TinyCodeTest Evaluation Stack
Deterministic verifiers, difficulty buckets, and multi-model pass@k benchmarking
Notes on how the TinyCodeTest evaluation stack turns model runs into reproducible, inspectable benchmark results.
Experiments
Longer notes on systems experiments, evaluation design, and engineering tradeoffs behind selected projects.F1 Strategy Lab Notes
Scenario design, deterministic grading, ablations, and stress tests for strategic reasoning
Experiment notes from building and evaluating an RL-style environment for race-strategy reasoning.
Compression Benchmark Notes
Repeatable measurement, SIMD-aware implementation work, and fair algorithm comparisons
A closer look at the benchmarking discipline behind Compression Bench.