Part 9 of 9 · Use-case explorer

Find a small feature worth trying

Start with a behaviour your users need. Explore sixteen feature sketches, find the judgment inside each one, and choose a small experiment you can evaluate.

Design sketches, each with the closest official cookbook linked.

Lab

Explorer

Pick a goal, a surface, or both. Cards show the primitives as pills, link to the pattern that composes them, and name the closest official cookbook.
Goal
Surface
Question sketches are abbreviated: real criteria carry a description per option or level, an escape option, and the field paths the instruction points at. Part 5 covers how to write them.

From the behaviour you want to the judgments you need

Write down the change a user should see: a better ordered list, a relevant label, a filled field, or a case sent to someone who can help. Then work backward to the evidence and judgment it requires.

  1. Show. Ranking, sorting, highlighting. You need a comparable number per item: a Noul per item against the same question, or a Score on a shared rubric. Code sorts. Pattern: composite scoring when several dimensions matter.
  2. Select. Routing, dispatching, picking one of many. You need a winner from a closed set: a Choice with an escape option, plus branch-specific questions asked in the same request. Pattern: speculative fan-out.
  3. Change. Extracting, normalising, filling fields. You need the model to pick which existing span or allowed value applies, never to produce a value. Code finds candidates; a Choice selects. Pattern: select, not generate.
  4. Hand off. Verifying, escalating, gating. You need a probability that something is wrong, per checkable unit, and a threshold per consequence. Nouls, aggregated in code, feeding a review band. Pattern: confidence-gated cascade.

Before sending a request, check whether a rule or lookup already answers it. For the judgments that remain, choose an observable success criterion: fewer wrong routes, fewer missed cases, or less review time.

Where to go next

Reference

Official docs

The API contract, SDK signatures, primitive pages, and the current model facts. The documentation index lists the available topics.

 Ideas

Use-case map and cookbooks

TypeSafe's own map by industry, and the cookbooks each card above links to, with real datasets, run dates, and costs.

 Try it

Playground

Paste a state and a few questions and read the distributions before you write any code. API keys live in the same console.

Pick one feature and a handful of representative examples. Compare the proposed result with what a person would expect, then expand the evaluation before connecting it to real actions. The request builder is a place to draft the first version.