Pick a deck
One attempt per question — keys 1–4 answer, Enter advances, and SKIP counts as a miss. Code katas allow unlimited runs until the tests pass; prompt katas grade your written prompt against the five anchors (≥ 4/5 clears) and then show the editorial. Answer options are shuffled every run, so read — don't pattern-match. The clock runs for the whole deck, every explanation teaches something the question couldn't, and your best score per deck persists in this browser. Miss something? The end screen lets you re-run just your mistakes. Clear a deck, download the card, post it.
Code katas run on Pyodide + numpy — a real CPython compiled to WebAssembly, lazy-loaded (~10 MB) the first time you hit RUN & GRADE. Your code plus the visible test asserts execute locally; passing means your stdout ends with ALL TESTS PASSED. Prompt-kata deep grades are strictly bring-your-own-key: the request goes from your browser straight to api.anthropic.com, the key lives in this tab's sessionStorage only, and skipping the deep grade costs you nothing — tier-1 grading is fully offline. If a numeric drill stalls you, the relative tolerance is printed under the input — estimate, don't agonize.