Read the snippet → tap the answer. No typing, no code, no memorizing syntax.
Every recipe was live-tested on the real contest bot.
Yes — because of what the exam actually grades. You are never asked to write code. The notebook comes pre-written, the bot writes every change. People fail this contest not from lack of coding — they fail by asking the bot vague things ("improve this code") and accepting broken answers. Recognizing the task type and pointing at the right cell is a pattern-matching skill, and pattern-matching is learnable by anyone in days. That's this quiz.
You said: focus on the preprocessing and model cells only. Correct — of the 18 task patterns here, 15 are won inside exactly those two cells. The quiz tells you which cell on every answer. The other 3 are "process" rules (like: submit the baseline first) that cost you M1 attempt 1.
1. Tap an answer — guessing is fine, wrong guesses teach hardest (that's tested science: the "testing effect", Rowland 2014). Expect to miss most on round 1 — that's the method loading, not you being dumb.
2. Missed cards come back until you get them.
3. Do short rounds, spaced out — spacing beats one giant session by ~30–40% (Cepeda 2008).
| When | What |
|---|---|
| Now | Round 1: Spot the win (~15 min). Miss freely. |
| Tonight before bed | Round 2: Everything. Sleep locks it in. |
| Aug 3, before 07:30 practice | Round 3: Reflexes only (~5 min) |
| Aug 3 evening | Round 4: Everything. Target 80%+ |
| Aug 4 & 6 morning | 5-min warm-up pass |
1. Skip the story. Find the scoring cell and the baseline's "this simple approach doesn't…" confession.
2. Spot the trigger words → you'll recognize the pattern from this quiz.
3. The answer card told you which cell and which magic words — your prompt is now targeted, not "make this better" (which is what killed M1 attempt 1).
Recipes: ioai-prep prompt pack v3 + Day-2 wildcards, live-tested on the Gemma 4 E4B contest bot 2026-08-02. Science: Rowland 2014, Cepeda 2008, Kornell 2009.