Preregistered.
Then reported.
Gates are written down before the run, not after the numbers land. A failed gate is a result, not a draft.
Results worth
checking.
We publish what we learn, including the work that did not succeed. Every paper below ships with its full experimental record — gate verdicts, per-step metrics, and the preregistration written before the run.
Browse the record on GitHub ↗Papers.
How we publish.
Preregistered.
Negative results,
published in full.
The architecture below did not work. The record of why is more useful to the field than silence, so we published all of it.