Which metric to compute. Every field-backed measure is sentinel-guarded: the source encodes unknown values as an out-of-range fill, so values outside the measure's plausible range are dropped before it is aggregated. Ranges: volume 0–100,000,000; avgRate 0–25; avgLoanAmount 0–100,000,000; avgCreditScore 300–850; avgLTV 0–999; avgDTI 0–999; avgIncome 0–100,000,000; avgAppraisedValue 0–100,000,000; avgPurchasePrice 0–100,000,000; avgDtiFront 0–999; avgCombinedLtv 0–999; avgLoanTerm 0–1,200; originatedVolume 0–100,000,000; originatedAvgDti 0–100. units and originatedUnits are document counts and are not guarded. Each response reports how many documents its guard excluded, so a mean taken over 98% of the data can be told apart from one taken over 60%. NOTE on population: units and volume cover loan APPLICATIONS of every outcome — denials, withdrawals and secondary-market purchases included — and so does avgDTI, while originatedUnits, originatedVolume and originatedAvgDti cover originations only. Which metric each bar reports, AND what the bars are ranked by — the list is ordered by the measure you ask for, descending. Defaults to units, so an unqualified "top originators" means the busiest; pass volume for the biggest by dollars, or an average to rank by that average. Two things to know about the averages specifically. Buckets built from fewer than 5 documents are dropped from an average chart entirely, because a mean says nothing about how many transactions produced it and a single-transaction bucket would otherwise outrank real ones; no floor is applied to units or volume, where a lone large transaction is a legitimate top bar. And ranking a bucket list by a sub-aggregated average is approximate in the search engine — each shard contributes its own local top-N before they are merged — so treat the ordering of an average chart as indicative near the boundary rather than exact. Every bucket publishes its own count, so the sample size behind a value is never implicit.
Date window for loans / sales / market endpoints. Accepts every Period value plus rolling windows: last30Days / last60Days / last90Days (today minus 30 / 60 / 90 days through today), previousMonth (the last complete calendar month) and currentMonth (the 1st of this month through today). All windows are resolved in UTC with inclusive calendar-date bounds. Rolling windows are only accepted by the loans, sales and market endpoints (list, count, breakdowns, analytics summary / time-series / chart); originators / companies / branches / agents / offices endpoints return 400 for them.
