See it first: three of our figures compiled end to end (source, contract, safe sentence, blocked restatements, receipt, verification, build provenance), and the open Claim Contract Compiler you can run on your own figures today.
When an answer engine cites a statistic, the number usually survives. The sentence around it usually does not: the population, the period, the program scope, the attribution, and what the figure does not establish. We measured this on our own figures for six weeks, with a public instrument, and found the failure is mostly omission and drift, not invention. Accuracy monitoring does not see it. We can run the same instrument on yours.
You give us a page or a list of up to 20 published statistics. We administer them to the major answer engines and to API models under fixed conditions, grade every answer against the figure as you published it, and return a report you can act on.
| what you get | detail |
|---|---|
| Per-figure grade | value correct / contract met / partial / wrong / abstained, with the answer text archived and dated |
| Failure codes | 14 codes from our public taxonomy: scope missing, scope wrong, attribution drift, fabricated support, source laundered, stale value, causal leak, individual leak, recommendation leak, and others |
| Which engine does what | per-system rows, fractions of runs, never averaged into a score that hides the pattern |
| Your side of the ledger | every case where the engine was faithful and your page was stale or self-contradictory; we found thirteen such families on our own site, and we will find yours |
| Fixes | a contract-shaped sentence for each figure (the form that survived best in our experiments), and what to put in your structured data |
$950 for up to 20 figures, five working days. $1,900 with three monthly re-runs so correction latency is measured, not assumed.
Every figure on this site carries a receipt like ⟦FRC:national-fha-denial-rate-2025:22.1%:ebf6b00b⟧: claim id, value, and a hash of the claim that changes if the figure is ever corrected. Anyone, human or machine, can check a quote at /verify.html. We host the same for you: your claims.json, a passport and contract per figure (Claim Contract 1.0, an open specification), receipts, a /verify endpoint under your domain or ours, a machine-readable corrections feed answer engines can poll so a corrected figure retires the day you correct it, and a misquote ledger where readers report what they saw.
Founding price for the first three publishers: $149/month for up to 50 figures, $2 per additional figure per month, setup from your existing pages included. The price after the founding three will be published when the first founding publisher renews, not before. Cancel any time; your claims.json is yours and exportable.
Not SEO, not "AI visibility" ranking, not a promise that engines will cite you more. It is a measurement of what they do with what they already take from you, and a way to make a quoted figure checkable. The instrument, the taxonomy, the specification and the checker are open (CC BY 4.0); you can run them yourself, and we will show you how if you prefer that to paying us.
FinanceRateCalc is a one-person research publisher (Ziya Yetiş, 23 years in bank credit). It takes no money from lenders and sells no leads. All figures it publishes are historical aggregates from public federal records, never predictions about a person and never evidence of misconduct by any institution; audits are of AI systems' restatements, not of the publisher's correctness, which remains the publisher's own.