Operators, product owners, and people affected by the data & features decision.
Data & feature foundations
Fraud labels arrive late and online features disagree with training data.
A measurable problem, before a model.
semantic contracts, event time, availability time, as-of joins with version, owner, event time, and availability contracts.
Fraud labels arrive late and online features disagree with training data. Every solution must state latency, cost, capacity, and fallback limits.
Offline evidence plus a deployable resolution: Data contracts, point-in-time features, backfills, quality gates, and lineage.
Frame and baseline
Turn the production problem into explicit data, metric, and baseline contracts.
- 01↗
Data contracts, ownership, and lineage
A trustworthy dataset explains meaning, accountability, and ancestry—not just columns.
65–90 min · checkpoint · deterministic lab - 02↗
Point-in-time-correct features
Point-in-time correctness preserves the knowledge boundary of the original decision.
65–90 min · checkpoint · deterministic lab
Build and stress
Formalize the machinery, test counterexamples, and expose system limits.
- 03↗
Feature transformations that survive production
Ship the fitted representation contract with the model, in the same order and with the same edge cases.
65–90 min · checkpoint · deterministic lab - 04↗
Batch, stream, and training-serving parity
Parity starts with one temporal meaning, then proves every execution path implements it.
65–90 min · checkpoint · deterministic lab
Resolve and operate
Join model behavior to architecture, observability, rollout, and rollback.
- 05↗
Labels, snapshots, and backfills
A label snapshot records what was knowable by a date, not the database’s latest opinion.
65–90 min · checkpoint · deterministic lab - HW↗
Point-in-time feature system
Design a versioned training set and online lookup from transaction, account, and outcome events.
Rubric · staged hints · reference resolution
2 concepts make this faster.
Prerequisites are guidance, never hard gates.