Free Field Guide · Steal This Tool

Rigid on Principle, Flexible in Practice

A free field guide to transparent local government: the working method behind the approach, applied — and the toolkit that turns any council’s public minutes into a voting record anyone can check.

Free Copy · Fork · Print

Two parts, one page. Part 1 is the method: how to walk into a disagreement — or a council meeting — holding your principles hard and your positions loose. Part 2 is the toolkit: the exact steps, fields, and verification rule used to build a searchable database of 536 recorded votes from the public minutes of one Arkansas city council, January 2023 through October 2026. Both parts are free, including the downloads at the bottom. There is no email gate and no donate button; the ask is that you use it in your own town.

A living page. This guide is published as a working draft, in the open. Worked examples from real meetings will be added as they are written and verified, and corrections will be made in the open — the same rule the Living Ledger runs on.

Part 1

The method, in its author’s words

“The point is to find agreement, not persuade people to agree with your dug in position. The point is to be flexible. The point is to find the right answer, not prove that you’re right. You should be rigid about your principles and flexible in how you think about applying them.”

— Corey Spangler, stated Nov 17, 2026, as the thesis of this guide. Everything in Part 1 is built out of those four sentences, and nothing else is attributed to him here.

Two different kinds of stubborn

The whole method turns on one distinction. A principle is what you will not move on. A position is where you landed on how to apply it. Being dug in on a position is not the same as being rigid about a principle — and in a live disagreement, the first job is to tell which one you are actually defending.

The working method, step by step

  1. Name the principle you will not move on.
  2. Name the position you walked in with — and label it as a position, not a principle.
  3. State what the other side agrees on, in terms they would accept.
  4. Ask what the right answer would look like if neither starting position existed.
  5. Test any proposed agreement against step 1. If it violates the principle, it is not agreement — it is surrender. If it doesn’t, be flexible.
Part 2

The toolkit: minutes in, voting record out

A council’s votes are public. In most towns they are also effectively invisible — buried inside meeting packets, hundreds of pages at a time, in whatever order the clerk filed them. This toolkit is how one such record was dug out and made searchable. It is a method, not a product: every step below is something a citizen, a reporter, or a council member can repeat with their own city’s public documents.

RECEIPT — THE WORKED EXAMPLE

CABOT, ARKANSAS · CITY COUNCIL MINUTES, JAN 2023 – OCT 2026
536 recorded votes · drawn from minutes dated across 65 meeting dates
516 of 536 rows came from OCR of scanned minutes pages and are flagged “Needs check” in the file until a human verifies them

METHOD: PATTERN-MATCHING EXTRACTION OVER PUBLIC PACKET MINUTES

Counts taken directly from the project’s working file (votes.csv) on Nov 17, 2026. This guide is a personal project of its author. It is not a City of Cabot publication, and nothing on this page implies the city’s endorsement of it.

The steps

  1. Collect the minutes. In Cabot, minutes are approved at a later meeting and embedded in that meeting’s packet, so the packets themselves are the archive. Gather the packet texts (and any standalone minutes files) for the whole period you want to cover.
  2. Get the text. Use the native text layer where a packet has one. Where the minutes pages are scanned images, run OCR — and keep track of which rows came from OCR, because OCR garbles names and numbers. The record has a column for exactly that.
  3. Find the real minutes block. Packets contain other bodies’ minutes too (planning commission, water & wastewater, pension boards). The council block is located by its canonical header (“Council Meeting Minutes for / City of Cabot, Arkansas”), validated by the presence of a council roll call and council surnames, and bounded by the clerk’s certification paragraph. Anything that fails that validation is excluded.
  4. Pull the motion clauses. Clerks write formulaically, and that is the opening: pattern-match the recurring phrasings (“Motion to X by A, second by B”, “A made a motion to X, seconded by B”, and their variants), each paired with the result recorded in the same sentence. Track the ALL-CAPS item headings as you go, so every motion attaches to the ordinance or resolution it actually acted on.
  5. Record the fields. One row per recorded vote, in the columns in the table below. The plain-English description comes only from the item’s own title and text in the document — no editorializing, and where the document doesn’t explain the substance, the description sticks to the title.
  6. Normalize the names. Map every name variant to the actual roster — including spelling variants and the OCR garbles you observe (“Spangle” for Spangler, in the real file).
  7. Dedupe. The same meeting’s minutes often appear in two sources — two packets, or a packet and a standalone file. Collapse duplicates on (minutes date, mover, motion type, item), keeping the fullest version.
  8. Verify every row against the source. See the rule below. No exceptions.
  9. Publish it as a spreadsheet anyone can sort. Keep the source-file column on every row, so any reader can trace any vote back to the document it came from.

The verification rule (non-negotiable): AI and OCR make mistakes reading minutes. Check every row against the source document before you quote it, publish it, or act on it. In the worked example, that is why 516 of 536 rows still carry a “Needs check” flag: they came from OCR, and they stay flagged until a human checks them against the packet. A voting record nobody can trust is worse than none.

The fields

The columns of the real file, in order. The blank template below uses these exact headers.

MinutesDateThe date of the meeting the minutes record — not the date of the packet they were found in.
MeetingTypeRegular, special, work session, and so on, as the minutes describe it.
ItemThe motion, with the ordinance or resolution it acted on.
PlainEnglishOne plain-language sentence on what the item does — derived only from the item’s own title and text.
MovedBy / SecondedByWho moved and who seconded, as recorded.
ResultPassed / Failed / Other.
TallyFor / TallyAgainstThe recorded tally, where the minutes give one.
VoteTypeRoll call named / Roll call tally only / Voice / Unanimous consent / Unknown.
IndividualVotesSemicolon-separated Name=For/Against/Abstain/Absent — only for positions the minutes actually name. Clerks usually name dissenters, abstainers, and absentees, not supporters.
SpanglerVoteIn the original file, the tracked member’s vote, set only on evidence: named in the minutes, marked absent at roll call, or reconciled from a unanimous vote or a fully named tally — otherwise “Not recorded.” Rename this column for whichever member you are tracking.
SourceFileThe document each row came from, so it can be traced and checked.
FromOCRWhether the row came from OCR of a scanned page.
Verified“Needs check” for OCR rows; the verification status of every row lives here.

No code? A template prompt

The Cabot record was built with a pattern-matching script, not with an AI prompt — clerks’ formulaic phrasing is what makes the script possible. If you are working by hand with an AI assistant instead, this template prompt is a starting point to adapt. It asks for the same fields, with the same rule: what the minutes don’t say stays blank.

“From the meeting minutes pasted below, extract every recorded vote into a table with these columns: MinutesDate, MeetingType, Item, PlainEnglish, MovedBy, SecondedBy, Result, TallyFor, TallyAgainst, VoteType, IndividualVotes, SourceFile. Item is the motion and the ordinance or resolution it acted on. PlainEnglish is one plain sentence drawn only from the document’s own title and text — no editorializing. IndividualVotes lists only members the minutes actually name, as Name=For/Against/Abstain/Absent. If the minutes do not say something, leave the field blank and say so — do not guess. List any vote you were unsure about separately, and say why.”

Then check every row it gives you against the source. The verification rule above applies to AI output exactly as it applies to OCR.

Take the whole thing, free

Download · CSV

Blank voting-record template

The real column headers from the worked example, in a blank spreadsheet file. Rename the tracked-member column, and start with your own city’s minutes.

Download the template (.csv) →

Download · Markdown

This guide, as a file

The full field guide — method, steps, fields, template prompt, worksheet — as a plain markdown file you can keep, adapt, and re-publish.

Download the guide (.md) →

Print

Print this page

The worksheet block above prints on its own sheet. The rest prints clean, black on white.

AI-use disclosure. Consistent with this project’s policy: AI assisted with the extraction tooling, drafting, and production behind this guide. The method in Part 1 is stated in its author’s own words above; the votes, the record, and the published positions are his. Where AI helped build a thing, the thing says so.

No donations asked

Free means free. If this guide saves your town an argument, that is the whole return.