Accounts receivable
Who owes us, and what to do about each one
We read every receivable on this book against the filed paper, put an outside model's estimate beside each one, and order the worklist by the dollars that estimate puts at risk. The question the estimate answers: is this bill still unpaid ninety days after the due date printed on it, given that it has not been paid?
✉
The open book
The book from the filed paper · the risk column from submitted/invoice_risks.csv
Reading the book…
How this page was made
Where the numbers come from
- Every balance is read out of
corpus/invoice/and netted against the advices incorpus/remittance/, at the desk’s own date: the book is derived on every read and stored nowhere. - The risk estimate and the action beside a row come from
submitted/invoice_risks.csv— a file produced outside this repository, by a model that is not in it. It is not company paper: no features, no training months and no coefficients travel with it, so this page prints the estimate, says whose it is, and prints no arithmetic for it. It is joined one record to one open bill, in both directions, and a file that covered part of the book is refused rather than drawn as a shorter column. - The action beside a row is that file's own string, verbatim, out of a vocabulary of three — and on this book the three of them hold the aging buckets one apiece, so it reproduces them exactly and the action chart restates the age chart. What day rule earned which string is not readable from this file: no bill on this book stands between 5 and 39 days late, so the groups are observed here and the thresholds belong to a model that did not travel with the numbers. The office's own instruction on collection is
corpus/policy/POL-COLLECTIONS.md; nothing on this page asks a crew to stop. - What this column replaced. Until 2026-10-16 the same column was fitted inside this repository, as a discrete-time hazard on the filed year, anchored closer to the due date than this one is, and it printed its own ceiling, its held-out months and its calibration loss beside every row. That code, its tests and its derivation scripts were removed; the model exists only in git history up to commit
43922b9, and the figures it last reported are kept indocs/bookkeeping-desk.md— which is where this page stops quoting a number it can no longer derive.