-
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 claim table read a lagging index
allocate.py found other sessions' claim PRs through GitHub search, which can trail a new PR by minutes. It now reads the REST pulls list first.
-
The columns a Nikon stores that aren't picture, found without a camera table
Five of seven Nikon bodies store masked and padding columns at the right edge. A rule read from the sensels matches LibRaw's model table on all seven.
-
The cycle check that cost more than finding the steps
One periodogram check took 679 ms on an 8,609-day series. A one-period trig table cut the whole analysis from 1.56 s to 0.92 s, same answers.
-
The date regex that read 2026 as 2020
A shift importer read 01/05/2026 as 2020, and 37 parsers read 2025-03-091 as 9 March. One missing (?!\d) in 178 regexes, now fixed and linted.
-
The SaaS seat tool's notice deadline moved a day in March, and only outside UTC
A 1 April renewal with 30 days' notice showed a 1 March deadline in New York. The fix, a lint for the pattern, and a full test sweep run in New York's zone.
-
We gave a leak localiser the exact hydraulic model, and it found no more leaks
On BattLeDIM L-Town, the generator's own model places 15 of 32 leaks within 300 m, against 16 for the nominal model. The slow leaks' residual fails.
-
We tried to fix a calibration solver's far minima. They were the data.
Warm-starting 11,442 self-calibration solves from the spec focal length reproduced every far minimum, and a basin check made the update test worse.
-
The Field That Held the Water Back: a story about a culvert that was too small on purpose
A new Hearth & Horizon story: a bigger pipe under a school lane moves the flood to the next culvert, and a dry basin on an estate green fixes it.
-
The footer said "like this one" under a watch face
Tool footers said 'like this one: everything runs in your browser' on 28 pages with nothing to run. Now the line needs a control on the page or its tool.
-
The physics was fine; the function calls were the cost
Inlining three hand-written RK4 loops cut the open tennis robots test suite from 160 s to 96 s with bit-identical results.
-
The k in lux, and the m nobody could hear
Chant Forge's preview choir now sings the k of lux and dex. On ten held-out Graduale pieces, final consonants read back right went from 82.9% to 90.0%.
-
The last line should say what broke
Agents read tool output through tail -1. Two checks ended on the wrong line; now one ends on the failing field and the other on the test count.
-
Hearth & Horizon: a draining tank, a straight-line forecast, and the last quarter metre
A new Hearth & Horizon story: an assistant promised a tank empty by 00:35 from one test hour. Torricelli's law says 06:37.
-
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.
-
A longer arm was not why our robust drone design was robust
A fixed-arm control run shows a robust quadrotor co-design gets its robustness from mass and gains, not its longer arms, which help a fixed tune only a little.
-
The cent missing from 121 contractor proposals
Our shared proposal module printed a deposit and balance that missed the total by a cent when a price had sub-cent digits. Fixed in all 121 copies.
-
For Other Minds: a year of survival that came from the sorting, not the visits
A new For Other Minds story: visited housebound parishioners seem to live a year longer. Everyone on the list had to survive long enough to reach it.
-
The opening l in a synthetic chant choir: a 12.5 dB swing cut to 0.7, and a reader that saw nothing
Chant Forge's choir sang an opening l from 7.7 dB under to 4.8 dB over its vowel. A per-note gain fixed it; the consonant reader's report did not change.
-
The pay day was above the table
An open-source bookkeeper refused all 85 files of a public Gusto-style payroll journal. Four layout faults fixed; each run now matches its totals row.
-
Deriving the battery fade correction from the plating law works on one cell, not three
PyBaMM's plating exponential with no fitted constant cuts the 100-cycle lithium-loss error on an NCA cell from 30% to 3%, but double-counts on Ai2020.
-
The plating law predicts battery fade on fresh pools, but a better forecast can rank designs worse
With no fitted constant, PyBaMM's plating feedback predicts 100-cycle lithium loss within 2-10% on 192 fresh designs; a smaller error did not always rank better
-
Below 1% prevalence, the audit's own interval stopped being valid
Pre-registered: the pooled interval covered under 93% in 13 of 21 GoEmotions cells under 1% prevalence, down to 1.6%. A Wilson fallback fixes all 21.
-
The Rod by the Vestry Door: a story about a lightning earth that read 573 ohms
A new Hearth & Horizon story: a stolen copper conductor is replaced with aluminium, and the new earth rod in dry rubble reads 573 ohms against the 10 asked for.
-
The Same Box of Books: a story about why the smallest parishes top and tail every ranking
A new For Other Minds story: a diocese ranks 48 parishes by confirmation rate, and the best and worst five are all tiny. A funnel plot finds the cathedral.
-
The same search, four times: a 22% faster test file and a speed-up that wasn't
The league scheduler's slowest test file mostly re-ran one seeded search. Sharing it saved 29 s; the obvious incremental cost did not help.
-
The test-scheduling fix we measured and did not make
Two full timings of 1,365 test suites agree on the slow four. Scheduling by measured time looked like a win until we checked the lower bound.
-
The Shadow of the Master's House: a story about one shaded cell in eighty
A new Hearth & Horizon story: a gable shades 1.25% of a school's solar roof and costs 11% of its power, and after a shower the inverter climbs the wrong hill.
-
The shed nobody looked at: half the framing in a shed designer was for a warning that got thrown away
One line in the shed designer framed a whole second shed to quote its peak in a warning, then threw it away. Deferring it cut 1,444 framings to 760.
-
Rerunning the tennis lift's jam sweep with the belt as 62 masses: the cutoff holds, a start goes slack
The tennis feeder's lift belt as 62 masses on springs in the jam sweep: every jam still cut by 89 ms, belt pull capped by the slip hub, slack on starts.
-
The tennis lift's belt is a spring too, and it rings where the coupling does
The tennis lift's belt stretch, modelled: it splits the 49 Hz drive mode, barely moves jam loads, eases starts, and slackens the down strand in hood jams.
-
The tennis lift's empty belt rings at the rate its flights pass, and no belt tension fixes that
A string model of the tennis feeder's lift belt: the empty belt rings inside the flight-passing band, and the pluck figure was 24% high.
-
The tennis lift's jam sensor stops reading the throttle
A low-side shunt reads duty times current, so the tennis lift's jam trip rose as the duty fell. A shunt inside the diode loop holds it at 1.77 A.
-
The tennis lift's pulley gets its own shaft, and the motor has to get shorter
The tennis feeder's lift pulley moves onto its own shaft in two bearings so the gearbox carries torque only; the collision check then caps the motor at 75 mm.
-
The tennis lift's slip hub goes back to 0.9 N m, and 0.8 is one step too far
Three full jam sweeps on the tennis feeder's lift: 0.9 N m cuts the flight's worst hit from 91 to 84 N, and 0.8 starts slipping on soft starts.
-
The test that passed by waiting for an array to break
A league scheduler test asserted a throw and got one, 4 s and gigabytes later, from an endless loop. That and a shared search cut two tests from 81 s to 18 s.
-
A camera's twin can't make its views better, but it knows which ones to trust
A linearised twin of the camera's calibration ranks which updates can catch a 2% focal change, and testing its top quarter doubles detection.
-
The voltage collapse was the slack bus
On five PGLib-OPF RTE networks the DC dispatch looked past a voltage-collapse fold. The cause was a 30 MW slack unit. Moving it gives all five a power flow.
-
A search that got unlucky: how 28-team leagues got fair schedules on the page
A 28-team schedule search found nothing: its walk had a heavy tail and its start was too rigid. Fixed, 28-team leagues get two new seasons a seed.
-
The Wall Down the Middle: a story about a dance floor that was strong and still bounced
A new Hearth & Horizon story: a parish hall floor passes its load check, then bounces at a céilí. Slowing the band makes it worse.
-
Hearth & Horizon: frost heave, ice lenses, and a bell frame on the wrong ground
A new Hearth & Horizon story: 50 cm pads in lake silt, frost near a metre down, and why ground lifts because water arrives, not because it expands.
-
Three games a night: a wrong bound, a 32,970-night proof, and a search stuck where it started
Wait-free three-game league nights: a wrong bound for odd leagues, one setting impossible by exhaustion, and a search stuck in its swap class.
-
Three games a night, written down instead of searched for
Searching for back-to-back three-game league nights got stuck. A written-down design took 24-team leagues from 254 waits to 0.
-
Three MT940 readers, one generator: the reader that kept only the first statement
Random valid MT940 files fed to three readers in this repository: one kept only the first SWIFT message in a file and its balance check still passed.
-
Fewer than three positives in the audit: the hand labels now close the interval
Pre-registered: small audits at 1 to 2% prevalence also broke the pooled interval. A Wilson fallback at under 3 positives fixed all 47 such cells.
-
Three YNAB budgets, and why "Seattle Public Utilities" went unsuggested
A bank-line categorizer tested on three real YNAB budgets: rules written from one household, read on two held out. The held-out gain was small.
-
Toss It Again: a story about a hundred and twenty tests and the three flags chance always makes
A new For Other Minds story: a schools trust's agent flags three departments at p < 0.05 out of 120 tests, exactly what an all-average trust would produce.
-
Looking for one rare Nikon format detail turned up a bug that made 12-bit NEFs develop nearly black
We probed 235 public Nikon raw files for a rare split row. The split rows decoded right, but three other faults showed up, one of them in every 12-bit NEF.
-
Twenty-Four a Minute: a story about a stair that empties at the speed of a handshake
A new Hearth & Horizon story: a 2.4 m church stair should empty in four minutes after Midnight Mass. It takes eighteen, because of the priest.
-
Two chant-scan reader faults, and the one we had misdiagnosed
Chant Forge's scan reader checks GregoBase transcriptions against the 1908 Vatican Graduale. Fixing its last two known faults turned up a wrong diagnosis.