Record how to find out whether a tutorial teaches

docs_examples proves every listing in docs/ compiles and still matches the file
it was quoted from. It cannot prove a chapter teaches anything -- a document can
be made entirely of verified excerpts and still be unfollowable, because what a
reader needs is the glue between them.

So AGENTS.md gains "Testing a Tutorial": hand the chapter to a subagent on a
weaker model with the library, its headers and the art but not the finished
program, and make them build and run the game. The weaker model is the point; it
will not paper over a gap with knowledge the document did not give it.

The section carries the rules that make the result mean something -- copy the
tree with examples/ and docs/ removed rather than asking them not to look, let
them read the public headers because a real reader has them, do not make them
hand-author a .tmj, give them a screenshot helper, and verify their binary
yourself -- plus the sandbox recipe, and the warning to prove the build path
works before handing it over.

It also records how to read the report, because a weaker model states its own
mistakes as documentation defects with total confidence. Reproduce before
believing; and a rejected finding usually still leaves something behind.

The evidence for doing any of this is in the table at the end: four read-only
reviews of chapters 20 and 21 found real gaps and missed all three of the
defects the first build-and-run found, including a published example output the
chapter could not produce.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KzBDV2fqgnUAcqCKqKvc71

docs/MAINTENANCE.md

@@ -4,6 +4,13 @@ This file documents the harness that keeps `docs/` honest. It is not a chapter
 a reader learning libakgl never needs it — and it is checked by the same harness
 it describes, so the file that publishes the convention is held to it.
 
+**The harness checks that every listing is true. It cannot check that a chapter
+teaches.** For that, the tutorials get a second test: the chapter is handed to a
+subagent on a weaker model with the library, the headers and the art but *not*
+the finished program, and that reader has to build and run the game. See
+`AGENTS.md`, "Testing a Tutorial". Four read-only reviews of chapters 20 and 21
+missed three defects the first build-and-run found in minutes.
+
 ## Documentation examples
 
 **Every fenced block in `docs/*.md` is compiled, linked, run, or matched against