> ## Documentation Index
> Fetch the complete documentation index at: https://docs.amlok.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Bank benchmarks

> Compare bank portfolios and evaluate AMLOK across consistent products, vintages and evidence populations.

A bank is an institution context, not one lending product. Evaluate consumer and business portfolios separately before producing an institution-wide summary. Keep bank-level reporting distinct from borrower-level underwriting.

## Financial-risk measures

| Measure | Definition and comparison rule |
| - | - |
| Portfolio delinquency | Past-due accounts or balances divided by the matching eligible portfolio, segmented by product and vintage. |
| Impairment and allowance | Preserve the institution's reporting definitions, accounting basis and stage classifications. |
| Realized credit loss | Net charge-offs or realized losses against a stated average exposure or origination cohort. |
| Concentration | Exposure by obligor group, sector, geography and collateral type, without double-counting related facilities. |
| Operational quality | Evidence completeness, analyst correction rates, processing time and unresolved exceptions. |

## Public benchmark sources

[OSFI's financial data guidance](https://www.osfi-bsif.gc.ca/en/data-forms/open-government-osfi-financial-data/working-open-government-data) identifies bank balance-sheet, income and capital returns. Use the relevant return dictionary and reporting perimeter when building peer comparisons. Reconcile mergers and reporting changes before adding institution totals.

These returns provide institutional context. They do not reveal the application snapshots and later outcomes needed to evaluate borrower-level models. Use product-specific [consumer](https://docs.amlok.io/research/consumer-underwriting.md) and [business](https://docs.amlok.io/research/business-underwriting.md) evidence for those tasks.

## AMLOK evaluation

Construct development, validation and later-period holdouts with entity-group separation. Where data rights permit, test on an institution excluded from development to measure transferability. Report both within-bank and held-out-bank results; pooling them hides deployment differences.

Measure calibration, decision-relevant error and evidence reliability by product and cohort. Operational evaluation should include analyst review, correction and abstention, not only inference time. An institution-wide average must disclose product weights and coverage.

For Signal, evaluate risk changes against a frozen prior baseline and report false alerts per monitored borrower-month. The bank retains decisions, policy thresholds and intervention. Apply the [governance framework](https://docs.amlok.io/research/robustness.md) and [evaluation protocol](https://docs.amlok.io/research/evaluation.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.