Reports now has an OSFI E-23 readiness view that shows, principle by principle, whether the records Guideline E-23 asks about exist in PulseAI, with an Appendix 1 table for every model. Six new health checks on the model record feed it, and every place PulseAI cites E-23 now uses the headings of the guideline itself.
OSFI Guideline E-23 takes effect on May 1, 2027. This release lays your PulseAI records against it: a view on Reports that goes principle by principle, the checks on the model record that feed it, and OSFI’s own wording wherever PulseAI cites the guideline. Two smaller changes make attestation cycles and KPI period close easier to finish.
Reports has a new E-23 readiness tab. It takes each of E-23’s twelve principles, grouped under the guideline’s three outcomes, and shows whether the records that principle depends on exist in PulseAI. Every row carries one of five states: evidence present, partial, no evidence recorded, performed outside PulseAI, or could not be determined.
Each row gives the reason in words, the record counts behind it (for example, how many of your active models have an owner recorded), what remains yours to do, and a link to the screen where a missing record is made. A principle reaches evidence present only when at least one record exists, so an empty model registry shows as the largest gap, not a clean sheet. A record that cannot be read is shown as could not be determined and is never counted either way.
Four further rows name the work E-23 expects that happens in your own processes and tools: data quality and bias testing, conceptual soundness and explainability, production testing and change control, and performance and drift measurement. PulseAI does not do that work. Each row says, as a separate fact, whether results from it have been filed in PulseAI.
The tab counts down the days to May 1, 2027, and is computed from your current records each time you open it. The weekly health report sealed in the Evidence Vault is the permanent record of the same checks.
The same tab lists every registered model against E-23 Appendix 1: the six minimum attributes OSFI expects for every model, and the eleven additional ones. Each set shows how many are captured, such as 5 of 6, and names the ones that are missing.
Captured means a value is recorded, not that the information is adequate. A risk rating counts only when the assessment behind it is recorded, and approved uses are the use cases linked to the model. Every registered model is treated as in scope for the additional attributes, because a negligible-risk exemption cannot be recorded in PulseAI yet. Retired models stay in the list, as E-23 expects decommissioned models to stay in the inventory, but are left out of the totals.
There is no total, no percentage and no “N of 12” anywhere on the view. A single number is the one a reader would quote to a board, and a count of records cannot honestly produce it.
The view reports whether a record exists, and says so at the top: it does not judge whether your framework is adequate or proportionate, and it is not a compliance assessment. It also states how many models it counted and that models held in other inventories are not represented. Principle 3.4, model review, never shows evidence present: a review record says whether the reviewer owns the model, not whether they were independent of its development, which is the question E-23 asks.
The Health tab has six new checks over the model record. Their findings appear there, feed the readiness view, and are included in the weekly health report:
Health findings, evidence records, the readiness view and the AI Impact Assessment wizard each named some E-23 principles in their own words. They now cite each principle by the heading of the guideline section that holds it, such as “3.4 (model review)”, so the same principle reads the same way on every screen. Evidence sealed before this release keeps the words it was sealed with.
Admins can now close an attestation cycle from its row on the Attestations page; until now that took an API call. Closing asks for confirmation first, because it is one-way. The cycle keeps its certified, pending and overdue counts, and any record still pending stays pending and counts as outstanding.
A KPI with measurements but no verdict blocks period close. The period close page now links each blocking KPI straight to its Close KPI form, offers a way back to period close, and keeps the retrospective you have typed so far.
At a glance
partnerships@pulseai.now