Textbook/Unit 07

Risk systems & models

Data contracts, decision engines, machine learning, experiments, and governance.

Unit 07 / The big picture

Decide fast. Learn on a different clock.

Two lanes separate live decisions from learning. In the live lane, an event feeds time-correct features, then policy and model, then a durable action. Recorded decisions and effects flow down to observation, evaluation using mature outcomes, and independent validation. A versioned release with rollback returns approved changes to policy and model.

Enlarge to read every label and explore the connections

The live path needs bounded latency and durable actions. The learning path needs mature outcomes, valid evaluation, and controlled releases back into production.

  1. Follow the upper lane from event to durable action.

  2. Trace observed effects into the slower learning lane.

  3. Require validation and rollback before the next release.

Inside this unit

5 chapters
Only use facts known at decision time
7.1 / 14 min read

Risk data contracts and event time

Build features that mean the same thing in analysis and production.

Read chapter
Reserve shared capacity once
7.2 / 15 min read

Decision engines, rules, and reliable execution

Turn evidence into one controlled action under time pressure.

Read chapter
A probability must agree with outcomes
7.3 / 15 min read

Risk models, calibration, and delayed outcomes

Evaluate predictions against the decision they will support.

Read chapter
Compare cohorts at equal maturity
7.4 / 15 min read

Experiments, causal effects, and risk tradeoffs

Measure whether a control improves outcomes rather than only changing them.

Read chapter
An assistant can draft; authority stays explicit
7.5 / 15 min read

Model governance and AI-assisted risk work

Keep model use bounded, reviewable, and accountable.

Read chapter