Underwriting Coverage,
Measured — Not Claimed

Coverage Tiers — and What "World-Class" Means Here

World-class coverage is not scenario count. No credible buyer benchmarks an underwriting platform by counting demo scenarios; past ~18–20 exemplars a demo bloats without persuading anyone. World-class means four things: (1) a complete, examiner-shaped taxonomy of what "coverage" even is; (2) honest tiering of each cell against evidence; (3) computed numbers that update when the registry updates; and (4) coverage ultimately proven by an evaluation suite (Wave 3), not narrated by a demo. This document is that instrument.

Tier discipline mirrors the Capability Matrix's "Demo-Validated Design" standard from the claims-reconciliation pass: a Demo-Covered cell means the behavior is exercised in the interactive demo on synthetic data — production implementation is validated during the 90-day pilot. Eval-Covered (Wave 3) is the stronger claim: ≥3 golden-file cases per cell executed by the automated evaluation harness on every release. Out-of-scope items are excluded from every denominator so percentages cannot be inflated by scoping.

Coverage by Stage — Computed

Weighted composite across the three dimensions (weights shown below, declared in the registry). The progression is deliberately honest: the eval-stage target is below 100% because disparate-impact analytics remains a dated roadmap item (Phase 2 · Q4 2026) and two cells stay roadmap — a benchmark that reaches 100% before the product does is marketing, not measurement.

Decision-Path Coverage

The paths a credit decision can travel — the dimension buyers probe hardest, because it is where governance lives. The original fourteen scenarios all showed the system succeeding — a systemic gap. Wave 1 (shipped) closed it: S15–S18 show the system encountering trouble — abstaining, holding, escalating, and being overruled — which is what a CCO, an MRM officer, and an examiner actually ask to see.

Product Coverage

Product coverage is claimed per-product, never as "every credit type you underwrite." Three products are declared out of scope with rationale and excluded from all denominators — stating a boundary is a strength in diligence; silence is the only wrong answer. Agricultural production credit shipped in Wave 2 (S20) — it was the highest-frequency community-bank objection.

Regulatory-Event Coverage

The compliance moments a file must survive. Five of the missing events are deterministic policy checks, not scenarios — which is why Wave 2's centerpiece — now shipped — is a single Guardrail Gauntlet scenario (S19) that fires flood determination, legal lending limit, Reg O, HVCRE, and the SCRA screen in one run, wired to the deterministic Policy Guardrail Engine (P0-4, 9 citable rules, 44-test oracle suite) rather than scripted text.

Four Waves to World-Class

Each wave has entry criteria, exit gates, and the audit P0 items it depends on. The sequencing rule: governance paths before products, products before proof, proof before production claims.

What This Benchmark Is Not

This taxonomy was curated by CAIBots from the structure of the OCC Comptroller's Handbook credit-risk booklets and FFIEC examination topics; it is a vendor-authored benchmark, not a certification, and no regulator has endorsed it. During a pilot, the partner institution should map this registry onto its own product set and exam findings and re-weight accordingly — the registry is in this document's source precisely so that is a five-minute edit, after which every number on this page recomputes. Percentages measure breadth of coverage, not depth or accuracy of any cell; depth is Wave 3's job. And a benchmark authored by the builder shares the limitation this platform's own SR 11-7 language now states about itself: independent validation is performed by the institution, not claimed by the vendor.