-
What size service does an all-electric house need? The backup strips decide
NEC 220.82 for a whole house, now in the service load check: in its all-electric example the backup heat strips move the answer from 125 A to 175 A.
-
When did this series stop updating? A date range for a stale feed
The changepoint detector now dates when a series began repeating its last value more or less often. Dating to the step failed its rule; a date range passed.
-
When stickiness hits its limit, a metric watch goes deaf: capping rho in a changepoint alert
15 of the 17 real changes our metric watch missed came from one place: rho stuck at 0.95. Capped at 0.8, it reaches 33 of 52 marked changes, not 29.
-
When to order, and how much: the exact answer is cheap, and the rules of thumb are usually close
Exact Wagner-Whitin order calendars for small sellers. On 6,000 invented cases Silver-Meal averaged about 2% over the optimum, but up to 26%.
-
Whose holiday is it? Canada's thirteen calendars and the prospect on the other end of the call
Call retries skipped the rep's holidays, not the prospect's. Canada's 13 provincial calendars as rules matched 855 of 855 published dates.
-
A 22-team round robin with the least possible carry-over, found by climbing
Annealing over starter timetables gives a 22-team round robin where every carry-over count is 0 or 1, and cuts carry-over 4 to 13% at 24 to 40 teams.
-
A calm stretch is not a change: giving each segment its own volatility, and why it still failed its test
A scale per segment stops the drift model cutting volatility bursts into jumps (17 cuts to 2.4), yet fails its pre-registered held-out rule on TCPD.
-
Testing a DNG decoder on a Canon raw file: 22 million sensels, zero differences, twice as fast
ShootFlow's DNG tile decoder read a Canon CR2's camera-written stream exactly as LibRaw does, all 22 M sensels, and now runs 2.2x faster.
-
A carry-over score in one pass: scoring a round-robin starter over p numbers
A starter timetable's carry-over reduces to one difference sequence. Scored that way, a local search cuts 24 to 40 team round robins up to 7% further.
-
A four-week memory only pays when you score the closed hours
A last-4-weeks window in an automatic staffing forecast did nothing on three real bakeries until closed hours were scored. Then it cut hours short or spare 16%.
-
A French bakery shuts for lunch: what a second real till export taught the shift planner
A real bakery's 232,711 item lines broke the shift-from-sales tool four ways. It now reads item exports and leaves the lunch closure unstaffed.
-
The .gitignore line git never reads, and a ** rule that changed in git 2.52
A free tool that names the rule deciding every path, as git does. Testing it against two gits found that ** after text changed meaning in 2.52.
-
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 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 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 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.
-
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.
-
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.
-
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.
-
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.
-
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.
-
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.
-
Don't mirror your home-and-away season
Replaying the first half with venues swapped doubles every carry-over count. Searching the second half on its own takes 20 teams from 2,136 to 1,678.
-
Eight searches instead of four, and the wait got shorter
The league schedule generator now spreads its carry-over search over every core, so it runs eight searches instead of four and still finishes sooner.
-
Eighteen jobs, four cores: the changepoint detector's several-column mode on threads
Each column's six-setting changepoint panel now runs on worker threads in the CLI and on Web Workers in the browser: same answers, 8.8 s down to 3.6 s.
-
We ran the rule authors' own good invoices through our e-invoice checker
The rule authors' own good invoices for Dutch, Romanian, Portuguese and Saudi rules found a hang, a severity bug and two false rejections. All fixed.
-
Your smart meter already knows whether you need a panel upgrade
A new free tool reads a year of utility interval data, finds the highest 15-minute demand NEC 220.87 asks for, and checks it next to the 220.83 calculation.
-
Checking every cell of the IRS's earned income credit table, and the bug it turned up
The ai-bookkeeper now matches all 10,992 cells of the 2025 EIC Table, and the IRS worksheet exposed an undercount for low earners with other income.
-
File by January 31 and the January estimate's penalty goes away
The ai-bookkeeper penalty worksheet now applies Form 2210's January 31 rule: tax paid with a return filed by then counts for the January 15 installment.
-
Three currencies, one page: changepoints across several columns
The changepoint page now takes several columns. On franc, pound and yen it finds the 2015 Swiss floor removal in the franc and the Brexit vote in the pound.
-
Half the overnight grade was spent on clips nobody would see
ShootFlow can now give full-resolution ProRes masters only to the clips its storyline picks. On a 12-clip test card the grade stage went from 164 s to 88 s.
-
Half the wait: the changepoint detector's tests went from ten minutes to five
Typed arrays and an inlined cost in PELT's inner loop cut the changepoint detector's test suite from 599 s to 301 s, with every answer unchanged.
-
A robust loss that made our changepoint detector worse, and why
A biweight loss cut false changepoints on heavy-tailed synthetic noise, then nearly doubled them on bank and bitcoin prices: the tails are in the walk.
-
Walk or noise? A heavy-tail check for changepoints on a drifting level
The drift model now says whether a series' heavy tails are outliers around a level or the steps of a random walk, which decides whether its jumps mean anything.
-
We finally read the IRS's IRIS guidelines: three faults in our 1099 file, and a rent form
Reading the IRS's IRIS template guidelines found three faults in our 1099-NEC file. The AI Bookkeeper now also writes the 1099-MISC for rent.
-
Joining short shifts looked cheaper, and was not
Letting short staffing peaks share one shift cut the planned cost on paper. On 72 held-out comparisons at two bakeries it was dearer 16 times, cheaper once.
-
From 22 changes to 3: a noise floor per column for joint changepoints
The joint changepoint fit now floors the noise of smooth columns: German unemployment goes from 22 changes to three, with white-noise power unchanged.
-
Kaplan-Meier, log-rank and Cox regression in the browser, pinned to two reference libraries
A new free tool draws Kaplan-Meier curves and fits Cox models from a pasted CSV, and its tests hold every number to lifelines and statsmodels.
-
Half the back-to-back home games in an odd home-and-away league
The free league scheduler now builds home-and-away seasons for odd leagues with about half the back-to-back home or away games: 25 teams, 262 down to 117.
-
Odd-sized leagues: from 10 back-to-back home or away games per team to under 2
A 23-team league now gets 44 back-to-back home or away games instead of 127 at the same carry-over, from a home/away pattern read straight off the timetable.
-
Picking a round-robin timetable for its home/away pattern, not just its carry-over
A 26-team league can keep the same carry-over with 74 back-to-back home or away games instead of 102, by choosing a different starter timetable.