Loading
Please wait...
CCD2 applies from 20 November 2026. Can you show which policy clause each credit decision rests on?See the 50-decision audit
Loading
Please wait...
FOR EU CONSUMER LENDERS · CCD2 APPLIES 20 NOVEMBER 2026
From 20 November, any customer can ask why an automated credit decision went against them. Aequitas shows, decision by decision, which clause of your credit policy each stated reason rests on, and flags the ones that rest on nothing.
Fixed fee · 10 working days · anonymised data is fine
LENDING AUDIT
A fixed-scope check a head of credit risk can approve without a procurement cycle.
50 recent automated decisions (approvals, declines, limits) and your credit policy. Anonymised extracts are fine. Nothing connects to your live systems.
Each stated reason is matched to the policy clause it relies on. Numeric rules are checked with plain code against the decision’s own data. Your own expert marks agree or disagree on every finding.
A signed report covering all 50 decisions: supported, unsupported, or can’t tell, each with the clause and data it rests on. Ready for CCD2 explanation and review requests.
THE ENGINE BEHIND THE AUDIT
Approve the revised counterparty exposure under current stress assumptions?
Approved with conditions: the liquidity trade-off is documented and bounded.
Toggle the sample: the gate can certify, and it can honestly decline.
HOW TRUST WORKS
Aequitas is built around the opposite commitment. Three rules govern every verdict it certifies:
Confidence comes from checks that actually ran, not from the model’s own say-so.
No positive evidence, no certification. Aequitas would rather tell you it can’t certify than hand you a polished guess. Absence of a red flag is not proof.
Degraded runs read as degraded. Uncalibrated confidence is labeled uncalibrated. Screening tiers are labeled screens. What you see is what actually happened.
HOW IT WORKS
Multi-perspective deliberation is the easy part. Aequitas adds the hard part: a verification layer that decides whether a verdict has earned the right to be certified, and refuses when it hasn’t.
The proposer
A structured multi-perspective debate works the question from many angles, and an independent adversarial audit challenges the draft verdict. This is where the reasoning is built; nothing is certified yet.
The gate
Each claim is checked against the rule or source it relies on, and anything that can’t be confirmed is marked unconfirmed. Verification findings outrank the draft’s own confidence, and an engine-health check refuses to certify anything produced in a degraded state.
The honest verdict
If the bar is met: a verdict with its basis disclosed (exactly what was checked, in plain language). If not: an honest “can’t confirm,” with the limits disclosed instead of smoothed over.
SEE IT IN ACTION
Three sample records across regulated domains: a certified verdict, a certified verdict with the conflict surfaced, and one Aequitas declined to certify.
FINANCIAL RISK
Illustrative sampleShould this counterparty exposure be approved under current stress assumptions?
Approved with conditions: the liquidity trade-off is documented and bounded.
SHA-256 30683768a5ada532…
HIRING & TALENT
Illustrative sampleShould this candidate be advanced past the automated screen for the senior analyst role?
Advanced: the decision rests on job-related criteria, with the trade-off documented.
SHA-256 75f659f95c4a5526…
UNDERWRITING
Illustrative sampleIs this premium adjustment justified by legitimate, non-discriminatory risk factors?
Not certified: one factor could not be shown to be non-discriminatory. Returned for human review with the concern documented, not papered over.
SHA-256 d3cd0a7e9f01a020…
Illustrative examples. Sample hashes are real SHA-256 digests of each sample’s content; check them if you like. Every Aequitas MAX judgment carries its own certificate.
TIERS
Every certified tier runs the same verification floor; the tiers differ in how broad the deliberation is, not in which checks run.
A fast, deterministic screen that flags contradictions and gaps in minutes. Honestly not a certified verdict; it’s a rapid consistency check.
Multi-perspective debate and independent audit, with each claim checked against the rules and sources it relies on, and a verdict with its basis disclosed or an honest “can’t confirm”, plus honest confidence.
Everything in Lens, plus structural analysis of irreconcilable conflicts: it maps why positions can’t all hold at once.
The complete pipeline at maximum breadth, premium model set, strongest evidence. A verdict with its basis disclosed, or an honest “can’t confirm”, backed by the deepest analysis.
BY INDUSTRY
The defensibility layer for AI builders in lending, hiring, and underwriting: an auditable reasoning record built for the scrutiny compliance teams apply.
Credit decision audits for CCD2, with each reason traced to your own policy.
For HR-tech: the bias audit trail and decision record enterprise procurement now demands, in the terms NYC LL144 asks about.
For InsurTech: underwriting AI with the documented reasoning Colorado SB 21-169 and NAIC market-conduct exams ask about.
AN HONEST INSTRUMENT
Aequitas discloses degraded runs and uncalibrated confidence rather than hiding them. A screen is labeled a screen. A verdict discloses exactly what was checked. That honesty is the product.