Lab notes
experiments with receipts — including the ones that failed
Anatomy of 1,821 nanosecondscase study
How a rare ladder re-anchor was costing 2.6µs at p99.9, and what fixed it.
2026-06-20
The 32σ study: a signal that's real and still losescase study
65,536-symbol backtest. OFI is genuinely predictive — 55.28% over 5.9M samples — and the spread eats it anyway.
2026-06-05
io_uring vs mmap: a null result, honestly spentnull result
Weeks understanding the ring. The mapped file won anyway. Worth writing down why.
2026-05-30