Training Dashboards

Live views re-publish every ~5 minutes; each page also auto-refreshes itself. Timestamps inside each page show data freshness.

TEACH-2 — 1-word cert bandit (LIVE)

Word ledger: what each word fetched in depth pay vs CE pay; deterministic teacher; distinct-4 GRPO.

ENGLISH-0 — pretraining (LIVE)

From-scratch 0.5B on FineWeb-Edu with the depth gates; sample generations, CE, gate depth.

ENG-2 / SEARCH-1 (other thread)

Gate-movement / cert reward arms on the ENG-1 corpus.

TEACH-1b — frozen record

The length-price run (stopped at c881): collapse, "english" era, free-lecture lock-in.