What BenchSloth actually checked

The source-linked browser desk and deterministic handoff work without a model. Neither a cited passage nor an elapsed timer proves scientific progress.

TestObserved result
Full private SOP ingestion5/5 imported; 300/300 exact quotations (source integrity, not scientific accuracy)
Python tests163 passed on uploaded source
JavaScript assertions60 passed on uploaded source
Gemma 3 4B Context Recovery3/27 attempted cases gave complete exact handoffs; 26 returned; 1 timed out; 1 of 28 unattempted
Paired deterministic handoff27/27 complete
Independent human scientific review0 completed; unknown accuracy and time savings

Gemma is experimental. The static site does not call a model. At least one invented edit was rejected by the application. Full synthetic comparison reports are in the public source and evidence archive.

Back to BenchSloth