Dais · Accuracy
Two blind tests against a board's own adopted minutes.
Two regular meetings of a California school board were drafted from the public recording and the posted agenda alone. The drafts were then compared, line by line, with the minutes the board adopted at a later meeting. Every item was there. Every motion matched on who moved, who seconded, the tally and the outcome. The misses were in the narrative, and they are listed below.
| Meeting | Draft built from | Items matched | Motions correct | Non-unanimous votes |
|---|---|---|---|---|
| Regular meeting, August 3, 2026 | Posted agenda, public recording | 60 of 60 | 29 of 29 | One 3 to 1 vote, the single No attributed to the right member |
| Regular meeting, September 14, 2026 | Posted agenda, public recording | 50 of 50 | 25 of 25 | None that night |
| Both | 110 of 110 | 54 of 54 |
Motions correct means the mover, the seconder, the aye and nay count, and carried or failed all match the adopted minutes. A motion that matched on three of the four would have counted as wrong.
Method
How the test was run
- Two public inputs.
The agenda as posted on the district's agenda portal, and the meeting recording as posted on the district's public video channel. Nothing from the district's staff, and nothing from inside the room.
- Draft first, answer key later.
For the September 14 meeting the draft was finished on September 16. The board did not adopt its minutes for that meeting until September 28, so no version of the answer existed in public when the draft was built. For the August 3 meeting the adopted minutes were already public; they were downloaded and set aside unopened until the draft was done.
- Machine comparison.
A script matched every numbered item by its agenda number, then compared each motion's mover, seconder, tally and outcome. The August answer key was a scanned PDF with no text layer, so it was OCR'd first; three apparent mismatches traced to OCR damage in the scan ("Nay: Q" for "Nay: 0") and were confirmed as matches by reading the page image.
- Human read of the rest.
Narrative paragraphs, names, titles, clock times and length were read side by side and the differences written down. Those are the misses reported here.
Findings
What matched, and what a clerk would still correct
| RIGHT | Structure Every section and every numbered item, in order, in both meetings. The August draft also reproduced the board's own layout decisions without being shown them: consent items pulled for discussion moved into their own section, and each consent item printed with its own motion block. | 110 of 110 |
| RIGHT | Actions Mover, seconder, tally and outcome on all 54 motions, including the one split vote and the one item whose seconder differed from every other item that night. When the chair misspoke a tally and corrected it on the recording, the draft took the correction. | 54 of 54 |
| RIGHT | Blanks Call to order, the public comment period before closed session, the closed session report-out and the adjournment time all happened off the public recording. The draft marked each one for the clerk and guessed nothing. | 4 flagged, 0 guessed |
| MISS | Names in the narrative The adopted minutes name the trustees, student board members and administrators who spoke during reports. The draft kept several of them generic ("a trustee asked"). The agenda does not list who will speak, so the draft had nothing to reconcile against. | Clerk fixes |
| MISS | One title, one surname A presenter's title came out as a near synonym of the one in the adopted minutes. A public commenter's surname, which appears on no agenda, kept the speech recognizer's spelling. | Clerk fixes |
| MISS | Length and register The draft wrote a sentence in a few places where the clerk wrote "None", and ran about ten percent shorter than the clerk overall. Three motions matched on every fact but used different wording for the motion sentence. | Clerk fixes |
Limits
What this does not show
- One boardBoth meetings are from the same district, in the same minutes format, with a chair who announces every tally aloud. A different format is a different draft; the first meeting with a new district is the real test.
- Voice votesWhen a chair takes a voice vote and announces no count, the draft leaves the tally blank and flags it rather than printing a number nobody said. That board's third meeting had fourteen of those on consent.
- Off cameraAnything that happens before the stream starts or after it ends is a blank with a flag. The draft never carries attendance or times forward from a prior meeting.
- Two meetingsTwo is a start, not a track record. The next adopted set of minutes gets scored the same way and the table above gets a new row.
The fastest way to see what this looks like on your own minutes is your next regular meeting. Send the date and the draft comes back to you.
Send your next meeting datetyler@districtfiscaltools.org. If you got an email from us, replying to it does the same thing. About Dais.