Functional check

Qwen browser execution check

The saved four-message run for caching and batching.

This saved browser check used four real messages, with two passes per path. Every path got only 1 of 4 correct. It checks that the execution paths run; it does not establish useful moderation accuracy.

All 24 Qwen layers ran. No layers were skipped. Loading and warmup are excluded; instruction-cache creation and copying are included in cached-path time.

Qwen 0.5B q4, one WASM thread, 64 message tokens maximum. This differs from the trained float32 Python classifiers. Four messages are too few for a general speed or quality claim.