Free tool Tier a model against OSFI E-23 in about five minutes — seven risk dimensions, the controls your tier requires, and a report you can print. Open it →
osfi-gate

A walkthrough

Seven views over a year of evidence for Northwind Trust, a bank that does not exist. Its people, repositories and findings are invented. What is real is everything the gate did with them: the scans, the policy decisions, the hashes, the chain and the signatures.

Each of these is a capture of the live demo, which you can open and click through yourself.

Control status, the whole estate

Control status: repositories grouped by team, each row carrying a tick or a cross under each control identifier, with the commit and the age of the evidence on the right.
28 repositories, one column per control. Every cell is a decision a gate run recorded, not a self-assessment. Two repositories are failing; one is stale, meaning its status describes an older commit — which the page says rather than quietly showing green.

A repository that is failing

A repository view: MR-06 listed as not satisfied, an AI and model inventory table, waivers on the default branch, and the list of runs.
MR-06 is not satisfied: a model here does not carry every control its materiality tier requires. Below it, the model inventory read from the repository, any waivers on the default branch, and every run with a link to its evidence.

One run, and where it came from

A run view showing the change metadata, provenance including the policy digest, signature mode, manifest digest and scanner versions, and the verification notes.
The provenance column is the part that matters: the policy digest the decision was made under, the manifest digest, and the exact version of every scanner. "Verified" means the ingester re-ran the hash, signature, chain and replay checks rather than taking the bundle's word for it.

AI call sites, found rather than declared

The AI model registry: five detected call sites with repository, file and line, provider, model identifier, region and inventory ID, above two alerts.
Call sites come from static analysis over the source, then join to the model inventory. The two alerts are the point: a model now running a different version from the one that was validated, and a validation record maintained for code the gate can no longer find. Neither shows up in a spreadsheet somebody keeps by hand.

E-23 materiality, as a record in the repository

An E-23 model assessment: a credit adjudication scorecard rated high materiality, scoring 19 of 21, with its risk dimensions listed against E-23 section references.
Risk dimensions, a computed tier, and the governance controls that tier requires — versioned in Git next to the code. The score and the assessor are recorded, and MR-05 fails the run when an assessment goes stale or nobody is named.

Accepting a finding, with an expiry

The waiver register: four risk acceptances, one expired, one expiring in ten days, and two reported as no longer needed.
Four acceptances, each with a term and a named owner. Two are flagged as no longer needed — the control now passes without them — which is the kind of thing a register kept by hand accumulates silently for years.

Where the console admits what it does not know

The ingester health page: a green banner saying the ingest job is running, and a warning that one repository has not run the gate in 80 days.
The page that matters most for trusting the rest. Note the wording: northwind/etl-jobs has not run the gate in 80 days. Its evidence was checked for new runs and there are none, so this is the repository, not the ingester. A dashboard that cannot tell you its own data is stale is worse than no dashboard, because it looks the same either way.

The interface is translated. The evidence is not.

The same run view with the interface in French — Changement, Vérification, Violations — while the verification notes and violation messages remain in English.
The console is bilingual. Changement, Vérification, Violations and Télécharger le lot de preuves are translated; signature verified (key) with key… and MRM-0204 is high materiality and lacks board_oversight are not. Those are reproduced from the signed bundle, and translating them would change the text of something that was signed. Most products would translate everything and quietly lose the property.

Open the demo Verify a bundle yourself