Where does this data actually come from?

The fields we read

These are groupings of reported line items— the raw fields the metrics are computed from. Mnemonics are the regulators' own, so any figure can be traced back to the schedule it came from.

They are not the same thing as the platform's analytical domains, of which there are 8. Those are how the institution view organises what it reports; the groups below are what it reads. Each has its own read, its own provenance block and its own section on the institution view, and is named here beside the route segment it is served under:

The rules engine

The Audit Engine is deterministic: 50 rules per institution, of which 36 fire from the filed figures alone and 14 are attestation rules answered by a named human. Every rule maps to a published citation — FIL-104-2006 / SR 07-01 (71 FR 74580) for CRE concentration, 12 CFR 324 for capital adequacy, and so on. No rule is AI-driven; each is a threshold comparison you can check by hand.

Rule behaviour has been backtested over ten years of historical filings — 197,400 institution-quarters, with 20 rules ported into the study and scored — measuring hit rate, lift over the base rate, and lead time.

The outcome measured was bank failure, taken from the FDIC's published failed-bank list, and nothing else. It is not an enforcement study and the lift figures are not enforcement correlations. Failure is a real and severe outcome, but it is also a rare and lagging one: most of what an examiner finds never becomes a public event of any kind, so a hit count is a floor rather than a truth, and a rule that passed should be read as “trust this one more than that one” rather than “this rule is right”. Where a rule performed poorly, that is recorded rather than hidden.

How position is reported

There is no single 0–100 composite. The regulator issues no readiness score, so MDRM IQ reports the deterministic facts — which signals are in breach, approaching, drifting, passed or not evaluated; where the institution sits against its published peer group; and which way the series is moving — and leaves the judgement to you.

AI-generated content

MDRM IQ uses Anthropic's Claude to write briefs, narratives and news analysis. Those outputs are labelled and are informational only: they synthesise figures already computed and shown elsewhere on the page. Generated text may not introduce a number that is not in its input — the report path refuses to generate rather than fill a gap with a plausible figure.

Regulatory news

Agency feeds are polled hourly and tagged by source: the Federal Reserve, FDIC, OCC, CFPB, FinCEN and NCUA, plus industry publications. The public enforcement register is narrower on purpose and states its own coverage on the page: it holds the actions the OCC and the Federal Reserve publish in machine-readable form. The FDIC publishes no such register, so for an FDIC-supervised bank an empty result is not evidence of a clean record — and the page says so rather than letting silence read as an answer.