Parents ask it often: what if kids hand every bug to someone else? In our Markham classroom the honest answer is simple — name the check they skipped before asking. Owning the first look is not the same as knowing the fix. Four questions on paper beat handing the mouse away.

A real classroom case: Scratch maze sprite stuck in a wall

Last month at afterschool in Markham a Scratch maze froze mid-run. The sprite sat in a wall. Someone waited for help instead of reading which costume, direction, or if-touching block had changed. We paused and wrote four questions on paper before anyone took the mouse.

Our 2026-09-26 Xiaohongshu note used this exact classroom beat: name the skipped check, then try once before handing the mouse away.

Why this still matters when AI can write code

Claims below are grounded in that teaching note — not an external product case with published accuracy metrics.

1. AI does not remove the need to name the skipped check. “Get the sprite past the wall yourself once, and say which check you skipped” is a different job from “make any maze look finished.” If a kid (or an AI prompt) muddles those, the next fix that comes back solves the wrong problem.
2. The useful state is tiny on purpose. You only hold costume, direction, and if-touching on the active sprite — not the whole maze. That is abstraction: keep what the next check needs, drop the rest.
3. The four questions and edge cases are the algorithm. Say the skipped check → change one thing → predict → play → handle unread script, lookalike sprites, sensor on the wrong sprite, and a classmate grabbing the mouse. The same first-look check transfers to a bike chain that came off, wifi router lights, or a missing worksheet against a backpack list. Contest-style questions inspired by CCC Junior, CCC, BCC, and school contests reuse those moves — kids who want to sit those register on their own; CodeSky is not an official contest program and is not affiliated with Waterloo or CEMC.
4. Languages are tools on a path, not the answer. Early work often uses Python so the checks stay readable; JavaScript when the same idea becomes something clickable on a screen; C++ later when speed and structure matter. Tools change; name the skipped check → try once before handing the mouse does not.

So AI’s ability to type code does not cancel a thinking-first start. It raises the value of kids who can name the skipped check, change one thing with a prediction, and handle the edge cases before they ask a model — or a classmate — to fix it.

Three steps kids can name

We practice the same shape on the bench:

1 · Name — the check you skipped

2 · Four questions — Goal / Remember / Done when / Edge

3 · Try once — before handing the mouse

• Decomposition — Goal / Remember / Done when / Edge
• Abstraction — hold only costume, direction, touching color (not the whole maze)
• Algorithms — say skipped check → change one thing → predict → play → handle edges
• Pattern recognition — same “look first” on bike chain / router lights / backpack list / contest-style work

Follow CodeSky

Instagram @codeskyca  ·  小红书 / RedNote  ·  WeChat 公众号: search codeskyca

Sources: CodeSky Xiaohongshu note pack 2026-09-26-孩子一卡住就甩给别人改怎么办 (Scratch maze sprite-in-wall case; Python → JavaScript → C++ pathway; contest-style disclaimer).