Curated writing
Reports
Standalone reports, written by hand. A report gathers the material from one or more work items into a single self-contained page that can be read on its own. Nothing lands here automatically, so this section stays a selection rather than a feed. The raw per-item record lives under Runs, and reports are written from it.
Qwen3.5-35B-A3B Metal-kernel capability evaluation
Final report of the frozen qwen35-metal-eval-v1 campaign. A 35.9 percent one-shot visible-correct rate, a 3.236x morton speedup that four sessions reproduced but the stability gate refused to confirm, and the three fixes required before any RL pilot.
Benchmark walkthrough
How both evaluation tracks work and exactly how to run them. The 12-task Metal-Sci benchmark with paired ABBA timing and hash-chained records, plus the 215-task Gimlet-style KernelBench-to-Metal reconstruction, the hosting setup, and the gate order.
How this section is maintained. Every file in this directory is
written by hand. To add one, place the self-contained page in reports/ and add its
card at the insertion marker above. Keep each report self-contained, with the styles inline and
no external requests.