About
Compliance as code
Infrastructure stopped being a thing people configured by hand, and became a thing people declare, review and version. Security followed. Compliance mostly did not. It is still a quarterly round of assertions, kept in documents, evidenced by screenshots, and trusted because of who signed it.
osfi-gate treats a control the way the industry already treats infrastructure. The rules that decide whether a change is acceptable are code, in a repository your second line owns. They are reviewed in a pull request, versioned, and identified by commit. They run automatically on every change, and what they produce is not a status in a dashboard but a signed artefact that records the decision and everything needed to reach it again.
That produces three things a document cannot:
- A policy with a history. You can see what the rule was on the day a decision was made, not only what it is now. Every bundle names the commit of the policy that applied.
- A decision that can be re-run. Not "we attest this passed" but "here are the inputs, here is the rule, run it yourself and see." Offline, with nothing of ours running.
- A gap that surfaces early. A control the gate could not evaluate is reported as unknown, never as passing, so it shows up in your own console rather than in somebody else's finding.
Why it exists
Canadian federally regulated institutions have no shortage of control frameworks. What they tend to lack is evidence — the kind somebody outside the team can check without taking the team's word for it.
The usual arrangement is a control attested quarterly by the people who operate it, recorded in a platform that is itself trusted, and reviewed by someone with no practical way to reproduce the finding. Every step is reasonable. The result is a chain of assertions with no point at which anyone can independently confirm anything.
This takes the opposite position: the evidence should survive not being trusted. If the tool that produced it disappeared tomorrow, the evidence would still check out.
That constraint is what shaped the design, and it is also what rules things out. It is why approvals happen in Git rather than in a web form, why the tool refuses to report "unknown" as "passing", and why there is no hosted service that would require sending you evidence you were told you did not have to trust us about.
Who builds it
Blake — one person, not a team. Twenty-five years leading enterprise technology teams, working as a digital transformation and AI architect, now building this.
What that means for you is on the front page under Where this is, plainly: what is finished, what is not, and what nobody can tell you yet.
osfi@melx.buzz Connect on LinkedIn
What is on this site
- The live demo — a console over a year of evidence for a fictional Canadian bank. Twenty-eight repositories, 1,418 gate runs, and a short guided tour. Nothing in it is real, and it says so on every page.
- Verify it yourself — what verification actually checks, and what it does not.
- Controls — the 24 controls in the reference policy, what blocks and what only warns.
- Free tools — the E-23 model risk tool, which runs entirely in your browser.
- The roadmap — what has shipped, by milestone, and what has not.