Leader track ⚖️ · six sessions, no code
For whoever signs the approval. What the risk actually is, why challenge fails in organisations that have all the right job titles, which models earn real review, how to read the report, and when refusing is the correct answer.
What model risk actually is
Risk is about consequences, not accuracy. Four factors set it, and only one of them is about the model. Plus what the April 2026 guidance changed.
🗣️ Leader session 2Effective challenge
Expertise, independence and standing. The third is the one that fails, because the power to stop something is granted rather than appointed.
🔍 Leader session 3What a validator is checking
Conceptual soundness, outcomes analysis and ongoing monitoring, walked through four real findings without a metric being defined.
🗂️ Leader session 4Materiality and the inventory
Which models earn three weeks of review and which earn a line in a register, and why an immaterial model is not an unwatched one.
📄 Leader session 5Reading a validation report
Findings, severity and conditions. What approval with conditions actually buys, and how to tell a control from a sentence.
✍️ Leader session 6Signing it off
The questions to ask before your name goes on it, what to require in writing, vendor models, and when refusing is the correct answer.
Practitioner track 🧪 · ten sessions, hands on
For the people who do the work. The inventory, the design review, the leak hunt, replication, challengers, outcomes analysis, out-of-time testing, monitoring, findings, and a capstone bench that scores your call against what is true.
The inventory and the tiering call
What a register has to hold to support an aggregate view, and how to score one model's materiality without adjectives.
🧱 Practitioner session 2Conceptual soundness
Reading the design before the numbers: purpose, assumptions, data selection, and the quality and extent of the developmental evidence.
🕳️ Practitioner session 3The leak hunt
A field written after the outcome window scores 0.748 on its own. Why a noisy leak survives review, and how to find one.
🔁 Practitioner session 4Replication
Reproducing the number in the document, and what a gap between the document and your re-run actually means.
⚔️ Practitioner session 5Benchmarks and challengers
Three columns reach 0.676 against the champion's 0.682. The benchmark belongs early, because it reframes every other finding.
🎯 Practitioner session 6Outcomes analysis
Predicted against actual, banded. The aggregate gap is 0.006 and the bands run to 0.142, in opposite directions.
📉 Practitioner session 7Out of time and stability
A later vintage, a relationship that moved, and a book that drifted from 19.0 to 34.0 percent thin files.
📡 Practitioner session 8Ongoing monitoring
Thresholds, triggers and what actually fires. Designing a monitoring pack that can fail for the right reason.
📝 Practitioner session 9Writing findings
Severity, conditions, remediation and documentation, including what a model card does and does not settle.
🧪 Practitioner session 10The validation bench
Seven tests against a model whose truth is known by construction. Run them, collect the findings, and make the call.
The whole thing on one page
Every idea in the course and the session that owns it. Click any node to jump there.
Three ways through
Everyone starts at session 1 of their track. These are the shortest useful routes if you cannot take all sixteen.
What is real here
Nothing on the bench is a simulation of a result.
| The applications | 6,000 from a structural model written out in assets/mr-live.js, fixed seeds, every coefficient visible |
| Every model | A real logistic regression fitted in the page by iteratively reweighted least squares, on click |
| Every AUC | A rank-based calculation over the scores those regressions produced. Nothing is hard-coded |
| The leak | Real by construction: the collections field is generated from the outcome, so 0.748 alone is a fact about the data |