Historical decision review
Review past lending decisions.
Keep the evidence in your environment.
Explore a working example of a locally operated report-and-recheck workflow. See the result first, then tell us about the review your team needs to do.
Development workflow. Public-data example. Customer availability and terms confirmed separately.
This example uses FL-BSA measurement mode over recorded decisions, a separate development track not part of the current simulation-only release.
What your team supplies
A prepared file containing recorded decisions and comparison groups, the population and period being reviewed, and a clear definition of the outcome being counted.
Preparation and source-selection rules stay with the review. The demonstrated flow uses one cohort column, 2–20 declared groups and at least 30 rows in every group. This floor does not establish statistical adequacy. Borrower data stays in your environment.
What you can review
Group counts and recorded outcome rates, a findings document, a PDF summary and a saved package that can be checked locally using the retained prepared file, exact analysis specification and matching software version.
Source extraction is checked separately. The result describes your supplied records; it does not explain causes or predict future decisions.
Inspect the result
A worked public-data example
This technical example uses public 2024 US mortgage records for Ally Bank. We chose Home Mortgage Disclosure Act (HMDA) data because its public loan-level records include lending outcomes and applicant sex, allowing this example to be checked without customer data. The lender did not participate in or endorse the exercise. It demonstrates a reporting workflow, not a UK/EU consumer-credit benchmark or a fairness verdict.
Reported origination or approval, by recorded applicant sex
Conventional, first-lien home purchase; one-unit, site-built principal residence. The declared selection retained 2,079 of 6,403 public source rows.
| Recorded applicant sex | Selected rows | Originated or approved* | Share |
|---|---|---|---|
| Female | 796 | 705 | 88.57% |
| Male | 1,283 | 1,153 | 89.87% |
| Total | 2,079 | 1,858 | 89.37% |
* Includes reported originations and approvals not accepted. Denials form the rest of this selected population. Withdrawn, incomplete, purchased, preapproval-only and other actions were excluded. Other/missing sex codes were excluded. Public rows are not proven unique applications.
The observed difference is 1.30 percentage points. This is unadjusted: it does not establish discrimination, causation, statistical significance or differences in a wider population.
What was checked?
For this declared example, a separate calculation reproduced the source selection, counts, rates and displayed results across 75 checks. The packaged workflow also created and checked the report offline, and rejected a changed input.
This checks the declared example. It does not establish customer time savings, results for other data or production availability.
Read the report and its context
Download the findings and preparation context first, then open the compact PDF summary.
These are unchanged outputs from the tested example. The report’s “Fairness metric” labels are field names, not fairness judgments. The two downloads are a reading sample, not the complete replay package. The retained prepared file, exact analysis specification, matching software version and full output package are required for replay. Source/exclusion receipts support a separate extraction check. Outputs are unsigned; they are not certificates.
A defined review, operated by your team
We are exploring one defined historical-decision review, operated in your infrastructure, with agreed documentation and support. Scope, availability, software terms and any support fees are confirmed before an evaluation.
This is not a hosted data-analysis service. The workflow measures supplied historical decisions; it does not execute your lending model, make decisions, predict a new policy or certify compliance.
Discuss this workflow
What decisions do you need to review, and who needs to check the result? Please do not send borrower records or other confidential data.
Equilens will reply by email to discuss the task and current availability. Any evaluation, price or delivery commitment is agreed separately.