Parents ask it often: what if one change breaks the whole project? In our Markham classroom the honest answer is simple — change only one place and name what moved. Changing a big chunk is not the same as debugging. Four questions on paper beat rewriting the whole script.

A real classroom case: Scratch chase-score speed bump

Last month at afterschool in Markham a Scratch chase-score game was mid-build. A classmate raised the speed. Suddenly the score and the touching-color rule both broke. We paused and wrote four questions on paper before anyone rewrote the script.

Our 2026-09-27 Xiaohongshu note used this exact classroom beat: change only one place, name what moved, then try again.

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 what moved. “Change only the speed, and say what just moved” is a different job from “make any chase game look finished.” If a kid (or an AI prompt) muddles those, the next edit that comes back solves the wrong problem.
2. The useful state is tiny on purpose. You only hold the speed change against the score rule and the touching-color block — not the whole chase game. That is abstraction: keep what the next check needs, drop the rest.
3. The four questions and edge cases are the algorithm. Change one place → name what moved → predict → play → handle double edits, green-flag spam, delete-restore blocks, and a full script rewrite. The same one-change habit transfers to a Lego tower that collapses when one brick is added, a recipe that goes salty when one spoon is off, or a backpack zipper that sticks when one book is shoved in last. 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 “only move one place” stays readable; JavaScript when the same idea becomes something clickable on a screen; C++ later when speed and structure matter. Tools change; change one place → name what moved → try again does not.

So AI’s ability to type code does not cancel a thinking-first start. It raises the value of kids who can change one place, name what moved, and handle the edge cases before they ask a model — or a classmate — to rewrite the script.

Three steps kids can name

We practice the same shape on the bench:

1 · Change one place — only the speed (or only that one brick / spoon / book)

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

3 · Name what moved — write it on paper, then try again

• Decomposition — Goal / Remember / Done when / Edge
• Abstraction — hold only the speed change against score and touching-color (not the whole chase game)
• Algorithms — change one place → name what moved → predict → play → handle edges
• Pattern recognition — same “one change” on Lego tower / recipe salt / backpack zipper / contest-style work

Follow CodeSky

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

Sources: CodeSky Xiaohongshu note pack 2026-09-27-孩子改一步就全乱了怎么办 (Scratch chase-score speed bump case; Python → JavaScript → C++ pathway; contest-style disclaimer).