Reading code, just enough

Module 3.5 teaches the pattern-recognition floor a non-coder needs to recognize when an AI agent has gone wrong — without writing code from scratch. Four lessons, each strictly observational (per D-33): reading a file tree, spotting when the AI edited the wrong file, reading an error message back to a file pointer, and recognizing when 'use client' is missing. The floor is detection. Each lesson stops at "you can spot this"; explaining from first principles is out of scope.

Loop checks across the four lessons most often name evaluate — detection is the heart of the evaluate step (planner's judgment per RESEARCH.md Pattern 5).

What this module builds

By the end of this module, you can look at the code surface an AI agent puts in front of you — the file tree, its diff summary, an error message, a file that won't respond — and spot when something is off, without reading or writing a line of code yourself.

Each lesson builds on the last:

• Lesson 1 — Reading a file tree: open a Next.js project's file tree, name what each top-level folder is for at a glance, and spot one shape smell (a pages/ folder and an app/ folder both present) → sets up Lesson 2 by giving you the project shape that tells you where an edit should have landed.
• Lesson 2 — Spotting wrong-file edits: read the agent's diff summary against the intent of your ask and judge right-file vs wrong-file in seconds → sets up Lesson 3 by leaving you with the case where the file looked right but the page now shows a red error.
• Lesson 3 — Error message to file pointer: scan an error message and find the first line that names one of YOUR files, then paste the whole error back with that pointer named → sets up Lesson 4 by surfacing one error word — hydration — that points at a specific fixable symptom.
• Lesson 4 — 'use client' and the server/client split: recognize whether a file needs 'use client', name the symptom of a missing one (an interactive element that silently does nothing, or a hydration error), and steer the agent to the fix → sets up Phase 3, where you use this observation floor on the thread project every day.

The thread that ties it together: Module 3.5 is the OBSERVATION floor. In every lesson you observe, recognize, and steer — you spot the smell, name the file, paste the error, point at the symptom. The agent reads the code, decodes the error, and makes the fix. Every code term here ('use client', hydration) is a symptom you scan for, never a mechanism you have to understand.

Reference scaffold

All four M3.5 lessons read from a tiny pre-baked Next.js sample-app at sample-app/ (per D-34). The scaffold is REFERENCE ONLY — you do not run it, install it, or modify it. The scaffold ships in Plan 02-02; Plan 02-13 onward authors lessons against it.