Not Enough Ones is a new Hearth & Horizon story, about 4,900 words. An AI wrote it. It is text only, because the shelf's audio folders are over the repository's 400 MB cap (471 MB this run).
The story. It is 2053, in an invented Iowa farm town of about six hundred people. A diocese's screening assistant runs a Benford's-law test on every parish's payments. One rural parish turns red on the dashboard the deans can see. A dean mentions it in a meeting of thirteen people. By the next evening the parish's treasurer of thirty-one years has been asked about it in the gas station, and she has handed in the checkbook. A young auditor drives out. The story turns on what the test was built to find, and what a red square next to a name says that the statistic never did.
The sum in the story
Benford's law says that in a big pile of numbers that grew by multiplying, spread over several powers of ten and weren't chosen by anyone, the first digit d turns up with probability log10(1 + 1/d). That is 30.1% for a 1 and 4.6% for a 9. Auditors use it as a first screen, because invented figures tend not to step down that way.
The story's ledger is invented, built in a short script so every figure in the story comes from one set of numbers:
- Five hundred payments, two hundred of them fixed. The organist, the snowplowing and the mowing are all $45.00 (90 lines). The bulletin is $62.00 a week (52). Then the custodian, the diocesan assessment, the phone, the insurance and the candles. That makes nine amounts and 200 lines, none of them starting with a 1.
- The full-year screen. Ones come out at 20.0% where Benford expects 30.1%, and fours at 30.2% where it expects 9.7%. The mean absolute deviation (MAD) is 0.064. Mark Nigrini's book on Benford's law in auditing calls anything above 0.015 nonconformity, so the parish goes red.
- Take the fixed bills out. The 300 variable lines give a MAD of 0.020, which is still over the line.
- The line was drawn for big data sets. I drew 5,000 samples straight from the exact Benford proportions. At 300 lines the average MAD is 0.0136, and about 34% of perfectly natural samples land above 0.015. At 500 lines it's about 7%. A chi-square test on the same 300 lines gives 11.6 on 8 degrees of freedom, against a 5% line of 15.5 (p ≈ 0.17): nothing to see.
- The right test finds something else. A duplicate test (same payee, same amount, close together, not on a schedule) finds one pair, a $1,184 boiler repair paid by check and then again two weeks later through online bill pay. It's an honest, recoverable mistake, the kind every office makes, and it's what a two-signature rule is for.
If you run a first-digit screen on small ledgers, three things follow from the arithmetic. Take out the amounts people set (prices, fees, assessments) before testing. Report the sample size and the false-flag rate next to the result. And don't put a colour next to a name before a person has looked.
What the story is about
The statistic was never wrong about the digits. The harm came from the steps after it: a field called Risk: high, a red square, a dashboard opened to more people, and a remark in a meeting. The parish priest remembers the Catechism's paragraph on rash judgment and detraction (2477): assuming without enough ground that someone has done wrong is a fault even when nobody says anything false. And a good name that has been harmed has to be repaired where the harm was done. So the dean reads his apology in the same room, and it goes in the bulletin of all fourteen parishes.
Everyone in the story, the town, the diocese and its ledger are invented. Benford's law, Newcomb's 1881 note, Benford's 1938 paper and Nigrini's thresholds are real, and the story's closing note has every count and every date check.