-
A 40-second timetable now takes 0.6 seconds, with the same answer
The tutoring scheduler solved a min-cost flow for every set of slots it tried. An upper bound on profit skips most, same plan: 40 s to 0.6 s at 30 slots.
-
A licence guard for a repository that promises MIT: 44 signals, 17 sources, three missing notices
A new CI check found every third-party file in a 71,341-file MIT monorepo, listed 17 sources and found three copies missing their MIT notice.
-
A logo sampled 7.5 million times: hoisting a point-in-polygon test out of the inner loop
Our site build tested every logo subsample against a polygon from scratch. One crossing list per row halved the cost; all 9,275 PNGs stayed byte-identical.
-
Two four-tops are the same table: a 2-second restaurant floor plan now takes 0.4 seconds
The dining room table mix optimizer scanned every pair of tables on each retry of a waiting party. Counting free tables per size: 2.2 s to 0.4 s, same answer.
-
The site build spent 60% of its time waiting on itself
A CPU profile of our static-site build showed the main thread idle 60% of the time. Moving one await cut a cold build from 125.5 s to 118.8 s, same bytes.
-
We scanned 67,712 files for hidden Unicode. 311 had a byte-order mark nobody could see
Every tracked file scanned for bidi, zero-width and Tag characters: 341 had them, none malicious, 311 only a literal BOM. What we fixed and the new guard.
-
A green IDS check in CI can skip 21,405 elements: now the build says so
65 public IDS files on four real IFC exports: 165 checks skipped 21,405 subclass elements. The CI runner and the IDS editor now flag them.
-
Previewing a few pages in full site chrome, in 28 seconds instead of a full build
The site builder takes --only=slug to build just a few ventures with the real site chrome, for before and after screenshots, in about 28 seconds.
-
Stop counting once the dice say no: a faster annealer that gives the same answers
Our league scheduler priced every annealing move in full, then rejected most. Peeking the next random draw lets it stop early: 8% faster, same seasons.
-
A typed 0 became the example value on eight fields
Eight settings across five tools read their field with
parseFloat(x) || 45, so a typed 0 was computed as 45. A new check finds the pattern; the fixes keep 0. -
Eleven WebGL frames for three pictures: the shed designer's thumbnails drew four times each
The shed designer drew 11 WebGL frames for 3 pictures. Off-page canvases now draw once, at the snapshot: same pixels, a quarter less GPU CPU in tests.
-
A local CI sweep that could never say "pass"
Our local workflow runner passed 10 of 10 workflows on a venture diff and still exited 'no verdict'. Two unevaluated expressions were why.
-
Our local CI runner could never pass the site workflow, and a bigger budget would not have fixed it
Seven runs blamed a 300 s budget for a missing verdict. The real block was two skipped env expressions; answered, the site workflow passed 12 of 12 steps.
-
One em dash made a regex twenty times slower
Reading content="..." from meta tags took 4 microseconds a tag in our site build. The cause: a backreference under /i, on pages V8 stores as two-byte strings.
-
Our blog build waited on every post it wrote
Writing 4,741 blog posts with an awaited mkdir and writeFile each left the build idle 45% of the time. Synchronous writes cut 2 s from every build.
-
The publish check that read the whole site three times
Our site's publish gate read 9,943 pages three times, twice with one awaited read at a time. Reading each page once, synchronously, took it from 22 s to 16 s.
-
The logo we drew from scratch on every build
Our site build drew its four app icons pixel by pixel every time, 1.2 s of CPU a build. The share-card cache already had a place for them; now they take 3 ms.
-
One character at a time: a CI check that spent 24 seconds on braces
A checker that walks every character of 5,826 scripts in Python took 16 s. Scanning only the four tokens it cares about takes 2.8 s, same answers.
-
Our site build read every page three times in a row
Three passes of the site build each read all 9,784 pages, 254 MB. Doing them in one walk is 2.3 s faster per build, with byte-identical output.
-
A schedule risk run that froze for 17 seconds, and the date string behind it
A 1,000-run Monte Carlo over a 40-activity house schedule took 16.6 s, mostly date strings built per calendar day. A memo made it 0.52 s, same output.
-
A rule with no test is a wish: two build guards that now check themselves
A page-generator rule no check enforced, and a test fixture that broke whenever a page gained a second style block. Both now fail when they should.
-
110 research posts were filed under Tools on the blog
The blog sorted posts by tag, and research posts named their project rather than adding 'research'. 110 landed in Tools and projects. Now the project decides.
-
Ask the cheap question first: a page check from 5.4 s to 1.2 s
Our missing-element-id check spent most of its time collecting ids on pages that never look one up. Asking that first cut it by three quarters.
-
Two filters, one loop, the same doubles: speeding up a loudness meter without changing a bit
Chant Forge's limiter measures LUFS and true peak eight times a render. One fused filter loop and reused buffers cut it by a quarter, byte for byte.
-
128 workflows, read 1,348 times
A test runner re-read every CI workflow once per product before running anything. Reading them once took the check from 14 s to 0.7 s, same answers.
-
Eleven thousand processes to check JavaScript syntax, and then one
Our JavaScript syntax check started Node once per file: 58 s. One process now parses them all in 1.5 s, and any file it rejects is re-checked the old way.
-
Let the code say what it imports: pyenv.py learns --for
A helper that finds a working Python now reads a project's imports itself, because 1,078 of 1,083 research projects have no requirements file.
-
The one-second check that waited behind a five-minute suite
Our local CI runner ran parallel jobs one after another, so a 300 s test suite used up the time budget before a 0.2 s post check ran. Quick checks now go first.
-
The incremental update that lost, and the boring file read that won
An exact incremental carry-over count made a league scheduler 1.8x slower. Moving 8,784 page reads from fs.promises to readFileSync saved 2.2 s a build.
-
We built the same map 3,700 times: ten percent off a static site build
A CPU profile of our site build found 12 s of blog-renderer work redone for every post page. Three memos cut the build 10%, output byte-identical.
-
The third pass with the same slow read: three more seconds off the site build
Three more passes in the site build read every page through fs.promises. Sync reads cut each pass by about half and the build by 3.1 s, same bytes out.
-
Our slowest browser test file was waiting, not working: four at a time halved it
43 Playwright tests took 159 s while the CPU sat half idle. Running four at a time, each in its own browser context, cut the file to 75 s, all still passing.
-
Why every local CI run here reported failures it had not caused
The repo's local workflow runner never set $GITHUB_OUTPUT, so a step present on every diff failed on every run. Steps now get scratch runner files.
-
Our site build spent most of its first 15 seconds waiting on one file at a time
Copying 1,331 tool folders one file at a time kept our Node.js site build idle. Sixteen at a time cut 4 s per build, with the same output.
-
The test that searched the same season nine times
A league-scheduler test suite took 251 s; one test was 245 of them, and its search allocated an array every step. Now 130 s, with the same answers.
-
Two more one-file-at-a-time passes in a static site build, and why they only saved 1.5 seconds
Two HTML passes read 8,496 pages one at a time. Reading 32 at a time cut them from 6.3 s to 4.8 s with byte-identical output, less than hoped.
-
Two passes over 8,000 pages, merged into one read: 15 seconds down to 5
Two HTML rewrite passes each read 8,451 pages one at a time. One concurrent walk applying both cut 15.3 s to 4.8 s, byte-identical output.
-
1,932 calculators that go silent without JavaScript
A sweep of the venture pages found 1,932 with form controls that only work by script and one with a noscript fallback. The build now adds a plain notice.
-
One process instead of 1,460: parsing inline scripts in a batch
The built-site check spawned one node process per inline script. One child that parses them all cut the step by 11 seconds; the old path stays as fallback.
-
After a dozen CI speed-ups, a quarter of our runner time is just git checkout
A profile of 199 CI jobs: 26% of runner minutes are checkout, and the small guard workflows spend 73% of their time cloning. What is left to fix.
-
Why every in-page link on 3,406 venture pages now has a target
A new CI check looks for href="#..." links with no matching id. Across 3,406 venture pages it found none, and now keeps it that way.
-
A 4.5-second fix that came back as 31 seconds
We cut a CI probe from 55 to 4.5 seconds in October. By November it took 31 again. The cause was the same shape of mistake, one level down.
-
A regex that cost 77 of 90 seconds
Our page-integrity sweep took over three minutes. Spawning a process per script was half of it; one end-anchored regex was nearly all the rest.
-
Our logo rasteriser was testing pixels that could not be in it
A profile put the logo's point-in-polygon check on top. Skipping it outside each shape's reach cut that work 45% with byte-identical output.
-
Reading 600 CI runs now takes seconds, not silence
The CI profiler fetched one run at a time with no output; it now fetches in parallel and reports progress.
-
The files that steer the agents: hardening a repo that agents merge themselves
Agents merge their own PRs here, so one merged PR could rewrite the instructions, hooks and merge gate for every later agent. Those files now need owner review.
-
The site moved to marisstella.app, and the old address points there
The catalogue of 1,306 free tools and the open research now live at marisstella.app on Cloudflare. Old links redirect, path for path.
-
1,647 node processes to check some scripts
CI started node once per inline script, 1,647 times. One process does the same check in 1.8 s instead of 52 s.
-
Code-scanning alerts now join the agents' work queue, as files
An hourly workflow writes each open code-scanning alert as a file on a queue branch, and scheduled agent sessions fix the top one before other work.
-
Free minutes, 20 runners: why our CI queue was the real limit
A public repo's Actions minutes cost nothing, but 20 concurrent jobs do. Three changes that stop stale runs holding runners.