what we measured
Evidence
Every entry is a lab record: a claim, the bench that tested it, and what survived. A failed result is evidence too, so it gets an entry like any other. Newest first, each one linked to its canonical artifact.
provenance for every number lives on the benchmarks page
rss2026-08-08
evidence
What mixing taught me about evals
Judgment under unreliable perception is a solved craft. Music solved it with structure, and my synth now carries a 3,050,113-assertion regression suite to prove the transfer runs both ways.
Shrey Patel
2026-08-08
evidence
A model in the room
Building a synthesizer plugin with an AI pair: three jobs the model has in a creative tool, and the fence that keeps it useful.
Shrey Patel
2026-04-19
evidence
Tenant fairness on shared inference
KVWarden Gate 2: 53.9 ms quiet TTFT, 61.5 ms under flooder pressure, and 26× better tail behavior than FIFO.
Shrey Patel