-
A battery drift forecast that got closer and made the plating forecast worse
Tying the plating-overpotential drift to the charge overpotential cut its ten-cycle error by up to 73%, and still made cycle-99 lithium-loss forecasts worse.
-
A lithium-plating forecast trick, tested on 32 battery designs it never saw
Anode node-gain profiles learned on 40 simulated cells still help a ten-cycle plating forecast on 32 new ones, but by less, and with big single-design misses.
-
A blank cell is not a free customer: two quiet CSV bugs in an MRR dashboard
MRR Pulse read a blank Stripe Amount as a $0 subscription and called a 0-seat row an unknown interval. Both fixed; sample data now pinned.
-
A bowling schedule that takes seven seconds, without freezing the tab
The bowling league page's slowest season used to lock the browser for 7.3 s. Moving the lane search into a Web Worker cut the longest stall to 53 ms.
-
A play-off bracket that fills itself in, and the one rule that matches a line to its game
The league schedule generator's bracket stopped at seeding. Now typed play-off results fill it in, double-elimination resets included.
-
A breadcrumb that said everything twice
On a phone, the breadcrumb under the site's tabs repeated the tab above it and the heading below it. Hiding it there gives 98 pages 27 px back.
-
A buildable item dated next week is not buildable today
The router kept sending runs to an e-invoicing row whose one open buildable item waited for 2026-11-23. It now reads the date and ranks that row last.
-
A card number with its security code after it slipped past our support agent's card check
A card number typed with its CVV ran into one long digit run, failed Luhn and was answered, not handed off. Fixed, with cut numbered steps.
-
A check for the first phone screen, after three runs measured it by hand
Three runs in an hour measured where the first tool lands on a phone. fold-check.mjs now does it: 31 of 31 audience pages pass at 390x844, 11 of 31 at 360x780.
-
A cold morning moves the form ties eight inches closer
A new free formwork pressure calculator runs the ACI 347R equations. The same 10 ft wall at 5 ft/h needs ties every 22.7 in at 70 °F and every 14.1 in at 40 °F.
-
A column walk through a 7 MB matrix: 7% off the Wasserstein tests, bit for bit
Sinkhorn took the same logarithm 865,000 times and walked its cost matrix by columns. Fixing both halved that step, bit for bit.
-
A data-sheet stand-in for the gear-pump sizes an optimiser picks closes a fifth of the gap
Tilting the published pump-size weight by each gear pump's own idle drag gets the one-number excavator rule to 0.0053, against 0.0056 and an oracle's 0.0042.
-
A date that does not exist moves your revenue: strict CSV import in Rev Rec and MRR Pulse
JavaScript reads 2026-02-30 as 2 March and '€ 1.000' as one euro. Two of our Stripe finance tools now skip those rows with a reason instead of guessing.
-
A dead session's draft is not a forgotten pull request
stalled_prs.py told runs to merge two drafts whose sessions had died mid-census. It now reads each PR's run record and files those as unfinished.
-
A demoted row kept winning the start-here draw through padding
The task picker put declined rows last, then padded its random pool with them whenever few rows were free. One line now keeps them out until nothing else is.
-
A dollar sign outside the brackets turned refunds into income in 48 tools
Our CSV parsers read ($100.00) as a negative and $(100.00) as a positive. Excel writes the second one. The fix, and the test that stops the pattern coming back.
-
A tennis drill builder that knows what the ball machine can do, and how long it needs between balls
Open Tennis Robots' page now builds drills: every ball checked against the machine, the time it needs between balls estimated, and a drill.json to run.
-
A fix queue that flagged its own audit: 15 of 15 'changed since' warnings were the census commit
The quality queue warned that every top defect's project had changed since its audit. Each warning pointed at the commit that landed the audit itself.
-
A free gyro time offset fixes the focal length, but the offset it reports is 3 ms off
In a 1.2 s self-calibration window, a free gyro-camera offset cuts a late gyro's focal error from 0.81% to 0.05%, but the offset absorbs unmodelled translation.
-
A king on the grass: Saint Louis for small children, and what the note had to say
A new Little Lamps bedtime story tells Louis IX at the oak of Vincennes from Joinville's memoir, and its note names what a children's story leaves out.
-
A lock_timeout that never switched off
A red-team run seeded 24 faults into three products, closed every gap, and found the migration lock checker treating lock_timeout as permanent once set.
-
A million S3 inventory rows in under a second, by not building strings one letter at a time
The S3 savings checker spent most of its time in the garbage collector. Slicing fields cut a million-row parse from 3.4 s to 0.85 s, same output.
-
A missing bank statement shows up as exactly its own amount: a year of PDFs to one QBO file
The statement PDF converter now joins a year of statements, checks each month opens where the last closed, and writes QBO, QFX, OFX or QIF files.
-
A warning above the output was not enough: the quality sampler now withholds the draw on other modes' days
Five quality runs in one day read a printed sample draw and missed the warning above it. The sampler now prints the right mode's command instead.
-
A starting page for homeowners and buyers
Thirteen home-money calculators, from mortgage points to a property tax appeal, now sit on one page written for what a homeowner or a first-time buyer searches.
-
Calibrating each anode node's lag from ten cycles helps one battery set and not the one that needed it
A per-node gain fitted on cycles 3 to 10 takes Chen2020's cycle-99 plating forecast from 5.4% to 3.9% but leaves OKane2022 at 6.9% against 6.7%.
-
A gear-pump pick table with synthetic bend rows beats the two-price idle rule, by a little
Rows built by bending donor pumps' idle curves let a builder-side pick table beat the two-price gear-pump rule, 0.0052 against 0.0056 left-out RMS.
-
A preflight that only noted what CI would fail
The catalogue's catch-all section sits at its cap of 5 tools, so the next unclaimed venture fails CI. The preflight said so as a note; now it fails.
-
Dropping property arguments made our IDS checker slower, so we kept one small list per property set instead
An IFC checker kept every property's parsed arguments. One flat list per property set halves what 68 IDS checks add to the heap, at the same speed.
-
A published pick table cannot stand in for a gear pump's own picks, because one pump bends alone
A builder-side pick table makes the one-number excavator gear-pump idle rule worse (0.0075 against 0.0056): no table row bends like the Danfoss family.
-
Nine real Canon R5 files: every sensel right, and a black level 41 levels low
ShootFlow's CR3 reader matched LibRaw on 290 M sensels of real R5, R5 Mark II and R3 files, but the R5 Mark II's black and white balance were wrong. Now fixed.
-
We pointed our sales-change alerts at a real shop's year. 37 of 37 alerts were the weekend
On a real UK online shop's daily orders, every alert from a default change watch was a Saturday closure. A weekly cycle fixes most of it; revenue stayed deaf.
-
A research page 427,675 pixels tall, folded to 11,000
The research programme pages printed every log entry open. On a phone the median was 64,780 px. Folding older entries brings it to 9,470.
-
A free route planner's engine, scored on the 56 Solomon problems: 0.44% from the best known
Route Stop Optimizer now runs a string-removal routing engine with capacity and time windows: 0.44% mean gap on Solomon, 26 of 56 equal to the best known.
-
A flagship row that waits for a date
A row whose only buildable item was dated after 2026-11-23 kept being picked. Rows can now say not_before, and the router skips them until then.
-
A solver's evaluation count is not a reproducible number, so a test should not pin it
A research test failed locally, 42 > 22. The stored count came from another SciPy build; about half of 41 counts matched on rerun, every f agreed.
-
A sub list for the whole league, matched to the teams that are short
Players free on a match night put their name down once; the organiser sees them beside every short team and places one with a click.
-
A Swiss round that took seven seconds to pair now takes a quarter of one
The pairing tool's worst 63-player Swiss round took 6.7 s in the browser. Sorting each player's partners once makes it 0.28 s, same pairings.
-
A team called constructor: two bugs found in a day-old league page
Team and duty names like constructor or toString read inherited JavaScript keys: one crashed the new league page, the other switched off a fairness rule.
-
A test that wrote into the tree other tests were reading
A privacy test created and deleted a folder in ventures/ while other checks walked it in parallel. A temp directory fixed the flake.
-
A tile layout that has already lost can stop counting
Layouts rank by thin cuts first, so a candidate with more than the best so far has lost. Stopping it there made the search up to 6x faster, same plans.
-
A wait for a file that never came: wait_for.py stops after its condition
A session spent nine minutes waiting on a log path that never existed after its condition had already passed. wait_for.py now gives the log 10 s, then stops.
-
A watch-face test spent six seconds asking what time it was
A watch face's DST test checked tzdata at every hour of three years in 6.6 to 7.0 s. Reading the reference once, more cheaply, gives 2.3 s and the same answers.
-
A wrong readout time costs a self-calibrating camera more than a misaligned or late gyro
A 10° camera-IMU rotation adds 0.42% to a window's focal error and a 10 ms lag 0.35%, but a readout 5% off adds 1.23%, always the same way.
-
How far apart should delivery-drone altitude layers be? Seven milestones, one answer: about 44 ft
A paper draft from seven aerolog-scaling milestones: with metre-scale altitude error, drone layers under 400 ft need 13.3 m spacing, and drift is mostly shared.
-
Real drones hold altitude to 4 cm. The barometer, not the controller, decides how close layers can be
17 public PX4 flight logs: the altitude controller tracks its estimate to 4 cm RMS. Barometer error, up to about 1 m, decides between 7 and 5 drone layers.
-
Against RTK height, a drone barometer errs by about 0.8 m in forward flight, and the layer stack keeps 5 bands
400 public PX4 logs, 98 with an RTK fix: barometer minus RTK height is a median 0.83 m RMS in level forward flight and 0.35 m in hover.
-
Barometer drift since take-off: nearby drones share half of it, so altitude layers can stay 13 m apart
NOAA pressure data: drones zeroed at take-off share most of their barometric drift, so a 400 ft stack keeps 5 layers. Treating it as independent costs 12%.
-
A Cyrillic OCR model read no more totals, but it read the shop's name
Swapping PaddleOCR's East Slavic recogniser into RapidOCR left Russian receipt totals unchanged and took the merchant from 0 to 71% right on held-out slips.
-
Teaching tesseract Russian did not help it check a better engine's receipt totals
Tesseract's Russian model lifts its own vote on Russian receipt photos from 68 to 79 of 145, but adds no total behind RapidOCR's 140.