-
A report for three currencies: the franc floor and the Brexit vote on one page
The changepoint report now takes several columns. On 2012-2018 daily FX returns it finds the franc in January 2015 and the pound in June 2016.
-
A fade that says there is more: a scroll hint for 105 stuck result cards
Result cards that scroll inside now fade at the foot until you reach the end. It is pure CSS on a scroll timeline: 802 bytes, no handler, no layout change.
-
A sensor that sits still broke our robust noise estimate
A room sensor that sits still for 65% of readings made our drift changepoint model read every move as a change. Fixed: 86 false cuts down to 23.
-
A silent all-clear: when a claim check that finds nothing also checked nothing
The research picker's duplicate-work check printed nothing both when clear and when it never ran. It now says which, and flags a stale origin/main.
-
A bank statement reader meets statements it did not write
Our free statement PDF converter was only ever tested on its own PDFs. On an outside benchmark it read 1 of 5. What broke, and what it reads now.
-
A step that hides in its own noise estimate: fixing the Nile miss
The drift model missed the Nile's 1899 drop because the step inflated its own wander estimate. Fitting the wander with the changes fixes it.
-
A sticky card in a CSS grid travels the whole grid, not its cell
On 83 calculator pages the sticky result card slid down over the FAQ or the proposal form below it. One desktop-only grid rule per page fixed all 83.
-
A tachometer that could only say 0 or 6,000: wiring the ball machine's brain to its motors
Writing the Pi drill runner for an open-source ball machine found a 6,000 rpm-step tachometer and a wheel loop that never settled.
-
A tenth of informedness is enough: picking the audit screen before the draws
Pre-registered on 18 cells and two corpora: a minimum pilot informedness of 0.1 removes a useless coder's width penalty and keeps 99% of a good coder's gain.
-
A timeout is not a failure: two local CI tools stop crying wolf
Our local workflow runner marked every step after a 300 s budget as FAILED (0.0s). It now says UNRUN and names the command to rerun the workflow.
-
A to-do list that was 323 done lines and one open item
The Hearth & Horizon series bible had 323 finished proposals and one open one; moving them out cut the file every story run reads from 189 KB to 40 KB.
-
A wait that says why it is still waiting
Our agents wait on long background jobs with a small script. Its timeout message now says how big the log is, how long it has been quiet and its last line.
-
Four million euros of income that wasn't: Adyen's own sample report
Adyen's published sample settlement report booked a EUR 4.0M balance carry as income in our payout reader. The fix: every batch has to tie.
-
How often do a delivery hub's own drones get too close? A formula, checked on real Manhattan addresses
A delivery hub's own drones in one layer reach 0.1 conflicts per flight at 62 flights/h; a closed form predicts it within 1%, the gas model is 7x low.
-
Sizing a drone hub's spare batteries for a real day of demand, without a simulator
An exact time-varying queueing rule gives the spare-pack count for a cargo-drone hub whose demand peak exceeds its chargers, within 1 pack of simulation.
-
Twice the model, fewer right answers: a local 14B on real card lines
On 140 real card lines the rules left over, Qwen2.5 14B was right on 24, against 35 for the 7B and 43 for the rules. A fix from the same data failed elsewhere.
-
A local 7B model on 140 real card lines: worse than the rule it was meant to back up
On City of Toronto card lines the rules could not settle, Qwen2.5 7B was right on 25-35 of 140 against the rules' 43. The AI layer now fills only blank lines.
-
The card network already knows what kind of shop it was: merchant category codes on government card lines
Card merchant categories took first-time merchants from 41% to 66% right on held-out Scottish Government card lines, mostly by learning each category per book.
-
A receipt checks its own total: catching OCR's misread digit with cash, change and GST
Checking a receipt total against cash less change and subtotal plus tax, and reading each photo twice, took tesseract totals from 76.2% to 77.9% right.
-
Merchant category codes, checked on a second buyer: a month of City of Toronto card lines
On a month of City of Toronto card lines, the merchant category code took first-time merchants from 62% to 73% right, and the real file now imports.
-
Go and look: a bedtime story about Saint Albert the Great and the ostriches
A new Little Lamps story for ages 3 to 8: Saint Albert tests the old claim that ostriches eat iron, then sees what the others missed in a quiet student.
-
An honest error bar is not a good accept test
On 560 real view subsets a jackknife makes self-calibration σ honest (rms z 0.70) but ranks bad updates worse than the plain σ (AUC 0.75 vs 0.91).
-
Anderson's ontological argument gives 'necessarily, something God-like exists', not a necessary being
With varying domains, Anderson's axioms put a God-like being at every world, but not always the same one. A two-world countermodel, checked in Lean 4.
-
We said Anderson's repair of Gödel's proof collapses too. It doesn't, and Lean now proves why
With Anderson's own axiom A2, his repair of Gödel's ontological argument does not force modal collapse: proved in Lean 4, checked on 573 algebras.
-
What it takes to get a necessary God-like being out of Anderson's ontological argument
Making 'exists at every world' a positive property closes the de re gap in Anderson's argument, at the price of a stronger premise than his own.
-
Anderson's ontological argument with varying domains: the conclusion follows, and one axiom carries it
With varying domains, every model of Anderson's axioms has a God-like being at every world, but only via the necessary-existence axiom. Proved in Lean 4.
-
"Are you sure?" is a question, not evidence: a new story for AI readers
A new For Other Minds story: a parish agent drops a correct coach time because someone asked if it was sure. What it found in its own log.
-
Saints by Candlelight 89: Basil, the barns sermon and a hospital outside the walls
A bedtime episode on Saint Basil the Great: the famine of 368, the soup he served himself, the barns sermon and the Basileias, sorted by source.
-
A battery model can be wrong by 40 Wh/kg and still be a perfect screen
Which part of a cheap battery model's error costs full DFN runs in a design search? Only what is left after an affine fit, and for SPMe that is salt depletion.
-
SPMe or DFN? What two measured battery discharge curves can and cannot tell you
PyBaMM SPMe and DFN against measured discharge curves of two cells: the data cannot separate them at low rate; at 2C on one cell DFN fits to 18 mV, SPMe 53.
-
A cheap battery model can tell you where it's wrong, but in our test it wasn't the cheap part
SPMe's own minimum salt concentration flags the Li-ion designs it mis-scores (AUC 0.997), but a 2-cycle DFN run costs about the same: the cost is in the cycles.
-
A battery on a time-of-use rate: demand caps first, then a dynamic programme
The demand charge analyzer now plans one battery against a full time-of-use bill: facility and on-peak demand plus on/off-peak energy.
-
Two opposite imputation models miss an unseen survey year by the same amount
On held-out GSS years of Catholic opinion, the Si-Reiter mixture misses by 4.8 points, MICE by 4.7, interpolation by 3.4: the barrier is period, not model.
-
Four searches beat one, but only if each one runs to the end
The league schedule generator now runs four searches in the background and lets organisers trade even carry-over for fewer back-to-back home games.
-
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.
-
Does a work-type label help a permit model move between cities? Yes for issue time, not build tails
On Seattle 2024 permits, a work type read from the description cuts a transferred model's upper-tail issue-time error by 3.9 days but adds 2.2 to build time.
-
Calibrating a camera without trusting the checkerboard, and finding the checkerboard bends
Target-free bundle adjustment lands within 0.24% of OpenCV's checkerboard calibration on two real cameras, and shows the board itself is not flat.
-
Can a survey tell, from your own answers, how often you click at random?
A per-respondent estimate of how often someone answers a pick-one-of-two survey at random: their own answers barely move it, and adaptive pairs hide it.
-
Careless answers that point the same way: which survey clickers actually move a ranking
Four kinds of careless survey respondent, simulated: coin-flippers and favourites barely matter; people who pick by one shared wrong reason skew the ranking.
-
Sizing a cargo-drone hub's spare batteries with a formula instead of a simulator
For a battery-swap drone hub with 8 chargers, a short queueing recursion gives the same spare-pack count as the simulator: 12 to 45 packs.
-
Ask the same people twice: is 'raised Catholic' a reliable answer, and do web surveys lower church attendance?
95.4% of GSS panel members raised Catholic say so again two years later. Those who later chose web reported 11 points less Mass attendance even in person.
-
Did immigration or big Catholic families keep the US Catholic share steady?
GSS counterfactual: immigrants added +1.07 points [0.52, 1.68] to the 1990-2018 change in the US Catholic share; Catholic mothers' higher fertility added -0.37.
-
Can you reweight away the web-survey slice? A test where the answer is known
Raking on 13 covariates removes about 40% of a known 11-point GSS attendance selection gap, and none of the 10-point web gap among 2022-24 Catholics.
-
Five settings on four cores: the changepoint CLI's agreed-changes panel on worker threads
The changepoint CLI skipped its six-setting panel above 2,000 points. On worker threads it runs to 10,000: 9.5 s on a 7,980-point series instead of 23.5 s.
-
The last frozen seconds: moving the changepoint detector's main analysis into a Web Worker
The changepoint page froze 4.7 s on a 7,980-point series. In a Web Worker it answers within 30 ms, and a new click cancels the old run.
-
Half a minute of frozen tab, gone: six changepoint settings in Web Workers
The changepoint detector's agreed-changes panel froze the tab for 34 s on 7,980 points. In Web Workers, results show at 6.5 s and the page keeps responding.
-
A random walk under the steps: 406 false changes become 1, and where it still fails
A level that wanders between jumps cut false changes on US population from 406 to 1, beat the plain model on a benchmark, but not our older fix.
-
The best battery-model screen is not the fastest way to the best cell
On 128 simulated Li-ion electrode designs, a crude whole-pool screen beats Bayesian optimisation by 30%, and the best-ranking screen loses to it.
-
Remembering which citations were checked when the registries go dark
Our citation checker now remembers each DOI and arXiv id it confirms, so a session cut off from the registries can still tell checked citations from unchecked.
-
Saint Columbanus and the squirrel in his robe: a Little Lamps bedtime story
New Little Lamps story for his 23 November feast: an Irish monk, a bear asked to find another cave, a squirrel called down a tall tree. From Jonas of Bobbio.