This course teaches you how to put a narrow, typed judgment step in front of a decision, using stock research as the running example and Jev, TypeSafe’s typed narrow-judgment API, as the interface. You will work entirely on fictional, deterministic data: a synthetic market of four made-up tickers over 300 trading days, with dated fundamentals invented for teaching. Nothing here describes real companies. In the first half you build a research pipeline you can inspect. You compute relative and signed cumulative volume proxies, compare a clean breakout with a failed one, and prove that event-time features never depend on future bars.
You then implement the CAN SLIM letters as code: quarterly growth, three-year CAGR, consistency checks, leadership plots and market context, and assemble a seven-letter evidence matrix with pass, fail and unknown states. From there you define four typed questions (Choice, Score and Noul outputs), answer them through a mock adapter, route cases by a confidence floor, read reliability diagrams, and generate auditable research notes from a template.
The second half applies the same pattern beyond finance. Interleaved practical chapters cover reading TypeSafe’s public technical claims critically, PII detection with a human-review route, detecting mismatches between confidentiality labels and document bodies, restricting browser automation to pre-approved DOM candidates, and comparing open, locally runnable API-compatible implementations with an independent evaluation plan. A capstone brings everything together in a one-page decision contract with designed counterexamples.
