AssureCLA · Conformance twin
Every RBI co-lending control, independently recomputed — and proven against an adversarial twin
A control is only trustworthy if it catches the thing it’s meant to. Our conformance twin exercises all 24 controls — each with a compliant pass and a violation that must be caught — on a deterministic, replayable synthetic book. The run is seeded and wall-clock-free, so anyone can reproduce it byte-for-byte. If the engine ever disagreed with the oracle, the run would fail.
24/24
controls proven
104
full-overlay runs
PASS
as of 2026-08-29
182
aged cohorts (90-day book)
How it’s proven
Not a checklist — an adversarial twin
Compliant + a violation, per control
The twin generates a clean book (all 24 controls GREEN) and, for each control, a book engineered to breach exactly that control — then checks the engine's verdict against an oracle authored separately from the engine. The twin is ours and the data is synthetic — this is adversarial self-testing, not third-party attestation.
Fail-closed aggregation
A non-maskable breach can only go RED (13 of 24 controls). A maskable one degrades to AMBER. Missing data is UNKNOWN — never a silent pass. GREEN can never hide a FAIL.
Deterministic & replayable
Every run is seeded, so any failing day replays byte-for-byte. Recomputed independently from each RE's own records — which is what keeps it usable when partners or inspectors disagree.
Control coverage
24 controls · each caught by a named violation
Every control is recomputed against a compliant book and a book engineered to breach exactly that control. Coverage spans the RBI (Co-Lending Arrangements) Directions, 2025 and the RBI (Priority Sector Lending) Directions, 2025 — the full text of both is public; which paragraphs we enforce, how each is evaluated, and the violation scenarios that trip them are part of our rule packs and are not published here.
2
critical controls
17
high-severity controls
13
non-maskable (can only go red)
2
RBI instruments enforced
The control-by-control matrix — paragraph mapping, severity and maskability per control, and the named violation scenario each is proven against — is available to assessors, auditors and partner compliance teams under access, together with the machine-readable proof. Request assessor access.
The aging book
Proof that persists — a 90-day rolling record
A one-day green run is a snapshot. The twin keeps a persistent, aging book of cohorts across a rolling window (2026-05-30 … 2026-08-28) — 182 cohorts, all conformant — so conformance is a track record, not a moment.
182
cohorts in the book
60
< 30 days
120
30–90 days
2
90 days +
What this proves — and what it doesn’t
This proves the control logic is correct and fail-closed against adversarial inputs — recomputed independently, replayable, honest about UNKNOWN. It runs on a synthetic booktoday; on a live arrangement the same engine runs on each RE’s own records. AssureLocker is a Technology Service Provider: we surface signals; the lender decides and funds. We never lend, price, decide or hold funds.