivable. Burlington Telecom’s $16.9 million moved in the dark for years, and the network did not survive the daylight. Monitoring is not a temperament. It is a clause with a calendar — and the next section is where the clause goes.
Across an 1895-lineage irrigation statute, a 1995 fisheries law, a 1998 software constitution, a land-trust bylaw set amended through 2024, and a 1975 town-forest enabling act, the same eight clause types recur — nesting threaded through them all. The Champlain Housing Trust bylaws carry the cleanest modern set, quotable clause by clause.
Membership. A member is “Any person 16 years of age or over who has paid annual membership dues of $1.00” and who “Expresses interest and support for permanently affordable housing, the Corporation, and its purposes”; voting belongs to members “living in the Corporation’s three-county service area of Chittenden, Franklin, and Grand Isle counties” (Art. II §§ 2–3). The dollar is the anti-Schwyz: a boundary anyone can cross on purpose.
Reserved powers. Member assent is required for “removal of Members or Directors,” “sale of certain lands,” “alteration, amendment, or deletion of the Corporation’s limited appreciation formula,” and “dissolution or merger” (Art. II § 3(C)).
Petition floor. A special meeting takes “at least 5% of the voting Membership or 20 voting Members (whichever is less)” (Art. II § 5(C)); the New Hampshire town-meeting parallel takes “25 or more registered voters or 2 percent… whichever is less” (RSA 39:3).
Open books. “The records and minutes of all Membership and Board meetings of the Corporation shall be open to, or available for, inspection by any person upon reasonable request,” with enumerated exceptions; executive session requires a two-thirds vote and an enumerated purpose (Art. VI § 2; Art. III § 11). Burlington Telecom is this clause’s absence — same city, $16.9 million tuition.
Graduated sanctions. Written charges filed with the Secretary; the Board votes whether they are sufficient, and dismissal is final; notice with a cure option — “The Notice may state that a Member may avoid suspension or expulsion by taking specified action”; a hearing before a three-person committee, “one selected by the affected Member; one selected by the Board…; and one selected by the first two”; removal only by “two thirds of the voting Members present at the next Membership Meeting” (Art. II § 8(C)).
Entrenchment. The bylaws amend only by “two thirds of the entire Board of Directors and the affirmative vote of 75% of the voting Members present” on two weeks’ notice (Art. VII), and the resale formula — “the cornerstone on which the Corporation… rel[ies]” — gets its own double procedure (Art. IX).
Dissolution with asset lock. On dissolution, assets pass “[t]o the local non-profit ‘Community Land Trust’ Corporation(s) serving the area(s) closest to the area(s) in which the Corporation owns land,” failing that to another CLT or federation, failing that to a 501(c)(3) that “agrees to administer the assets… in accordance with its goals and purposes” (Art. VIII). The institution can die; the land stays common.
Nesting. One-third of CHT’s board seats are held by municipal officials “from four different cities or towns”; Maine’s councils propose to a commissioner while a nonvoting legislator watches; Randolph’s commission delegates to named clubs; Debian’s treasury lives in separate corporations with “no authority regarding Debian’s technical or nontechnical decisions.”
The whole part compresses into a table of contents you can draft against — sixteen sections, each keyed to the working document that proves the clause earns its ink:
| § | The clause | Proven by |
|---|---|---|
| 1 | Name, purpose, the thing held — description plus map or roll | NH RSA 31:110; acequia water-right roll |
| 2 | Membership — eligibility, nominal dues, one member one vote, service-area voting | CHT Bylaws Art. II |
| 3 | Declaration — each member commits to exactly one unit | 12 M.R.S. § 6447(7)(B) |
| 4 | Proportional contribution — assessed work, or fee in lieu | NMSA § 73-2-21(A)(1), (A)(5) |
| 5 | Officers with defined duties and a fixed reporting calendar | NMSA § 73-2-21(B); Debian Const. § 5 |
| 6 | Reserved powers of the membership | CHT Art. II § 3(C); Debian § 4.1 |
| 7 | Rule-change referendum with supermajority | 12 M.R.S. § 6447(6); Debian § 4.1(2) |
| 8 | Petition floor | CHT Art. II § 5(C); NH RSA 39:3 |
| 9 | Open books — written accounts on a calendar, records public property | NMSA § 73-2-21(D); CHT Art. VI § 2 |
| 10 | Graduated sanctions — charges, cure, peer hearing, supermajority removal | CHT Art. II § 8; 12 M.R.S. § 6447(3) |
| 11 | Conflict resolution — internal first, written reasons, cheap external appeal | NMSA § 73-2-21(E); Debian § 6.3(6), § 7.1(3) |
| 12 | Nesting — seats for neighbors, named user groups | CHT Art. III § 3(C); Randolph ordinance; § 6447(4) |
| 13 | Asset separation — the treasury cannot rule | Debian § 9; the Patagonia split |
| 14 | Tiered amendment — ordinary rules cheap, purpose clauses expensive | CHT Art. VII + IX; Debian 3:1 |
| 15 | Dissolution with asset lock to the nearest sibling commons | CHT Art. VIII |
| 16 | Anti-coercion — volunteers owe no labor, only non-sabotage | Debian § 2.1.1 |
Sixteen sections is the ceiling, not the entry fee. A fire district or a town forest will want most of them; a community fridge or a tool library needs the same questions at one-page scale — who may use it, what exactly is held, who restocks and repairs, who can change the rules, where the money note is posted, what happens after abuse and what cures it, who settles the argument, and where the freezer or the tool wall goes if the group dissolves: asset-locked to the nearest sibling commons, not to whoever still has the key. The questions do not change with the square footage. Only the paper does.
Almost none of it needs writing from scratch: the acequia association’s 400-plus bylaw sets, the Champlain Housing Trust’s bylaws public as a PDF, Debian’s constitution on an open mirror, New Hampshire’s one-sentence enabling act — drafted, court-tested, free to steal. What no template supplies is the adoption: the words taken on, in public, as binding. The first Monday in June has already passed this year; by statute, in the ditches beyond the cottonwoods, the spring’s money is on the table in writing — public property, open to anyone who carries water. That is the oath layer with dirt on its boots. Your forest, your fridge, your fiber deserve the same sentence. Write the clause. Give it a Monday.
For the selectman, the forest committee, the fire district board:
For the renter in the city:
Part VIII
The apprentice pays for himself — if you design the work so he can.
In October 2014, a shuttered hundred-year-old school building on Main Street in Warren, Rhode Island — a town of about 11,000 — opened its doors onto the smell of bread. The renovation cost $3.2 million, financed in large part by a $2.99 million loan from USDA Rural Development’s Community Facilities program — the same federal channel that builds firehouses and town halls, here building a place to work. Inside the old walls: 17,500 square feet, three shared-use commercial kitchens — one a dedicated gluten-free bakery — and an event space where the assembly hall was. The building is called Hope & Main, and a decade in the ledger reads: more than 500 food and beverage businesses helped; annual launches up from 36 in 2023 to 61 in 2025; 148 food startups producing out of the Warren kitchens right now, roughly 120 more in the pipeline.
Hold the picture: it inverts the assumption that kills town economic planning. Warren did not recruit an employer or offer abatements to a distribution center and pray. It took a dead building, borrowed against the federal government’s cheapest rural credit, and built a room where work could start. The town that waits for a company waits forever. The town that builds the room gets the company as a byproduct.
This part is the operational manual for growing work in place, on two instruments. The first grows the worker: the registered apprenticeship, faster to start in 2026 than it has ever been, failing in well-documented ways you can design against. The second grows the firm: the shared plant — the kitchen, the shop floor, the machine no one in town can justify alone. Two readers at once: the employer or selectman in a town of 500 to 50,000 who could convene a program, and the knowledge worker — anywhere — whose task list is being eaten by AI tooling and who needs work that is physical, local, and license-protected. The book argued the why. Here is the how — forms, offices, dollar figures, two complete program designs.
Start with what the registered system actually is in 2026, counted honestly.
The Department of Labor’s apprenticeship.gov dashboards are fed by RAPIDS — the Registered Apprenticeship Partners Information Database System — covering 25 federally administered states and 16 State Apprenticeship Agencies, data through March 18, 2026. Ask how many active apprentices America has and you get two answers from one dashboard at different snapshot dates: 678,014 and 702,610. RAPIDS totals settle upward as late entries post, so the honest count is the range between them. The count itself is a lesson in record-keeping.
The trend underneath the range is real — roughly 680,000 actives in FY2024 against about 318,000 a decade earlier, a 114% increase, with annual completers up 143% — and an April 2025 executive order set a goal of 1 million. Two caveats: roughly a quarter of the headline count is military or federal-prison programs; and the leak — of the roughly 167,000 apprentices who started a registered program in 2017, about 78,000 — 46.8% — completed within six years; in 2023 alone, more than 115,000 left early, and RAPIDS does not record why anyone leaves. Every success number in this part stands on that denominator; both designs below are built to plug it.
What changed in 2026 is speed. On March 9, 2026, DOL’s Employment and Training Administration issued three circulars and a bulletin rebuilding the registration machinery — program design, the role of State Apprenticeship Agencies, how completion rates are calculated, with a public completion-rate portal — and Bulletin 2026-35 committed the Office of Apprenticeship to registration determinations within 30 days, with a public “shot clock” page tracking its own timeliness. “We’re removing the administrative roadblocks that have prevented Registered Apprenticeship from scaling to meet demand,” said Secretary of Labor Lori Chavez-DeRemer announcing the package. The six-month registration limbo is over; the turnaround times are public.
One more piece matters most, because it makes a registered apprenticeship a real thing rather than a title. Federal regulation — 29 CFR § 29.5(b)(5) — requires every registered program to pay “a progressively increasing schedule of wages to be paid to the apprentice consistent with the skill acquired,” starting no lower than the FLSA minimum. Time-based programs start around 50% of the journeyworker rate and step up roughly every 1,000 on-the-job hours to 100%; competency-based programs tie steps to skill sign-offs. The schedule is written into the registered standards — binding, not a guideline — and it is the single feature that most distinguishes a registered apprenticeship from an internship or a “trainee” title. Nor is it only for the trades anymore — more than 40 states now run or pilot registered teacher apprenticeships; Wisconsin added dental-assistant, paramedic, and surgical-technologist programs in 2025; California put $30 million into healthcare, education, and technology tracks that October. The white-collar door is open.
Here is the entire federal process for a three-person shop — the folklore about needing a compliance department is out of date.
Step one. Submit the “Express Interest in Starting a Program” form at apprenticeship.gov, or contact your Office of Apprenticeship state office. In the 16 State Apprenticeship Agency states — Connecticut among them — the state agency, not federal OA, is your registration agency.
Step two. Draft standards in the free Standards Builder: the occupation from the official Occupation Finder list, the training approach — time-based at the classic 2,000 on-the-job hours a year, competency-based, or hybrid — the work schedule, a related-instruction plan against the customary 144 classroom hours a year, the wage progression, and an equal-opportunity pledge.
Step three. The registration agency reviews the standards — within 30 days under Bulletin 2026-35, shot-clock performance published where you can audit it.
Steps four and five. Sign each apprentice on ETA Form 671 — the agreement that triggers RAPIDS entry, recorded within 45 days, federal paperwork estimate 45 minutes — then track and report through the free RAPIDS portal. Registration automatically lists you as a WIOA Eligible Training Provider, unlocking Workforce Innovation and Opportunity Act money for tuition and on-the-job-training reimbursement, awarded by your local workforce development board.
Government fees for all of this: $0. The real costs are wages, mentor time, and instruction, and the best federal evidence prices the overhead precisely: across the 46 American Apprenticeship Initiative grantees, 2015 to 2021, the average cost to generate one new apprentice fell to $5,171 by the final grant quarter — and grantees with prior apprenticeship experience spent $4,867 per apprentice against $8,702 for first-timers. Experience halves the cost — the argument for borrowing someone else’s.
That is what intermediaries are. Apprenti — the national tech intermediary — publishes real numbers: 93% of its apprentices finish the classroom work, 86% of those finish the on-the-job training, 89% of completers stay with the hiring employer, across 350-plus employer partners in all 50 states. FASTPORT does the same for transportation; trade associations like the Pool & Hot Tub Alliance hold DOL contracts to register employers at no charge — the federal contract pays them. The fee reality: beyond “federally subsidized means free to the employer,” price schedules are mostly unpublished — Apprenti negotiates per cohort — so get the number in writing; no rate card exists.
Then the states pay you to do it. Colorado wrote the richest credit in the country: the New & Emerging Industries Apprenticeship Tax Credit — refundable, meaning a small shop with no tax liability gets a check — up to $6,300 for the first six months of employing a registered apprentice plus $1,050 for each additional month, to a maximum $12,600 per apprentice per tax year, ten apprentices, a $126,000 ceiling; construction excluded by design; filed two-step through Apprenticeship Colorado, running through 2035. Connecticut’s corporation business tax credit covers apprentices in manufacturing, plastics, and the construction trades — for manufacturing, the lesser of $4 per hour worked, 50% of actual wages, or $7,500 per apprentice; confirm your entity type’s eligibility with CT DRS before budgeting it. South Carolina, New York, and Arkansas run smaller credits. No general federal credit exists — the federal layer is the Inflation Reduction Act’s prevailing-wage-and-apprenticeship bonus on clean-energy projects, plus grants. Your state’s credit is a fact to look up; FIRST ACTIONS names where.
Below the registered tier sits everything else — bootcamps, fellowships, income-share academies, for-profit training chains. The last decade ran the experiment without the registrar’s paperwork; the results deserve names.
BloomTech — formerly Lambda School, the income-share-agreement flagship — ended under federal order. On April 17, 2024, the Consumer Financial Protection Bureau issued a consent order against BloomTech, Inc. and its CEO, Austen Allred: the $20,000-tuition income-share agreements were loans carrying an average hidden finance charge of about $4,000, and the marketing claimed job placement “as high as 86 percent” while internal metrics showed rates “closer to 50 percent and in some cases as low as 30 percent.” BloomTech was permanently banned from consumer lending, Allred from student lending for ten years; collections halted for graduates who never got the promised jobs. The mechanism was not the curriculum: the claims ran ahead of the measured outcomes, and the financing instrument concealed its own cost.
Multiverse — the best-funded UK import: $220 million Series D, $1.7 billion valuation, 16,000-plus apprentices trained globally for Microsoft, Citi, Google, and Morgan Stanley — withdrew from US registered apprenticeships in July 2024 after losing $50 million in a year. Vice president of corporate affairs Tim Smith did the autopsy: “We carried almost the entire burden of registration for our employer partners, but more than half chose not to engage with the formal registration process, as the reasons for doing so weren’t obvious.” The investor Ryan Craig generalized it in the same reporting: “Without public funding, it becomes a pet project of a CEO. When budgets tighten, the CFO cuts it.” The system itself grew 6% that year while new tech apprentices fell 28%: the growth is real; the venture-scale business model on top of it is not.
And TechShop — the for-profit national makerspace chain — closed every US location without warning on November 15, 2017 and filed Chapter 7 with more than 9,000 active members still paying dues — memberships that could not cover industrial equipment, trained staff, and power across ten company-owned shops. Hold it for the shared-plant section: the failure was the business model, not the demand.
The tier is not poison, though — it is unaudited, and the survivors submitted to someone else’s audit anyway: Year Up United’s federal PACE randomized trial measured earnings 30% above control six years after graduation, the largest effect in any US workforce RCT — and Per Scholas, free and in more than 20 cities, held a 14–15% advantage through year eight of its own trial. Trials someone else ran, against BloomTech’s self-reported 86%. So the consumer-protection sentence of this part: never pay tuition for a program that will not publish audited placement and wage data, and prefer programs that pay you. The registered tier’s paperwork is not bureaucracy. It is the audit, built in — the credential gate is the human gate, where a program either submits to verification or sells you a story.
One more development answers the employer’s 2026 question: why train juniors at all when the AI writes the boilerplate? Revelry Labs, a New Orleans software consultancy, has run a 12-week software apprenticeship for fourteen years — its current chief technology officer is a graduate — and on April 9, 2026 published its AI-era redesign: two weeks of onboarding and deliberate practice with and without AI tools, then nine to ten weeks of real project tickets under a discipline Revelry calls AI apologetics — the apprentice may generate code with Claude Code or Cursor, but must verbally defend every pull request to a senior engineer before merge, “a reasoned, systematic defense of their code.” The program’s binding ritual becomes the defended pull request rather than the logged hour — and DOL institutionalized the same question with its AI in Registered Apprenticeship Innovation Portal, launched April 29, 2026: free program-design tools across healthcare, finance, education, and advanced manufacturing.
Stuart Page gave the why-bother its permanent form: “One day, the apprentice you didn’t hire won’t be the senior who notices that the LLM is on the warpath again… Or hire the apprentice, train the senior and sleep peacefully.”
The case for apprenticeship always points at Switzerland, and the Swiss system runs not on cultural reverence for the trades but on firm-level profit. The national cost-benefit study — Strupler and Wolter, 2012, on 2009 data — found Swiss employers’ gross training costs of CHF 5.3 billion against apprentice productive output of CHF 5.8 billion: a net gain of roughly half a billion francs while the apprentices were still in training. Per position, gross cost runs CHF 20,000 to 30,000 a year against productive performance of CHF 18,000 to 36,000; German firms, by contrast, run net costs and recoup through retention — the Swiss break even or better because apprentices do real productive work early, at a true training wage. Two of three Swiss young people enter an apprenticeship, across roughly 250 occupations — scale that exists because the economics work firm by firm. The translation rule, and the design law of this part: an apprenticeship the employer loses money on is a charity program and will die with the budget cycle; one designed around early productive work survives.
America’s largest attempt to import the Swiss model wholesale has published its own honest accounting. CareerWise Colorado launched in 2016 — founded by the Denver manufacturer Noel Ginsburg after a 2015 Switzerland visit, funded with $9.5 million from JPMorgan Chase and Bloomberg Philanthropies — with a stated vision of 20,000 student apprentices in its first decade. As of May 2026: about 1,200 students through CareerWise Colorado programs, plus roughly 2,000 more through programs it spun up or inspired — about 16,800 short. Ginsburg’s own verdict on the targets: “well-intentioned and ill-informed.”
The mechanism of the shortfall was employer churn, and the churn had reasons with names. It launched with about 45 employers; roughly 25 returned in year two; 29 are active in 2026. Eagle County Paramedic Services quit after one year — the required certifications could only be earned at 18, and insurance would not cover student ambulance drivers. The construction firm RA Nelson quit because school-year scheduling could not match job-site hours — “The hours and the pace didn’t really align,” said regional manager John Halloran. Behind the churn, parents defaulted to college and district internships competed for the same students.
Then the part that earns the post-mortem its keep — what worked. Eagle County: a single full-time coordinator, housed at the local chamber of commerce, and a program shortened from three years to two. Vail Health has employed 27 apprentices to date. Not a platform, not a state office, not a marketing budget — one named human being doing the paperwork, the scheduling, and the school liaison, in the institution local employers already trust. That design fact is the seed of Program Design A below.
The other American translation that held is the network. Aon launched its two-year US apprenticeship in 2017 with City Colleges of Chicago, then co-founded the Chicago Apprentice Network with Accenture and Zurich North America — community-college instruction on company time, degree-optional, registered. Aon has onboarded 400-plus US apprentices since 2017; the network passed 2,000 in July 2024. The structural lesson: the network — not any single employer’s program — was the unit that scaled, because no single firm had to carry recruiting, curriculum, and credibility alone. A town of 5,000 cannot copy Aon. It can absolutely copy the network.
The manufacturing translation owns the best wage number in the literature. FAME — the Federation for Advanced Manufacturing Education, begun by Toyota at its Georgetown, Kentucky plant in 2010 and now run nationally — is a five-semester earn-and-learn: three days a week of paid work, two days of community-college classes, toward an Advanced Manufacturing Technician credential. The Opportunity America/Brookings study — October 2020, Kentucky state earnings data — found FAME graduates earning nearly $98,000 five years after completion, including overtime, against roughly $52,783 for comparable non-FAME manufacturing CTE completers; roughly 80% completed. The largest measured wage premium of any US apprenticeship-style program — measured once, in one state, on pre-2021 earnings; carry the vintage with the number.
The state plumbing converges: Colorado stood up a unified agency plus the refundable credit plus grant money — the most complete state stack — and California now pays intermediaries per apprentice placed rather than by grant. Where your state has neither, the federal machinery works everywhere RAPIDS reaches.
Return to the school in Warren. The apprenticeship needs somewhere to happen — the second instrument is the shared plant, the room that lets ten businesses afford what none could alone.
Hope & Main’s revenue model is ordinary on purpose: hourly kitchen rental, membership tiers, classes, events, a farmers-market retail arm; the incubator reports roughly 40% of member businesses surviving seven years — its own figure against an unstated baseline, so treat it as program-reported. The financing is independently verifiable: the USDA Community Facilities program that carried Warren’s $2.99 million remains open, year-round, to towns of 20,000 or fewer — and for communities of 5,000 or fewer with median household income below the threshold, the grant share runs up to 75%.
The same pattern runs inland. Codefi opened November 7, 2014 in Cape Girardeau, Missouri — 17,000 square feet in the Marquette Tower, more than 350 members — seeded by a $750,000 federal Economic Development Administration award matched by nearly $1.5 million raised locally, and run by a local foundation, not a platform. Its own count: 80-plus startups, over $50 million raised, 300-plus jobs, 7,958 students introduced to programming — program-reported, attributed as such. Generator, the Burlington, Vermont makerspace — six technical shops, 400-plus dues-paying members — supplies the budget honesty: in its best public reporting year, 2018, it ran just over $500,000, at 42% earned income and 58% philanthropy and grants. The planning rule that yields, dated evidence and all: even a well-run makerspace covers roughly two-fifths of its cost from users; a town budgeting one should plan for permanent sponsorship — library-style — not self-sufficiency. That is the lesson TechShop’s 9,000 members paid for.
And the plant can be smaller than a building. A Tormach 1100MX CNC mill — the standard prosumer job-shop machine — lists at $29,479 base, with the automatic tool changer at $6,745. A working two-mill-plus-lathe shop — mills, lathe, bandsaw, tooling, CAM software, all single-phase power — lands well under $150,000 in equipment, under $250,000 with a year of rent and working capital in a rural market. PK Machining in Tennessee grew from one 1100MX in a garage to a five-machine shop on Xometry contract work plus overflow from local manufacturers — a manufacturer-published story, though the machine prices are list. What sells from such a shop in 2026: repair and obsolete parts for local industry and agriculture, fixtures and jigs for larger plants, short-run brackets and housings for hardware startups, prototype machining — the work too small, too fast, or too local for big shops to quote. Towns spend more than $250,000 on a culvert. This is a culvert that hires.
The college ledger, from the College Board’s November 2025 pricing study: in-state public four-year tuition and fees averaging $11,950 for 2025-26, total cost of attendance $30,990 a year, bachelor’s borrowers leaving with an average $35,639 in loans. The apprenticeship ledger, from DOL’s own page, verified live June 10, 2026: $80,000 average starting salary after completion and 90% employment retention — with the asterisk DOL itself footnotes: both figures trace to one state’s linked records — Kansas’s — averages pulled up by union construction scales. Evidence, not a floor.
Four years of in-state public cost of attendance: roughly $123,960 paid out. Four years of registered apprenticeship at a conservative blended $20 an hour, 2,000 hours a year: roughly $160,000 earned, zero debt, ending at a credential associated with an $80,000 start. The gross position swing approaches $280,000. The caveats ride with the number or it does not get used: only 46.8% of the 2017 cohort completed within six years — and an uncompleted apprenticeship confers far less than an uncompleted degree costs; the $80,000 rests on Kansas; lifetime-earnings comparisons still favor degrees in many professional tracks. The claim this guide will stand behind is narrower and stronger: for a worker who completes, the apprenticeship path runs cash-positive the entire way, and in the trades and advanced manufacturing the five-year earnings beat the median bachelor’s outcome.
The design, distilled from every case that worked: three employers — say, an HVAC and plumbing contractor, a machine shop or millwork operation, and the town’s largest white-collar employer, usually the bank branch, the school district, or the medical practice — form one group sponsor under a single set of registered standards, instead of three separate programs. One part-time coordinator, 10 to 15 hours a week, housed at the chamber of commerce, the library, or town hall, does all paperwork, scheduling, and school liaison. The precedents: Eagle County’s chamber coordinator; the Chicago network’s shared credibility; GPS Education Partners, grown from one employer — Generac Power Systems — and five students in 2000 into Wisconsin’s statewide consortium, 1,400-plus graduates across roughly 100 businesses.
The mechanics, in order. The three employers sign a consortium agreement; one entity — the chamber, the economic development commission, or the largest employer — becomes sponsor of record. First, the legal question: does this create an employment relationship — wages, workers’ compensation, child-labor rules? The state apprenticeship office answers the registration side; the town’s risk pool — CIRMA, for Connecticut members — rules on classification; town counsel handles waivers for minors. Then the coordinator files Express Interest — in Connecticut, registration runs through the CT DOL Office of Apprenticeship Training, 200 Folly Brook Boulevard, Wethersfield; that office, or your state’s equivalent on the apprenticeship.gov directory, is the exact desk where this begins. Standards go through the free Standards Builder — HVAC technician, CNC machinist, and medical assistant already exist on the occupation list; the 30-day clock governs the determination; related instruction contracts to the nearest community college — 144 hours a year is one evening class a week plus summer blocks; each hire signs ETA Form 671 within 45 days.
The budget, per apprentice per year: wages of $32,000 to $40,000 at a $16-to-20 training wage over 2,000 hours — offset by the apprentice’s actual production, which the Swiss data say to design for deliberately; instruction tuition of $1,500 to $4,000, often WIOA- or state-covered; coordination near the AAI curve — $5,171 per apprentice experienced, $8,702 first-time — roughly the $20,000-to-25,000 part-time coordinator split across a four-apprentice pool; offset by whatever your state pays, from South Carolina’s $1,000 to $4,000 to Colorado’s refundable $12,600 to Connecticut’s $7,500 manufacturing maximum. For scale: a volunteer fire department routinely sinks more than $35,000 into training and equipping one firefighter. Towns already invest at this size in people — just not in this.
The timeline: month one, employer commitments and the coordinator hired; months two and three, standards drafted and the college agreement signed; month four, the determination on the 30-day clock; months five and six, recruiting — high-school seniors and adult career-changers, by personal ask rather than broadcast, the only method the evidence supports; month seven, the first cohort working. Two-year term, per the Eagle County lesson. The pre-mortem, every failure mode documented: employer churn — CareerWise lost roughly 20 of 45 launch employers after year one — so the consortium agreement states the pool absorbs an apprentice when an employer exits; age-and-insurance walls — Eagle County Paramedic’s 18-year-old certification floor — so check certification ages and liability coverage before choosing occupations; school-schedule mismatch — RA Nelson’s job-site hours — so prefer adult apprentices or block schedules for seasonal trades. Retention — the 46.8% problem — is management, not mystery: recognition, matching people to fitting tasks, apprentices recruiting the next cohort one-on-one.
This design belongs to the other reader: the 35-to-50-year-old marketing manager, paralegal, or mid-level analyst whose task volume is being absorbed by AI tooling, and who wants work that is physical, local, and license-protected — without starting at a 19-year-old’s wage.
The design rule: aim trades-adjacent, not tool-belt-first. The roles that buy your existing skills are the coordination and diagnostic layers of the trades — building-performance and energy auditor, HVAC controls and commissioning technician, electrical estimator, facilities and maintenance management, advanced-manufacturing technician, construction project coordination. Each sits beside a licensed trade, pays within 10 to 20% of it, and rewards the spreadsheet-and-client skills the prior career built.
The path, month by month. Month zero: one visit to the local American Job Center — found through careeronestop.org — to test WIOA eligibility for funded retraining. That office is where the design starts; the intake is free. Months one through five: a structured bridge program, kept while still employed. Merit America — nonprofit, 20,000-plus learners, part-time for working adults, 15 to 20 weeks — reports average wage gains near $21,000 a year, from $27,800 before to $48,500 after, holding above $30,000 at three years; 2025 added a Semiconductor & Advanced Manufacturing track. Per Scholas carries the field’s strongest independent evidence — the 14–15% advantage still standing at year eight. Both satisfy the consumer rule: audited outcomes, and you pay nothing or pay on success. Months six through twenty-four: convert to an earning role — a registered adult apprenticeship through an IEC or ABC chapter or a union JATC, adult entrants standard, prior-learning credit shortening the term under Circular 2026-01’s competency-based approach — or direct hire into the adjacent role. In manufacturing regions, FAME chapters accept adult career-changers into the same five-semester program with the $98,000 five-year record. The wage reality, unsoftened: training wages are modest — the regulation guarantees a rising schedule from roughly half the journeyworker rate, not comfort. Month twenty-four and after: the license or certification — state electrical tracks, BPI building-analyst, or the equivalent in your trade; costs vary by state, verify locally.
The design buys what the registered system was built to guarantee: paid from day one, raises by rule, a credential gate with an auditor behind it. The book argued the apprenticeship is the institution that survives the AI transition. This is what it costs to enter one at 42.
If you run a town or a business in one:
If your job is the one being automated:
Either reader:
Part IX
“…necessity connected with the repair of highways, bridges, sidewalks and water and sewer systems and the care of the town poor, and then not more than one thousand dollars.”
— Conn. Gen. Stat. § 7-348, on the only spending a Connecticut town officer may undertake beyond an appropriation
In early 2016, some of the fastest internet in the United States ran into Pinetops, North Carolina — gigabit fiber, town-owned — and it arrived with an expiration date. The light in the glass came from Greenlight, the municipal network the city of Wilson, population about 49,000, had built through its own electric utility. North Carolina’s 2011 law, H129, fenced Greenlight inside Wilson County; in February 2015 the FCC preempted that law, and Wilson ran fiber across the line — to Pinetops, and to a Nash County family farm with nearly 300 employees. On August 10, 2016, the Sixth Circuit held the FCC had no authority to reallocate power between a state and its municipalities. The fence stood again, dated: cut Pinetops off by October, or put the entire network’s legal basis at risk.
What Wilson did next is this part’s founding move. It read the statute the way town counsel reads a statute — word by word — and found the seam: the law restricted selling service beyond the county. It said nothing about giving it. In September 2016 the council voted to give Pinetops its internet free for six months. The legislature answered in 2017 with a narrow special law, H396 — valid only until a private provider arrived; when Suddenlink built into Pinetops, Greenlight was forced out for good.
Wilson had capital, competence, customers, and the install base. What it did not control was the law — just a statute, an agency’s reach, and a court’s reading of the boundary between them. Every move in this guide — hash register, mesh, commons charter, apprenticeship pool, CERT roster — ends at the two questions Greenlight answered the hard way: is this legal here, and who pays? You Cannot Eat Code argues that the town is the unit of account. A unit of account has legal capacity and a balance sheet; this part is the map of both. The constraint layer is not the enemy of the Town Stack; it is the oldest layer of it — and the towns that move fastest walked the walls before they built.
Begin with the doctrine that decides what “town” means in your state. In 1868, Iowa Supreme Court Justice John F. Dillon wrote the rule that still carries his name — City of Clinton v. Cedar Rapids & Missouri River R.R. Co., 24 Iowa 455, 475 (1868), validated by the U.S. Supreme Court in Atkin v. Kansas, 191 U.S. 207 (1903): a local government holds only the powers the state expressly grants, those necessarily implied, and those essential to its declared purposes. Home rule is the counter-theory — a zone of local initiative the state cannot casually invade; Part I ran the national arithmetic.
But the label decides less than you would hope: the live constraint is preemption, and the preemption decade — 2020 through 2026 — hit the home-rule states hardest. Texas, strong home rule, passed HB 2127 in 2023 — the “Death Star” law — barring cities and counties from any field occupied by the state’s major codes, agriculture to labor to property; Houston, San Antonio, and El Paso sued and won at trial on August 30, 2023, and on July 18, 2025 a Texas appeals court reversed, leaving HB 2127 standing. Florida answered Orange County’s November 2022 rent-stabilization vote — nearly 60% approval, increases capped at 9.8% — with the Live Local Act, signed March 29, 2023, banning local rent control statewide; HB 1417 followed, undoing 46 tenant-protection ordinances across 35 cities and counties. Arizona’s HB 2686, February 2020, preempted municipal gas bans before any Arizona city had proposed one — prophylactic, pushed by Southwest Gas; Tucson Mayor Regina Romero: it “needlessly micromanages cities.” And a second sovereign waits behind the first: the Ninth Circuit held Berkeley’s pioneering gas-piping ban federally preempted in California Restaurant Ass’n v. City of Berkeley (2023). Even a home-rule city answers to two capitals.
Broadband keeps the starkest file. Tennessee confines Chattanooga’s flagship municipal fiber to its electric footprint by statute, Tenn. Code Ann. § 7-52-601, and the decision that re-fenced it, Tennessee v. FCC (2016), is the same Sixth Circuit ruling that turned off the light in Pinetops. Sixteen states still restricted municipal broadband as of 2024–2025, though the barriers are falling state by state. Towns win one now and then: Ohio’s legislature overrode a veto to ban local flavored-tobacco rules; Columbus and 20 other cities sued; and on July 9, 2025, Ohio’s Tenth District Court of Appeals unanimously held the preemption violated the Ohio Constitution’s home-rule provision — Columbus v. State, 2025-Ohio-2408. The precise lesson: constitutional home rule occasionally holds the line; statutory almost never does.
Connecticut, the worked example, is a hybrid — Dillon’s Rule baseline softened twice: by the Home Rule Act of 1957, CGS §§ 7-187 to 7-201, and by Article Tenth of the state constitution, adopted 1965 — no special legislation aimed at an individual town; towns may adopt and amend charters. The baseline grant is CGS § 7-148, as amended by PA 23-207 — a long enumerated list, construed in the Dillon manner: the operative question is not whether some statute prohibits a town’s enactment but whether there is “statutory authority for the enactment.” A charter adds three powers under CGS § 7-194 — finances and property, conveyances, contracts and evidences of indebtedness: bonds. When charter and statute collide, Cook-Littman v. Fairfield Board of Selectmen, 328 Conn. 758 (2018), draws the line: “in an area of local concern… general statutory provisions must yield to municipal charter provisions governing the same subject matter,” while “matters that concern public health and safety… have traditionally been viewed as matters of statewide concern.”
Goshen has never adopted a charter. Fifty-six of Connecticut’s 169 municipalities have not — 113 have — and the Litchfield hills are statutory country almost without exception: the book’s home ground runs on the state’s default operating system — selectmen, town meeting, and the § 7-148 list. The list is generous, but a move that falls outside it needs a charter, a special act, or a state statute — and knowing which before the meeting is the difference between counsel’s letter and counsel’s apology. The city reader holds the inverted problem: a charter, broad home-rule powers, and a legislature that has already occupied field after field. Her first question is never “may my city act?” but “has the state already acted?” — the LawAtlas state-preemption dataset maps the closed fields.
Now the money, where the epigraph began: no Connecticut town officer may spend or contract beyond a department’s appropriation, and the lone exception — highways, bridges, sidewalks, water and sewer, the care of the town poor — stops at one thousand dollars. In a statutory town the board of finance builds the budget and the annual meeting adopts it (CGS Chapter 106); the board may keep a contingent fund up to 3% of estimated expenditures; and mid-year money runs through CGS § 7-348 — in a town with a grand list of $20,000,000 or less, any additional appropriation that alone or cumulatively exceeds $10,000 per department per year must be voted by town meeting; above that grand list, $20,000. Last raised in 1990, by P.A. 90-23 — thirty-six years of inflation ago.
The threshold is the law’s own pilot-design rule; read it twice. As a constraint: any unbudgeted experiment above it is, by law, a public vote — size the pilot under the line, or build the case that carries the meeting. And as a discipline: a records register, a mesh seed, a CERT class all fit beneath it, but the threshold is a way to start, never to govern — a pilot that cannot win $12,000 at a town meeting has a problem no appropriation mechanics will fix. The line is an option and a calendar. It is not the answer.
Elsewhere the envelope itself is capped: 46 states and the District of Columbia restrict property taxation in some form. New York holds levy growth to the lesser of 2% or inflation — still 2% for 2026 — override by 60% vote; Massachusetts runs under Proposition 2½ (1980): levy capped at 2.5% of full cash value, growth at 2.5% a year plus new growth, override by majority referendum. Connecticut caps no town levy — the only brakes are the board of finance and the meeting itself — though it caps the motor-vehicle mill rate at 32.46 mills (PA 22-118). The reading in a cap state: every new recurring obligation competes inside a fixed envelope — assume the general-fund door nearly shut, and plan from the first sketch for override politics, district formation, or outside money. The city reader’s nearest town-meeting equivalent is the participatory-budget line, where her city runs one — a small door, but real.
When the project outgrows the annual budget, the town borrows — and the law is friendlier than the market. Connecticut authorizes municipal bonds at CGS § 7-369 and sizes the ceiling at CGS § 7-374 in multiples of annual tax receipts — 2.25 times for general purposes, 3.25 for urban renewal, 3.75 for water pollution control, schools and pensions classed separately, tax-anticipation notes and grant-backed bonds excluded. The market is the regressive part: cost of issuance averages 3.096% of principal on issues under $10 million against 0.741% above $75 million — the small issuer pays 4.2 times the big one’s rate, because the costs are fixed: one small California school district paid bond counsel $40,113, the rating fee $9,500. The escape hatch, built in 1986, is the bank-qualified bond of IRC § 265(b)(3): an issuer selling $10 million or less per calendar year may designate its bonds bank-qualified, so local banks deduct 80% of their carrying costs and buy the paper directly, no underwriting apparatus — worth, GFOA estimates, 25 to 40 basis points: roughly $232,000 to $370,000 over a 15-year $10 million bond. The cap has not moved since 1986 ($30 million briefly under ARRA in 2009–2010), so the hatch stays exactly small-town-sized. So the capital stack for a town of 500 to 5,000 is not a public offering at all: a bank-qualified bond or direct bank placement, a USDA Community Facilities loan, a state grant, and a capital reserve.
Connecticut keeps a second instrument on the shelf — the most underused legal object in this guide’s toolkit: the special taxing district of Chapter 105, CGS §§ 7-324 to 7-329, a self-funding unit smaller than the town, born from fifteen signatures. The mechanics, exactly: on petition of fifteen or more voters specifying the limits of a proposed district, the selectmen must call a meeting of the voters within those limits within 30 days; a two-thirds vote of those voting establishes it (200 voters or 10% of the district, whichever is less, may demand a referendum). The result is a municipal corporation — fire district, sewer, lighting, sidewalk, recreation, “improvement association” — levying its own tax, set at district meeting, collectible with the town tax’s lien power; hundreds exist statewide, some under special acts predating May 29, 1957. Know the tool, then weigh it: a district funds what the whole town will not — and routes around the town meeting doing it, sometimes the point, sometimes the wound; a legitimacy question before a legal one, taken up in Part VII where charters get written.
The patient instrument needs one vote and no counsel: CGS § 7-360 lets any municipality’s legislative body create, by majority vote, a “reserve fund for capital and nonrecurring expenditures” — uses widened by PA 16-180 to planning, construction, acquisition, revaluation. The town that wants a records vault or a communications layer in three years opens the fund this year — no debt, no counsel’s invoice, no issuance spread — and when a grant window opens, the match is already in the account.
Now the grant channels — and first the weather, because the climate changed in 2025. On May 8, 2025, the $2.75 billion Digital Equity Act ended by social-media post — the post pronounced the act unconstitutional; NTIA’s termination letters went out May 9 — killing its three programs: $1.44 billion for state capacity, $60 million for planning, $1.25 billion competitive. Connecticut got a notice; Vermont lost $5.3 million, Maine $35 million; more than 20 states sued on June 24, 2025, and a year later nothing had been restored. The library money ran the arc in reverse: a March 14, 2025 executive order moved to dismantle the Institute of Museum and Library Services; 21 state attorneys general won a permanent injunction in November 2025 in Rhode Island v. Trump; the administration dropped its appeal April 6, 2026, a parallel suit settled three days later, terminated grants reinstated — and the FY2027 budget proposal zeroes the agency anyway. The fight moved from courts to Congress; one fact survives every round — the library money flows through your state library agency, the office to call.
The structural lesson: a federal program can now die by post, resurrect by injunction, and face zeroing again inside one budget cycle. Programs are weather. Two rules follow. Never design a plan whose first domino is a federal funding notice. And treat the project file as the asset, because the windows open on Washington’s calendar, not yours: a current, approved hazard-mitigation plan — the eligibility ticket every mitigation channel checks first — a scoped project, a benefit-cost sketch, ready to file into whichever window opens. Most federal money reaches a town through the state, which collects, ranks, and forwards sub-applications on its own earlier deadline — the federal date is never yours. The steadiest window follows disaster: FEMA’s Hazard Mitigation Grant Program opens after a presidential disaster declaration, projects gathered and ranked by the state hazard mitigation officer — in Connecticut, at DEMHS. The town that gets funded is the town whose file was current the day the declaration came.
The standing channels reward the same preparation. USDA Rural Development’s Community Facilities program is alive, year-round, built for this readership: towns, villages, and tribal lands of 20,000 or fewer residents; money to purchase, construct, or improve “essential community facilities” — firehouses, libraries, town halls; a grant share up to 75% for communities of 5,000 or fewer below the higher of the poverty line or 60% of the state nonmetropolitan median income, its FY2026 notices publishing on schedule. Philanthropy built this book’s reader a channel: the Trust for Civic Life — launched 2024 by Rockefeller Brothers Fund, Stand Together, and Omidyar Network, now 21 funders committing $50 million over five years to rural civic life — has moved $16 million to 46 grantees across two rounds, grants of $150,000–$500,000 over three years, the 2025 round alone reaching 382 rural counties, and its $5,000–$25,000 Civic Experiment micro-grants, moved through hub grantees, are sized to a first move — 2026 priorities: the Black Belt, Central Appalachia, Tribal lands, the Southwest border, rural communities in economic transition. The states are the quieter survivors — the legislatures passed more than 160 broadband bills in 2025 — and Connecticut’s Historic Documents Preservation grants, the clerk-filed channel at $5,000, $7,000, or $10,000 by town size, are detailed where they belong, in Part II.
One body of law gets one sentence because Part II carries all of it: the e-records and signature statutes — UETA, E-SIGN, URPERA and Connecticut’s enactment — reduce for a town to a single principle, hardening is legal everywhere; substitution requires the records administrator’s blessing, and Part II walks every clause and case.
The question that kills more town programs than any statute is a fear: what if someone gets hurt, and the town gets sued? The honest answer is a perimeter you can walk. Connecticut’s CGS § 52-557n makes a political subdivision liable for its employees’, officers’, and agents’ negligence in the scope of duty — and immune for “negligent acts or omissions which require the exercise of judgment or discretion as an official function.” Two carve-outs do most of the litigating: ministerial acts — duties prescribed so exactly that no judgment remains; “inspect monthly” in a written policy converts discretion into ministerial duty — and the identifiable person facing imminent harm, where discretionary immunity falls away. From which follows the drafting rule that runs against every careful instinct — Part VI applies it to CERT charters; it governs every document in this book: the more precisely the papers command — “shall, within 24 hours” — the more immune discretion converts into suable duty. Write “may” and “as resources permit” unless the town intends to be bound.
The volunteer protections stack three deep — Part VI prices them for response teams. The federal Volunteer Protection Act of 1997 — 42 U.S.C. §§ 14501–14505 — shields the individual volunteer, compensation capped at $500 a year, and leaves the entity’s liability untouched. State volunteer and Good Samaritan statutes — Connecticut’s are cataloged in OLR 2013-R-0199 — forgive nothing grossly negligent. And CGS § 28-13 immunizes the state, its subdivisions, their agents, and “members of the civil preparedness forces” complying with civil-preparedness law, willful misconduct excepted, workers’ compensation attaching under chapter 568 and CGS § 5-142. Every layer turns on the same switch — enrollment, activation, paper: FEMA’s CERT Liability Guide (April 2021) walks the state variance, and the never-enrolled group or self-deployed member stands outside the structure.
What the town itself keeps, no volunteer statute removes: ordinary negligence in ministerial functions — maintenance of the thing you built, the mesh pole, the tool-library bandsaw; premises liability; workers’ compensation for volunteers who qualify as employees; civil-rights claims under 42 U.S.C. § 1983, which no state immunity statute can block; contract and procurement disputes. The management is clerical: enroll formally, activate in writing, inspect on a schedule the town can actually keep, insure. The insurance is cooperative — small towns buy through member-owned intergovernmental risk pools: California builds them on the joint powers authority of Government Code § 6500 et seq.; Connecticut’s is CIRMA, formed 1980 inside the Connecticut Conference of Municipalities, its Workers’ Compensation and Liability-Auto-Property pools serving nearly 90% of the state’s public-sector market. Part VI states the launch test; it generalizes: call the pool underwriter before the program goes live and ask what enrollment, inspection, and waiver paperwork would make it insurable — uninsurable means unaffordable, whatever the budget line says. One absence completes the perimeter — a frontier line, not an all-clear: no appellate decision from 2020 through 2026 about a novel town technology program — a mesh network, a records-anchoring layer — has yet surfaced; the litigated terrain remains roads, sidewalks, schools, police. The liability frontier lags the technology frontier; first movers will write the precedent — one more reason to do the paperwork as if a judge will read it.
Some Town Stack moves have no statute written for them; they have precedent, and it arrives in three fates — a survivor, a casualty, a felony — most legible in money itself. The bright lines are old and federal: 18 U.S.C. § 486 criminalizes making or passing metal coins “intended for use as current money,” whether or not they resemble U.S. coin, and 18 U.S.C. § 336 — the Civil-War-era Stamp Payments Act — bans notes under one dollar meant to circulate. Paper notes of one dollar or more, pegged to the dollar, treated as taxable income, and imitating nothing federal are lawful. That is the BerkShares design; BerkShares is the survivor: launched 2006 around Great Barrington, Massachusetts — roughly 140,000 paper bills in circulation, just over 300 businesses, six exchange offices at two local banks, more than $10 million circulated since 2006, a digital app since 2022 — par, 1.5% only to convert back. It lives because it is embedded in banks — regulated institutions that outlive founders. Ithaca HOURS, launched 1991, is the casualty: it decayed when founder Paul Glover left town, because Glover had personally been the matchmaking layer and no institution inherited the job. The Liberty Dollar is the felony: Bernard von NotHaus minted silver “Liberty Dollars,” and on March 18, 2011, after a 90-minute jury deliberation, was convicted under 18 U.S.C. § 486 — the U.S. Attorney called the scheme “a unique form of domestic terrorism” attempting “to undermine the legitimate currency of this country”; the December 2014 sentence: three years’ probation, six months home detention. The boundary in one line: paper yes, coins no; pegged yes, sovereign-pretending no.
The municipal-network fights compiled the catalog of lawful workarounds: give the service away where the statute restricts selling (Wilson, 2016); obtain narrow special legislation (H396, 2017); take the opt-out referendum where offered — more than 120 Colorado communities voted out of SB-152 before its 2023 repeal; stay inside the electric footprint (Chattanooga); repeal the statute (Arkansas and Washington, 2021; Minnesota, May 2024). Every entry came from reading a statute’s exact words rather than its reputation — why the first dollar of an unusual project goes to the lawyer who reads the text, not the vendor who quotes the build.
Where communities organize force rather than fiber, the line is brighter. After the Unite the Right rally, Charlottesville and its local businesses — represented by Georgetown Law’s Institute for Constitutional Advocacy and Protection — sued the armed groups that had deployed on its streets: City of Charlottesville v. Pennsylvania Light Foot Militia, filed October 2017, under Virginia’s constitution, anti-paramilitary statutes, and common law. Within eight months, nearly all the defendants — the Pennsylvania and New York Light Foot Militias, the III% People’s Militia of Maryland — had entered court-enforceable consent decrees barring their return as armed, coordinated units, and ICAP published a 50-state catalog of anti-paramilitary tools in February 2018. What endures is the chartered, subordinated volunteer — the CERT team enrolled under a state emergency-management statute (Connecticut’s Chapter 517), activated by the emergency-management director; the volunteer fire company, the oldest continuously legal community-defense institution in America. No successful challenge to a properly enrolled CERT program has surfaced — an absence, not a guarantee, conditional on the enrollment itself. Part VI builds to the rule; this part confirms it as law: community capability organized under civil authority is two centuries of settled law; organized beside it, a consent decree waiting to be signed.
Everything above compresses into a walk. It is research, not legal advice — the final word belongs to town counsel — and three offices answer most questions before counsel bills an hour: the town attorney, the Secretary of the State, and the state municipal league — in Connecticut, CCM, whose risk pool already serves the town.
Harden the record — Part II. First question: who is the legal custodian of this record, and what does the state retention schedule require before anything is moved, copied, or migrated? Who answers: the town clerk, then the state public-records authority — in Connecticut, the Office of the Public Records Administrator at the State Library: signed approval precedes any destruction, stricter still for pre-1900 originals. Redundant layers — microfilm, distributed copies, cryptographic anchors — need no one’s permission; replacing the official copy does. The money: the clerk-filed state preservation grant; the state library agency for the federal library money, if Congress keeps it alive.
Build the mesh — Part IV. First question: does my state restrict municipal communications networks — and is a mesh a “network” in the statutory sense at all? A LoRa mesh on unlicensed Part 15 spectrum, carrying no commercial broadband, generally sits below what even the sixteen restrictive states regulate — where a statute exists, ask first. Who answers: the state broadband office — in Connecticut, DEEP’s broadband group — CCM, and town counsel; the spectrum needs no license. The money: the § 7-360 reserve; a $5,000–$25,000 Civic Experiment micro-grant; EMPG pass-through ($319.5 million nationally in FY2025, 50% match, via DEMHS); USDA Community Facilities for gear housed in a town hall or firehouse.
Charter the commons — Part VII. First question: what legal wrapper holds the common assets — a special taxing district, a nonprofit, or a municipal board? Who answers: town counsel on the choice; the Secretary of the State for incorporation; OPM, in Connecticut, on districts. If the commons must tax, the § 7-325 district sits on the shelf — fifteen signatures, a called meeting, a two-thirds vote; if it must flex, a 501(c)(3) with the town as member or contract partner. The money: the district’s own levy; a community foundation; Trust for Civic Life hub grants, $150,000–$500,000 over three years.
Grow the work — Part VIII. First question: does this arrangement create an employment relationship — wages, workers’ compensation, child-labor and registered-apprenticeship rules? Who answers: the state labor department’s apprenticeship office — Connecticut’s DOL Office of Apprenticeship Training — on registration; the risk pool on volunteer-versus-employee classification; town counsel on waivers for minors. The money: registered-apprenticeship channels through the state DOL; USDA Community Facilities for the workshop in a town under 20,000; local match from the reserve or the school budget.
Stand up response — Part VI. First question: who activates us, and are we enrolled as civil-preparedness forces so immunity and workers’ compensation attach? Who answers: the municipal emergency-management director, then the state agency’s regional coordinator — DEMHS in Connecticut; CGS Chapter 517 governs, and FEMA’s CERT Liability Guide (April 2021) is the national reference. The money: EMPG at 50% match; the mitigation projects in the standing file, ready for the next post-declaration window; the pool consultation, free to a member town.
For the renter, the tree compresses to three questions. Has the state preempted my city here (LawAtlas; the city attorney’s published opinions)? What sub-city unit can hold money and act — business improvement district, special assessment district, community board, neighborhood council, the participatory-budget line where the city runs one? And which community foundation or civic-life funder covers this neighborhood? Smaller names; the same tree.
Back to the seam in the fence. Wilson never defied North Carolina, and never waited for permission: it read the statute in public, found the one verb the law regulated, and bought a neighboring town six more months of light while the legislature caught up. That is the posture. The walls are real — Dillon’s rule, the preemption codes, the thresholds, the caps. The gates are real too — the enumerated list, the bank-qualified hatch, the fifteen-signature district, the reserve fund, the standing file, the underwriter’s phone line. Walk the walls before you build. Then build.
For the selectman, the finance chair, town counsel:
For the renter in the city:
Part X
Twelve rungs, priced in hours and dollars.
In the first week of July 2021, with the Outer Cape running at full summer heat — Route 6 stacked with beach traffic, the parking lots full by mid-morning — the town of Wellfleet, Massachusetts counted its civic bench and found 46 openings on 47 committees. Not forty-six people serving — forty-six empty chairs. The historical commission, the seven-member panel that decides whether a historic building’s demolition gets delayed long enough for anyone to object, was down three members. Next door the weather was the same: Eastham, 30 vacancies across 42 committees; Truro, 33 across 37; Provincetown, 16 openings across 45 — with no applicants at all. Christine Legere put the numbers in The Provincetown Independent on July 7, 2021, and the story sank without a ripple — none of it was news to anyone who has ever tried to staff a town.
It is not a Cape Cod problem. In March 2025, Maui County’s 37 boards and commissions were running an 18% vacancy rate — 57 of 304 seats unfilled — with multiple boards, in Honolulu Civil Beat’s word, “paralyzed” for a year because they could not raise a quorum. Behind it, the old conveyor belt into every other kind of service — weekly religious attendance — had fallen from 42% of American adults in the early 2000s to 30% in 2024. The appointed tier of American government, the tier closest to your house, is short-staffed from one ocean to the other.
The elected tier is emptier still — here the anecdotes become a national accounting. In 2024, 70% of the races that appeared on American ballots were uncontested — an all-time high — and the gradient steepens as the office gets closer to home: 91% of regional races, 82% of county races, 74% of local races carried one name or none. This is not one bad cycle or one partisan data shop: the figure was 67% in November 2022 — against just 3% of federal races and 35% of state positions — and 70% in 2020. Three cycles, two independent databases, same answer: roughly seven of ten ballot lines in this country carry exactly one name, or none. By position: treasurer, 92% uncontested; district attorney, 88%. The most systematic academic study — Melissa Marschall and John Lappie, across roughly 10,000 mayoral elections — found 48% of American mayoral races uncontested, and found an uncontested race depresses turnout by roughly 9 percentage points: the vacancy is not neutral; it shrinks the electorate around it. Iowa ran the table at 85%; 21 states crossed 50%; and New Hampshire — the state that sizes its offices small — came in under 18%, the outlier that proves the door swings. Where office is built to human scale, people still contest it. Everywhere else, the bench sits empty.
The book this volume accompanies argued the town is the unit of account — where trust, judgment, and accountability are manufactured. This part, the guide’s spine and close, makes the operational claim: the power nearest you is not guarded by money, credentials, or connections. It is guarded by nothing. Between you and it stands a ladder of twelve rungs, each priced before you climb — hours, dollars, evidence. The bottom rung costs an evening. The top rung costs a career. Both readers can start this week: the reader in a town of 3,000, and the city renter, for whom the same vacancy structure repeats one tier down, in community boards and neighborhood councils with documented double-digit vacancy rates.
Before the first rung, the one mechanism the whole ladder runs on — measured for a quarter century, every measurement pointing the same way.
In the fall of 1998, Alan Gerber and Donald Green ran the first randomized field experiment in the modern get-out-the-vote literature — in New Haven, Connecticut, on the guide’s home ground. In-person canvassing substantially raised turnout; commercial phone banks did nothing — possibly less than nothing. The experiments since have settled the price: door-to-door canvassing produces roughly one additional vote per 14 people contacted — about $19 a vote at $16-an-hour canvasser wages. That is the retail rate for a stranger at the door.
Now make the contact a friend. In the 2018 midterms, a randomized trial by Outvote and Columbia University found a get-out-the-vote text from a friend raised the recipient’s turnout by 8.3 percentage points. A mass text from a stranger raised it by 0.29 points. The personal ask beat the broadcast by a factor of about 28. If you retain one sentence from this volume, that is the sentence. It replicated in unfavorable territory: in Florida’s 2024 cycle, the Center for Campaign Innovation found voters texted by someone they personally knew were 8.6 percentage points likelier to vote than controls, while a parallel 100,000-message stranger-texting program produced no measurable effect.
Inside that Florida study sits the failure case that completes the lesson: to recruit its volunteers, the program sent 424,179 broadcast texts to a universe of 538,078 low-propensity voters. It got 101 active volunteers — a conversion rate of 0.02% — who then texted 1,738 people they actually knew. The broadcast could not manufacture relationships; it could only locate the rare person who already wanted in. Every platform will tell you to scale the message. The evidence says the message does not scale — the relationship is the medium, and being asked — by a person who knows your name — has been a first-order predictor of civic participation since Verba, Schlozman, and Brady documented it in Voice and Equality in 1995.
The organizing science agrees. Hahrie Han’s comparative study of civic associations sorts their leaders into lone wolves, mobilizers who maximize headcount through transactional asks, and organizers who develop members’ capacity to lead — and finds the highest-engagement chapters blend the last two: broadcast to find people, then relational work to transform them — an assignment with real autonomy, a person who knows their name, a visible line from task to outcome. Keeping them is management, not magic: the Urban Institute’s study of 2,993 charities found the median organization retained 80% of its volunteers year over year, while nearly 3% retained none, and retention tracked recognition, training, matching people to fitting tasks, and volunteers recruiting one-on-one. A prior UPS Foundation study found two-fifths of volunteers had quit an organization over bad management, and issued a verbatim warning: “Poor volunteer management practices result in more lost volunteers than people losing interest because of changing personal or family needs.” Bad management loses more people than life does.
The mechanism, in full: show up in person, ask by name, give real work, recognize it. Every rung that follows is an application.
Rung one: get a library card and attend one public meeting. The program is whatever your town calls its governing body — board of selectmen, town council — or, in New York, your community board’s full-board session, agendas posted under open-meetings law. Two to three hours, once. Zero dollars. No prerequisite. The evidence that the act matters: civic attention at public meetings is now scarce enough that the Documenters Network pays residents — in Chicago- and Cleveland-style sites, $15 an hour plus transportation — simply to attend and take structured notes. The city reader needs no appointment and no standing: every community board’s public session is open, and attending one now is how the later rungs unlock.
Rung two: learn three neighbors’ names and do one favor. No program exists for this rung, which is the point — it is the informal layer everything formal stands on. One to two hours a month, zero dollars, no prerequisite. 54.2% of Americans — 137.5 million people — already operate in this layer of favors, tools lent, rides given. The evidence is the strongest in this entire part because it is meta-analytic: across 148 studies and 308,849 participants, people with stronger social relationships showed a 50% higher likelihood of survival over an average 7.5 years of follow-up — territory comparable to quitting smoking, larger than the measured effects of obesity or physical inactivity.
The city has already built rung two into hardware: the community refrigerator at 1 Pulaski Street in Bed-Stuy, started in March 2025 by civil-rights lawyer Keegan Stephan — restocked up to three times a day, coordinated through a neighborhood Signal chat, supplied partly by volunteers intercepting distributor surplus, and so heavily used by late 2025 that donations which once lasted until Monday were gone by Friday. Its governance is the smallest commons charter in this book: one host with standing — a yard, an outlet — one open channel, several restock partners. That is the entire constitution — and the formation capital is on the table: CitizensNYC grants up to $5,000 to resident-led groups — block associations, tenants’ associations, groups still forming, no 501(c)(3) required — plus Collaborative Impact Grants up to $12,000, window open through July 27. Three neighbors and an idea are an eligible applicant.
Rung three: make the personal ask — bring two people to the next meeting. The venue is rung one’s. One hour. Zero dollars. The prerequisite is rungs one and two — the ask only works from someone who has shown up and is known. The evidence is the mechanism section: 8.3 points for the friend, 0.29 for the stranger. This is the hinge rung — everything above it runs on the muscle built here. The raw material is abundant: in the year ending September 2023, 75.7 million Americans — 28.3% of the population sixteen and over — formally volunteered through an organization — what the Census Bureau and AmeriCorps called “the largest expansion of formal volunteering” since tracking began in 2002. But the commitment is thinning as it widens: average hours per volunteer fell from 96.5 a year in 2017 to 70 in 2023; the median fell from 40 to 24. America is full of people willing to show up twice. The ladder’s whole job is converting two appearances into a duty — and the conversion tool is a person who asks.
The next four rungs share a shape: a named program, open enrollment, a defined training cost, a credential or assignment at the end. All four work identically in a village or a borough, and none requires anyone’s permission.
Rung four: work the polls on November 3, 2026. The program is your registrar of voters — or helpamericavote.gov, which routes to every state’s signup; National Poll Worker Recruitment Day is August 11, 2026. The honest price is roughly 18 hours: a two-to-three-hour training class plus an Election Day running from about 5:00 a.m. until after the polls close — 15 to 17 hours on your feet. The rung pays rather than costs: New York City pays $250 for Election Day plus a training stipend, up to $2,750 for multi-day assignments; Bristol, Connecticut pays $275 to $340 a day by position. The prerequisite, in most states, is being a registered voter. The evidence is a staffing crisis in plain sight: more than 772,000 poll workers served in November 2024, 48% of jurisdictions reported difficulty recruiting them, and the majority of poll workers are over 61. The machinery of the franchise is run by retirees, short-handed, and hiring.
Rung five: take a paid Documenters assignment. The program is documenters.org — founded by City Bureau in Chicago, now spanning 25 communities across the country — where you complete a short orientation and claim meetings from a public assignment calendar. Three to five hours per assignment, no standing commitment in between, pay through the same site-specific stipends as rung one. The prerequisite is rung one, once. The evidence is scale: in 2025 alone the network published 3,253 assignments and paid its Documenters more than $369,000, crossing $1 million paid since launch. For the knowledge worker it converts a professional skill — structured note-taking — into civic infrastructure, and pays you while you learn how your town works. Where no network city exists, the rung still stands: attend, take minutes, publish. The scarcity is the attention, not the platform.
Rung six: CERT Basic. The program is FEMA’s Community Emergency Response Team — more than 3,200 local programs across all 50 states, tribal nations, and territories, findable through the locator at community.fema.gov. The price is 24 hours of instruction and exercises, typically spread across several evenings or weekends — fire safety, light search and rescue, team organization, disaster medical operations — then periodic drills at the local program’s cadence. Dollar cost: zero; many programs issue the gear. No prerequisite. Nor is it a relic: FEMA refreshed the curriculum and ran Train-the-Trainer deliveries into mid-2026; Austin ran trainings June 1–4, with another set August 21–23, 2026. In Connecticut the door is the town emergency management director or fire marshal; in a city, the borough or municipal Office of Emergency Management. It is where you stop being someone to manage in an emergency and become someone to assign.
Rung seven: the amateur-radio Technician license. This is the guide’s hardware rung — the licensed node of Part IV’s communication stack — and the cheapest federal credential you will ever hold. The program is a volunteer-examiner session, findable by ZIP code at hamstudy.org/sessions; the all-in cost is about $50 — a $15 exam-session fee and a $35 FCC application fee on grant. The exam is 35 multiple-choice questions drawn from a published public pool of roughly 410; 26 correct — 74.3% — passes. Study time is unofficial — prep-vendor estimates vary — so work the pool until practice scores clear the bar. No prerequisite. The questions are public, the fee is fixed, and the license is the difference between owning a radio and being a station.
Above rung seven the ladder changes character: evenings become seasons, and in exchange the rungs carry institutional weight — a credential, a roster, a fiduciary duty, a vote.
Rung eight: the extension credential — Master Gardener or Master Naturalist. The program is your state’s land-grant extension service, and Connecticut prices the worked example: UConn’s Master Gardener program runs a $495 fee and 100-plus hours of classroom instruction over 16 weeks, plus 30 hours of in-office service at an Extension Center or the Bartlett Arboretum and 30 hours of community outreach, all completed by the end of September. A reader acting now targets the 2027 cohort — applications historically open in August. The cheaper end of the same shelf: Texas Master Naturalist chapters — the largest such program in the country — charge $135 to $200 for 40-plus hours of basic training, with 40 volunteer hours and 8 advanced hours to certify; most states run analogues. No formal prerequisite, but rungs one through three make the service hours land locally instead of evaporating into a logbook. The evidence is the dose: the service requirement lands you at almost exactly 100 volunteer hours a year, and in a prospective study of 12,998 adults over 50, followed four years, volunteering at or above 100 hours a year was associated with a 44% lower mortality risk and a 17% lower risk of impaired physical function — and, honestly read, with no detectable effect on diabetes, chronic conditions, cancer, or hypertension incidence, though a separate cohort of 1,164 older adults found 200-plus hours a year cut the odds of developing hypertension by 40%, so the studies disagree on that margin and you should know it. The ladder does not promise immortality. It promises the one dose-response curve in this book where the dose is measured in casseroles and trail miles.
Rung nine: join the fire company’s civilian side. The program is Fire Corps — run by the National Volunteer Fire Council since 2004 — placing community members in non-emergency roles at local fire and EMS departments: administration, fundraising, public education, rehab support. Two to ten hours a month, zero dollars; rung six helps. This is the firehouse door built for the person who cannot answer a 2:00 a.m. tone-out, and the institution behind it is structurally short-handed: volunteers are 62% of America’s firefighters — 635,100 of 1,018,100 — and of the nation’s 29,452 fire departments, 18,873 are all-volunteer and another 5,335 mostly volunteer, donating time worth an estimated $46.9 billion a year. The ranks hit their recorded low in 2023 while call volume has more than tripled in 40 years; in 2024, 32 of the 72 American firefighters who died in the line of duty were volunteers. The operational path runs through MakeMeAFirefighter.org, where departments post openings, and there the price inverts: the department spends what can exceed $35,000 to train and equip one firefighter — you pay in hours, not dollars, Firefighter I typically earned on evenings and weekends at regional fire schools over several months, where Connecticut’s full-time career-track Recruit program, for scale, runs 800-plus hours over 14 weeks.
Rung ten: take a stewardship or trusteeship seat. The land-trust route runs through the Land Trust Alliance’s locator: as of the 2020 National Land Trust Census, roughly 950 member land trusts — operating in 93% of US counties, holding 61 million conserved acres — were supported by more than 250,000 volunteers, and the classic recurring role is stewardship monitoring: walk a conserved parcel annually, file the report. The library route runs through the town clerk or the library director: in Connecticut, library trustees on governing boards are public officers under Chapter 190 of the General Statutes — legally responsible for budget, director, and policy. Westport shows the mechanics: a 20-trustee board, half appointed by the Representative Town Meeting, half by the board itself, candidates submitting a resume and letter of interest. Price either route at two to six hours a month as a working estimate — no federal survey measures it; the documented norm is one board meeting a month plus committee work — and zero dollars. The prerequisite is rungs one through three plus the real credential, which no application form lists: a year of visible reliability. The city translation is the tool library, the commons with a membership ledger: the Chicago Tool Library runs pay-what-you-can annual memberships against a recommended $100; Baltimore’s Station North Tool Library prices membership at $1 per $1,000 of annual income and lends from an inventory of 3,000-plus tools. A renter can join one this week and be helping govern it within a year — same fiduciary muscle, different building.
Rung eleven: the appointed seat. Return to where this part opened: Wellfleet’s 46 openings were never a lament — they were a listing. In a Connecticut town the program is the standing vacancy list: Simsbury publishes one outright — “Vacancies on Simsbury Boards and Commissions as of February 1, 2025” — and most towns keep the same list at the town clerk’s or first selectman’s office, where the application process is asking for it. The seats are real: planning-and-zoning alternate, inland wetlands, ethics. In New York the program is the community board — 59 boards citywide, up to 50 unsalaried members each, two-year terms, appointed by the borough president, half the nominations coming from the district’s Council members. The 2026 season ran on hard deadlines — Brooklyn closed February 6 at 11 p.m., Queens February 13, the Bronx February 23, Manhattan February 27, Staten Island rolling — so a reader who missed it preps for the cycle reopening around December 2026 and keeps attending now. Price it honestly: official Brooklyn guidance says expect 6 to 8 hours a month, other guidance at least 10, a minimum of two committees, unpaid — call it 6 to 10 hours a month, $0. The prerequisite is demonstrated attendance — rungs one and three, plus a duty rung like five, eight, or ten — because appointing authorities choose faces they have already seen in the room. The evidence is the vacancy map itself: Maui’s 57 empty seats of 304, Wellfleet’s 46 of 47 committees. The gate is not locked. Most years it is propped open with a folding chair.
Before the last rung, the skeptic deserves a hearing, because the question — does any of this return anything? — has been studied at every tier of evidence.
The economic returns are real, and they are conditional. The NYU Furman Center found New York City’s Business Improvement Districts raised commercial property values inside their borders by roughly 15 percentage points over comparable properties outside — but the effect came almost entirely from large BIDs with budgets over $1.2 million; small BIDs, under $263,000 — half the city’s total — produced no measurable change. Organization moves value when it is capitalized; a lawn-sign association does not. The same ledger runs at garden scale: properties within 1,000 feet of a new community garden gained 9.4 percentage points within five years, the largest effects in the poorest neighborhoods. Compress the audit to one sentence: two hours a week of organized local service is associated with a mortality-risk reduction — 44% — in the same range as the best-studied lifestyle interventions, on cohort evidence, and the neighborhood-scale economic returns are real but conditional on actual capitalized organization, not sentiment. That is the defensible promise. Not transformation. Returns — audited, conditional, and large.
What the record cannot tell you is the upper rungs’ full hourly price, because nobody has measured it. The National Volunteer Fire Council states flatly that “there is no national average of the amount of time a volunteer firefighter gives to his or her community” — and no federal survey prices the selectman’s evenings either. The hour-cost of America’s smallest offices is mostly unmeasured — which tells you how long it has been since anyone who measures things thought they mattered. So the guide prices what is documented and flags the rest: the lower rungs cost 2 to 20 hours a year and $0 — or pay you; the middle rungs, 25 to 160 hours in year one and $0 to $500; the top rungs, 100 to 800 training hours, or a job. Which is exactly why they are scarce, and why towns repay them with authority.
The twelfth rung this volume can describe but cannot climb for you, so here is precisely what exists and what it costs.
In Connecticut the ballot door has two hinges, both statutory. The first is the party caucus: in most small towns, candidates for municipal office are endorsed by enrolled party members in caucus under Connecticut General Statutes § 9-390 — notice published in a newspaper of general circulation at least 5 days ahead, endorsements made between the 56th and 49th day before the primary and certified to the town clerk by 4:00 p.m. on the 48th. Walking into an announced caucus night with a handful of friends is frequently enough to be endorsed — the room is rarely crowded. The second hinge needs no party at all: a petitioning candidate qualifies with signatures of electors equal to the lesser of 1% of the votes cast for that office at the last election, or 7,500. Run the arithmetic and the number turns almost embarrassing — where 1,500 votes were cast for first selectman, the threshold is 15 signatures; where 800 were cast, it is 8. The signatures are free and the filing is free; a small-town race spends shoe leather and evenings before it spends dollars. A dozen neighbors and a filing at the town clerk’s office — that is the wall around the office.
The city runs the same arithmetic at a different address: Los Angeles has 99 certified neighborhood councils — elected bodies inside city government, each holding a real discretionary budget: $25,000 a year under the adopted city budget, down from $32,000. In the 2025 season, 88 of the 99 held elections, and total ballots cast came to roughly 0.3% of the city’s 2.1 million registered voters — a historic low. The translation: a candidate who personally recruits 50 voting neighbors — rung three, run at full power — can usually win an elected seat that directs $25,000 a year. The city reader’s caucus night is a community-board application deadline; the city reader’s empty selectman’s line is a neighborhood-council ballot; the mechanics differ and the vacancy structure is identical.
Price the seat in hours from the documentation that exists: 10 to 20 a month for the candidacy, through a municipal cycle. Then the office. Lincoln, Vermont — population about 1,200 — tells prospective selectboard members the board meets the first and third Tuesday at 6:30 p.m., that meetings run at least 2 hours, and that members should “plan to have almost weekly involvement in some aspect of this work, even if it’s just responding to email,” with a small annual stipend at year-end. Two formal evenings a month plus continuous low-grade attention — call it 8 to 15 hours a month, an estimate, because no survey has ever counted it. At the very top, the honest framing: in most of the 100-plus Connecticut towns governed by selectboards, first selectman is a full-time chief-executive job — the top rung is a career change, not a volunteer hour. The prerequisite is rungs one, three, and eleven. Or plain nerve.
And the evidence — the best single piece on the entire ladder — is the number this part opened with: 70% of American races uncontested in 2024, 74% at the local level, roughly half of mayoral races with no challenger at all. Set the two facts side by side and let them grind: the seat that governs your land, your library, your firehouse, and your kids’ school is, statistically, standing open — and the one mechanism proven to move a neighbor 28 times better than broadcast is a personal ask you could make tomorrow, free.
The book that sent you here ended on a question, and it was careful not to answer it for you, because the answer cannot be issued — it can only be enacted, by a particular person, on a particular street, in a particular town. This part has no authority to close that question, and it will not try. What it can do is what a ledger does — certify the price. Two hours for the meeting. One ask. Eighteen hours for the polls. Twenty-four for CERT. Fifty dollars for the license. A hundred hours for the credential. Six a month for the seat at the table. Fifteen signatures for the line on the ballot. Every rung named, every rung priced, every rung open this week — and above the last one, no gatekeeper of any kind. The truest description of American local power in 2026 fits in two sentences. The seat is not guarded. It is abandoned.
If you live in a town:
If you rent in a city:
Either reader: