From vague to prescriptive
The 2021 version opened with the CECL origin story and folded data review into broader process testing. The 2026 version drops the history, cites revised model risk management guidance, and adds a standalone data controls objective. One of the changes that matters most: examiners now have explicit if/then conclusions directing findings and escalation.
Where institutions are getting caught
Policy that doesn't match practice. The most common finding is also the most avoidable: practices such as annual reviews or evaluations without an action item associated with them. Effective challenge of model assumptions must take place in line with what the policy outlines.
Q-factors without a traceable why. Qualitative factors are drawing the heaviest scrutiny, largely because loss history is thin and Q-factors may carry a material share of the reserve. Scorecard adjustments moving each quarter while the supporting narrative sits unchanged. Both magnitude and direction need a documented reason. A factor that hasn't moved in two years will prompt a question about why it's still a factor at all.
Validation findings that never close. Institutions acknowledge findings, commit to a fix, then have nothing to show six months later.
Data lineage. Can you prove the balances on your core match what comes out of your CECL model? Englert's challenge: that reconciliation should take 15 minutes to an hour, not multiple days, and it belongs in your model validation scope.
Forecasting and validation
Two framing points worth taking to your board. First, the reasonable and supportable period is a methodology assumption, not an accounting policy election, so it needs the same governance and periodic re-evaluation as any other input. Second, multiple economic scenarios are allowed but never required. A single well-supported scenario works.
On backtesting, the question examiners are asking is not how wide the gap is between projected losses and actual charge-offs, but whether that difference is consistent. Conservative estimates rarely draw complaints. What draws attention is a model that doesn't move when credit quality does.
The takeaway
Most of this checklist is low effort and high value: reconcile your data, document your effective challenge, tie Q-factor movement to observable indicators, and close your validation findings. Scale the depth to your size and complexity. Then put it in writing.
Questions about your CECL readiness? Reach Zach Englert at Zach.Englert@EmpyreanSolutions.com.
Read more about CECL on our website https://hubs.la/Q04w6J530