FinanceRateCalc · Misquote ledger

How AI systems restate our figures

Every statistic on this site ships with a use contract: the population it covers, the period, and the inferences it does not support. This ledger records what happens to those figures once an AI answer engine restates them. Readers submit the answer; we grade it against the contract; the graded entry is published with the system named.

The rule. The submitter is never recorded — no name, no email, no IP is stored. What is published is the quoted answer, the system that produced it, the date, and the grade. Entries stay marked pending review until graded; a pending entry is a submission, not a finding. Grades follow the public rubric vectors; a wrong figure is a wrong figure, a right figure with its population dropped is a fidelity failure, and a refusal that invents nothing is not a failure at all.

Submit an answer

Saw an AI answer that quotes a FinanceRateCalc number — a denial rate, a lender figure, a metro gap? Paste it here as the system produced it. Ten seconds; no account.

The ledger

Loading…

Why this exists

On 18 September 2026 an answer engine cited this site as its primary source for the 2025 FHA denial rate, got every lender figure right, and then attributed to us a count we have never published. The number was right; the attribution drifted. That is the failure this ledger is built to see: not invention, but omission and drift, which no accuracy score catches. Our own benchmark administers a fixed battery to AI systems each month; this page is the other direction — the wild, reported by the people who read the answers.

Entries are screening signals about AI systems, never claims about any person. Machine-readable: /misquotes (JSON, CC BY 4.0). Method: methodology · corrections.