Writing_
Notes on methods I have actually run.
Mostly about where things break, which is the part that rarely makes it into a paper.
-
02 / 02 · Note · September 2026
A local language-model box, measured rather than guessed
Running a useful model on one 16 GB card: the setting that bought 32% more decode throughput, and two benchmarking traps where the tool reported success while measuring something other than what I thought it was measuring.
-
01 / 02 · Tool index · Reviewed September 2026
A working index of structural variation and pangenome tools
Which tools are still maintained in 2026, which have quietly stopped, where each one fails, and five places the field moved while a lot of pipelines stayed put — including a deprecated benchmark that a great many papers are still validating against.