Public BBEH preview
No key required. Same model, same sample boundary, routed differently. Preview run; use BYOK board-depth for public verification.
Free live runs are smoke tests. Small samples may show lower or higher uplift than board-depth or 100+ sample results.
For serious BBEH verification, run the 500-sample Full500 BYOK path; use 20 only as smoke.
Free smoke is available. Start the 20-sample route when ready.
Ready. Start the 20-sample route when ready.
BYOK matched-live harness
Select one lane. Omar keeps the model, sample depth, route boundary, raw outputs, route log, repro manifest, scorer/grader context, and report attached.
This run uses your key directly for the selected benchmark samples. Your key is not stored and is cleared after the run. BYOK is the matched-live reproduction surface: route logs matter only with the manifest, raw outputs, scorer/grader, settings, and numeric decision rule.
One current live lane per benchmark. Internal review and diagnostic paths stay out of the primary selector.
Omar/RCC shows where structured reasoning control improves performance, where it stays neutral, and where mismatch or drift should be flagged instead of hidden.
Routed Omar/RCC is primary. Forced RCC hard is a control mode.
Loading sample counts.
Loading access state.
Waiting for the server to start.
Execution trace / raw output
Ready.
