NKESSLING@TERMINALRTM 22:53:38 · UTC 20:53
NOTE ~/writing TOOLING
Benchmarks that fail the build
Keeping performance from drifting with a 5% regression gate on every pull request.
NOTE · PLAIN TEXT
Keeping performance from drifting with a 5% regression gate on every pull request.
Keeping performance from drifting with a 5% regression gate on every pull request.
Performance rots one small change at a time. A benchmark that runs on every pull request and fails at a 5% regression makes the drift visible the day it happens.
SUB NEWSLETTERMONTHLY
One note a month. No tracking pixels.
New research notes by email, plain text, with a one-click unsubscribe.
NOTE WRITING · RESEARCH NOTES8 NOTES
Keep reading.
Short write-ups of things that broke and how they got fixed. Plain language, real numbers.
SCRN WORK · SCREENER6 OF 8 SHOWN
The projects behind the notes.
Filter by language, sort any column, and hover a row for the preview. Each row opens the full case study.
ROLEKEY RESULT
TIDEA write-ahead log that survives the power cordtidewaterAuthor · maintainerRust2024–264,8121.92 GB/s append, one coreKILNA build cache that made CI six times fasterkilnAuthorGo2023–262,2076.1x faster CI, medianLOOMA Postgres job queue that stays fair under loadloom-queueAuthor · 3 contributorsTypeScript2024–251,39012k/s jobs on one nodeWRENRelease notes that write themselves from commitswrenAuthorTypeScript2025–261,04511 min saved per releasePRSMRead traces in the terminal and diff two runsprism-traceAuthorRust2025961O(n log n) trace diff, was O(n²)FERNA small config language with typesfernAuthorZig2025–266120.4 ms compile a 2k-line config