The bench · integrity
The records have to mean something.
Codeseum is a public ladder. A public ladder only works if every name on it is honest. Tape, replay, and signed findings keep the result attached to real work.
Tape required
Every record claim is on tape. No tape, no claim — and no exceptions, even for the leader. The screen, the camera, the keystrokes, all on the record. Tapes go public immediately; the 48h seat-lock protects the run, not the footage.
Replay verification
Every submitted solution is re-run deterministically against the trial's hidden test set. A run that doesn't reproduce its own MARK on replay is voided automatically before it can touch the ladder.
Human review
Judges review flagged tapes, record claims, and high-impact disputes. Findings are signed, visible, and separated from automated scoring.
What's allowed
Any approved engine. Multiple monitors. Notes. Music. Coffee. Friends in the room.
What's required
Continuous tape from preflight to verdict. Visible face cam. Visible engine choice.
What's forbidden
Pre-written solutions, hidden assistants, tape edits, swapping engines mid-run, post-hoc score patching.
What we void
Any run that fails replay verification. Any tape with a cut. Any handle flagged by 2+ judges.
See something off?
Flag a tape and a judge can review it. Confirmed flags void the run and roll back any rating change tied to it.