advisor-bench-v0.1
Paired with the methodology field note the-refusal-floor-is-trainable.
Benchmarks.
Open benchmarks for vertical-finetune evaluation — each one anchored to public sources, with the methodology field note that documents how it was built. 03 published.
hermes-brain-bench-v0.1
Paired with the methodology field note picking-the-hermes-brain-on-spark.
patent-strategist-bench-v0.1
0.625closed0.850retrieval0.950oracle
Paired with the methodology field note patent-strategist-v1-baseline-on-spark.
※ More benchmarks in preparation End of Jun 2026