Worked Case: From Records Request to Verified Findings

Corvin County, its officials and every record below are invented. The Idaho statute is real, and the steps work anywhere.

The tip

Nell Sorensen is a retired bookkeeper who runs a small citizen budget-watch group in Corvin County. In August 2026 a neighbor told her: "Ask about the county fuel cards. Somebody's filling up a lot." No name, no number. That's a tip. It's a reason to ask for records and nothing more.

Before starting, Nell sketched who would do what.

Stage 1: learn the subject

Nell asked a deep research tool how county fuel card programs usually work and what controls auditors look for. The report ran 11 pages. She kept four ideas: cards are usually tied to a vehicle, an employee PIN or both; logs usually record odometer readings; auditors compare gallons with tank capacity; and weekend or after-hours use is a common red flag that often has innocent explanations. She ignored its national fraud statistics, which linked to nothing she could open.

Then she found Corvin County's own fleet policy online. It said cards are assigned to vehicles, employees must enter an odometer reading, and personal use is prohibited. That policy, not the AI's summary, was the yardstick.

Stage 2: the records request

She used the Idaho FOIA / Public Records Letter Generator to build the request under Idaho Code Title 74, Chapter 1, then adjusted the list by hand.

Idaho Code 74-103 gives the agency three working days to grant or deny a resident's request, or up to ten working days if it notifies the requester in writing that it needs more time to find the records. Corvin County sent the notice on day three and the records on day nine.

Stage 3: reading the response

The county sent four things: a spreadsheet of 6,218 fuel transactions, a card assignment list, a 96-page audit, and a two-page memo about one employee, with some lines redacted.

The spreadsheet, card list and audit are public records about public spending. Nell uploaded them to a cloud AI tool. The memo was different. It concerned a named employee's discipline. She did not upload it anywhere. She read it herself in four minutes: it said the county had opened a review of one road department card in March 2026 and the review was "ongoing."

For the audit, her prompt asked for pages, not conclusions:

[audit PDF attached]
List every page that mentions fuel, fleet, vehicles, purchasing
cards or credit cards. One line per page: page number and the
exact sentence. If none, say NONE FOUND.

It found five pages. Page 61 held the goods: the auditor noted that the county "does not reconcile fuel card transactions to vehicle logs." She read pages 58 to 63 in full. She also searched the PDF herself for "fuel" and found a sixth mention on page 74, in a table the AI had skipped.

Stage 4: analysis

She asked the AI tool, with code, for the row count, columns, date range and duplicate count first. It reported 6,218 rows, matching the county's cover letter, and 12 exact duplicates. She asked the county's finance office about those; it said they were card swipes the vendor recorded twice and later reversed, and sent the reversal records. She removed them, leaving 6,206.

Fuel spending by fiscal year

Corvin County's fiscal year runs Oct. 1 to Sept. 30. The data covers all of FY2025 and 10 months of FY2026. Comparing a full year with a partial one would be wrong, so she compared the same 10 months, October through July, in both years.

The AI's summary said spending rose "about 40 percent." Nell did the math.

Which cards?

She asked for gallons by card, compared with the same 10 months a year earlier. One road department card, R-07, stood out: 1,840 gallons in the earlier period, 5,420 in the later one. That's 3,580 more gallons, or 29 percent of the county's entire 12,330-gallon increase (3,580 divided by 12,330 = 0.290). It was the card in the memo.

The AI also flagged 14 cards for "suspicious weekend fueling." Nell asked the road supervisor for the 2026 shift roster. Eleven of the 14 belonged to plow and patch crews scheduled on weekends. Two were sheriff's patrol cars, which run every day. One, R-07, had 63 weekend fill-ups with no matching shift on the roster. The other 13 flags would have smeared people who did nothing wrong.

Tank check

The card list said R-07 was assigned to a 2019 pickup. Nell looked up the manufacturer's spec for that model: a 26-gallon tank. The log showed 41 transactions on R-07 over 26 gallons. Nobody pumps 34 gallons into a 26-gallon tank, though cans can be filled. She noted it as a question, not a conclusion.

Stage 5: asking the people involved

No AI here. Nell called the fleet manager, who said the county had flagged R-07 internally in March and moved the employee to other duties. She called the sheriff, whose office was handling a referral; at the Sept. 9 commission meeting he spoke about it on the record. She reached the employee through an attorney, who said his client "followed instructions" and would say more later. The commission chair said a new fuel system that fixes the audit finding goes live in January.

She decided not to name the employee. He hadn't been charged, the review was open, and the public-interest issue was the county's missing controls. The SPJ code asks journalists to weigh that kind of call, and the same reasoning fits anyone presenting findings in public. No tool makes it for you.

Stage 6: putting it together

Her point: Corvin County bought 12,330 more gallons of fuel this year, nearly a third of them on one road department card, while an audit warned the county wasn't checking fuel purchases against vehicle logs.

She built a two-page fact sheet for the commission from her evidence table and used AI in a few narrow ways. An outline check found one gap ("commissioners will ask what happens to the employee"). A tightening pass on two long paragraphs, with the no-new-facts rule, cut 61 words. Ten title options gave her a starting point; she fixed the best one by hand: "Corvin County fuel use jumps 23%, one card accounts for nearly a third."

Then she checked the transcript summary the AI had made of the Sept. 9 meeting. It quoted the sheriff as saying "we found no misuse." At 47:12 in the recording, he said: "We haven't found misuse yet." Those are opposite claims. She used his exact words.

Stage 7: verify and go public

She pasted her final version and asked the AI to list every factual claim, one per line. It found 31. Her ledger had 29. The two extras: a sentence saying the pickup was "assigned to the road supervisor" (it was assigned to the department, not a person, per the card list) and the year the fuel card program started, which she'd added from memory. She fixed the first and cut the second.

She linked the audit and fleet policy from the county's own site and ran the fact sheet through Source Checker, which flagged one link to a news story about Idaho fuel prices as a retelling. She replaced it with the original source. Article Values Graph showed "employee misconduct" as a large circle. Her evidence for misconduct was thin by design; the issue was controls. She cut a paragraph speculating about the cans, and kept the tank numbers as a question the county hadn't answered.

Her disclosure line, at the bottom of the fact sheet:

An AI tool helped tabulate 6,206 fuel card transactions and locate pages in the county audit. Every figure was verified against the county's records and every quote against the meeting recording. No personnel records were entered into AI tools.