Samples
One example report for each type of model, as a real PDF you can open or download.
SR 26-2 model validation memo
Corporate PD Model v3.2. 4,000 corporate obligors, replicated from the declared specification and stressed across four illustrative scenarios.
May be relied upon, with conditions
12 pages · 12 sections · 3 finding(s) · 26 named as not performed · 392 KB
memo 3606d16b231438c9
SR 26-2 model validation memo
Application Scorecard Challenger v4. A scored application book with outcomes, score bands, segments and a stability window.
May be relied upon, with conditions
10 pages · 12 sections · 2 finding(s) · 19 named as not performed · 367 KB
memo c71ca71ff0e5bfe3
EU AI Act high-risk system dossier
Application Scorecard Challenger v4. The same scored book as the ML sample, under a declared EU supervisor — so the same evidence becomes the Annex IV dossier.
May be relied upon, with conditions
9 pages · 10 sections · 2 finding(s) · 19 named as not performed · 364 KB
memo fde7c63518038aed
ECB TRIM / CRR IRB validation
Corporate PD Rating System v3.2. A graded corporate population read against its declared master rating scale, grade by grade.
May be relied upon, with conditions
7 pages · 10 sections · 1 finding(s) · 15 named as not performed · 351 KB
memo dc733e0d7676b2bf
Fraud model validation, with the PSD2 reference rate
Card Fraud Scorer v7.1. 20,000 scored card transactions with matured fraud labels.
Inconclusive — the evidence is insufficient
6 pages · 10 sections · 0 finding(s) · 9 named as not performed · 333 KB
memo d5ea6546ec78795a
Transaction monitoring scenario validation
Transaction Monitoring Scenario Suite v6. 6,580 alerts across five scenarios, with a reviewed below-the-line sample.
Inconclusive — the evidence is insufficient
7 pages · 9 sections · 0 finding(s) · 7 named as not performed · 331 KB
memo 44c99e7131e25e13
FRTB back-testing and P&L attribution
Desk A VaR Model v2.4. A desk's daily VaR forecasts against actual and hypothetical P&L.
May be relied upon, with conditions
8 pages · 10 sections · 2 finding(s) · 6 named as not performed · 367 KB
memo ad9a53b22715de25
NIST AI RMF evaluation
Policy Assistant 2.1. Assistant traces with judged claims, planted contradictions and a judge measured against an anchor set.
May be relied upon, with conditions
13 pages · 8 sections · 4 finding(s) · 23 named as not performed · 400 KB
memo c1035073cf033d99
GenAI system assurance pack
Policy Assistant 2.1. The same assistant drawn ten times per prompt under a pinned system record, scored against the bank's expert benchmark.
May be relied upon, with conditions
10 pages · 8 sections · 4 finding(s) · 10 named as not performed · 379 KB
memo f55860189cf688c4
Actuarial model validation
Motor & Property Severity GBM v6. Closed motor and property claims, predicted against settled cost, by band and region.
Inconclusive — the evidence is insufficient
8 pages · 7 sections · 0 finding(s) · 8 named as not performed · 350 KB
memo 618e183827bb24a4
Interest rate risk in the banking book
Behavioural NMD & Prepayment Suite v4. Deposit and mortgage cohorts, modelled against realised balances over 36 months.
May be relied upon, with conditions
8 pages · 12 sections · 3 finding(s) · 4 named as not performed · 348 KB
memo 2c21e071f5549923
ICAAP / ILAAP assumption review
ILAAP Behavioural Assumption Set 2026. Funding by counterparty class, assumed outflow and rollover rates against realised behaviour.
May be relied upon, with conditions
7 pages · 12 sections · 2 finding(s) · 4 named as not performed · 346 KB
memo 79ebd982ca9ba647
Prepared 2026-10-07. Each card shows a fingerprint (hash) tying it to the exact report it came from, so a sample can't quietly go out of date. A sample report is a recommendation based on made-up data — it is never a certificate.
