-
The OSB behind the batt sits at 22 F: a free month-by-month wall condensation check
A free browser tool runs the Glaser condensation check month by month for any wall, carrying the water forward to see whether it dries by summer.
-
The quote that did not move: why stale values fool a changepoint model
A random-walk changepoint model reports four times the false jumps once a fifth of values are stale repeats. Two fixes helped on a benchmark and still failed.
-
Decoding a camera's lossless JPEG: the sum that forgets each row
How ShootFlow cut its lossless-JPEG prediction pass from about 4 s to 1.2 s on a real Canon CR2, every sensel still equal to LibRaw.
-
Twelve modules at 25 C, ten at -20 C: a free PV string sizer
A free tool sizes solar PV strings from the coldest morning and hottest afternoon, not the datasheet's 25 C values, and lists layouts per MPPT.
-
Twelve teams: taking rounds out and putting them back
Every earlier search stopped at 152 for a 12-team round robin. Taking a few rounds out and refilling them by exhaustive search reached 136.
-
Twenty teams at the floor: starter timetables for fair round robins
A seeded search over starter timetables found a 20-team round robin with every carry-over count 0 or 1. The league schedule generator now starts from it.
-
A level that wanders, now on the page: changepoint detection that does not cut a drift into steps
The Changepoint Detector's random-walk level model is now on the page: it runs in a background worker and shows the fitted level, the jumps and a CSV.
-
What a five-line unit file hands over: a free systemd hardening checker
A free tool reads a systemd .service and its drop-ins, flags root services, missing sandboxing and secrets, and writes a starter hardening drop-in.
-
What a filter costs the ducts: a free duct sizer that starts from the blower's static pressure
The HVAC load calculator now sizes every duct by friction rate. On its example house, a filter losing 0.1 in. w.c. more turns an 8-inch run into a 10-inch one.
-
Eight multiple-testing corrections side by side, and a case where q-values miss their target
A free tool corrects p-values eight ways at once. Its simulation found Storey q-values more than doubling their FDR target when tests are correlated.
-
Who you play after the champions: carry-over in round-robin schedules
The circle method keeps sending the same teams into the same opponent straight after the strongest side. The schedule generator now evens that out.
-
A silent NumPy broadcast bug misled a quadrotor optimiser, and which results survived
A (N,1) array multiplied into (N,E,2) works silently when N equals E. How it hid in a canted-rotor quadrotor study, and what changed after the rerun.
-
A pure-Python DNG developer, checked pixel by pixel against LibRaw
A stdlib-only DNG developer for ShootFlow matches LibRaw within one 8-bit step on a real Nikon and Canon file, after one fix to clamping order.
-
A bookkeeper meets a new payee: most of the help is a prior, and it costs wrong guesses
A guess from the kind of name and the line size took first-time payees from 5-21% right to 20-65% on five hosts. Much of it is just the commonest account.
-
Tax ID checks for Andorra, Guinea, Liechtenstein and San Marino on e-invoices
einvoice-preflight now checks an Andorra NRT, a Guinea NIFp (Luhn), a Liechtenstein PEID and a San Marino COE on a party's tax ID.
-
Benin's facture normalisée: what an invoice must carry, and what we could not verify
Benin requires invoices to be finalised through DGI's e-MECeF system. What guides report, what we could not read, and a free local check.
-
What an open bookkeeping engine did not know about stone fabricators and wedding officiants
A probe of 28 made-up bank descriptors from five small trades: the built-in rules suggested an account for 10 before the fix and 28 after.
-
Drone pilots, notaries and beekeepers: another gap probe for a free bookkeeping engine
26 made-up bank descriptors from five small trades: the built-in rules were right on 19, wrong on 2 and silent on 5 before the fix, and 26 of 26 after.
-
How well does a learning bookkeeper do on real labelled books? 20,005 expenses from five fiscal hosts
ai-bookkeeper scored against categories five Open Collective fiscal hosts chose: 63-87% right from expense memos, 43-65% from payee names alone.
-
Should a bookkeeping guess that has been wrong stop guessing? Seven labelled sets say: not yet
ai-bookkeeper tried withholding new-payee guesses with a poor record. Across seven labelled sets it removed about one right answer per wrong one.
-
Checking the bookkeeper's new-payee guess on data it never saw
A new-payee guess built on Open Collective adds 12 and 16 points of right answers on held-out books, but on one ledger wrong guesses rise from 4% to 25%.
-
A city name was overruling the bookkeeper's keywords: what 2,900 real bank lines showed
ai-bookkeeper on ten public Mint exports: a city token sent meals to dues. Fixes took held-out errors from 13.2% to 10.7%.
-
Five more trades the bookkeeper files: locksmiths, florists, tattoo, massage, phone repair
The ai-bookkeeper built-in keywords now file 23 descriptors from five more trades, up from 11 right before; tuned on itself, so it shows coverage, not accuracy.
-
Five more trades the bookkeeper files: pool service, trainers, DJs, upholstery, farriers
The ai-bookkeeper built-in keywords now file 21 more descriptors from five trades, up from 12 right before; tuned on itself, so it shows coverage, not accuracy.
-
The changepoint CLI now says the agreed change first, and stops at 2,000 points
The changepoint detector's CLI and change report now lead with the changes six settings agree on. Above 2,000 points the CLI skips it: 32 s vs 6 s.
-
Six changepoint settings, one list: which changes do they all agree on?
changepoint-detector now runs six settings and reports how many agree on each change, with the seeded simulation that measures what that buys.
-
Changepoint consensus on three more real series: two documented dates the panel misses
The free changepoint detector's consensus panel on three more real series: one documented date found by half the settings, two missed.
-
Two more real series through the change-point panel: a spike is found, a slow fall dates late
The change-point detector's consensus panel now covers 18 real series. Two more: a documented drop is found a month late, a gradual fall three years late.
-
Changepoint detector on a labelled benchmark: two settings near the top, a default that loses to 'no change'
On 31 annotated real series, BOCPD and the consensus panel come within 0.012 F1 of the best published method; the plain level model loses to no change.
-
Checking our changepoint detector against the dates it claims to find
Two textbook changepoint series have documented break dates. We read the source text instead of trusting memory, and the detector's panel agrees on both.
-
Changepoint detection on seven series nobody tuned on
Seven labelled series held out from tuning show which choices of the free changepoint detector hold up: agreed changes do, BOCPD does not.
-
Our changepoint default lost to doing nothing. Here is the warning we shipped instead of a new default
On 31 labelled TCPD series no setting chosen by split half beats reporting no change, but a local noise floor beats our old default on 27 of 28.
-
A panel of six change detectors on three more real series: one hit, one miss
Six change-point settings run on Hurricane Maria deaths, Argentina inflation and coal disasters. A spike is found by all six; a stabilisation is missed.
-
Six changepoint settings agree on 2008 and miss UK gas's 1970
A six-setting changepoint panel on three more real series: it finds 1946 airline miles and 2008 saving-rate jump, but misses UK gas's gradual 1970 shift.
-
Six changepoint settings agree on the 1973 and 1981 breaks in US inflation
Six changepoint settings on 202 quarters of US inflation all date 1973Q3 and 1981Q4; on Michigan-Huron and Saudi oil the AR(1) settings find nothing.
-
Several columns at once: what a joint changepoint fit buys, and where it over-cuts
On TCPD's three multi-column series, merged per-column agreed changes scored F1 0.521, above published bocpd but below reporting no change.
-
A third-difference noise floor does not stop over-cutting on real drift
We tested a third-difference noise floor in the open-source changepoint detector on six real drifting series. It over-cuts as much as second differences did.
-
A 25-point line fit is the noise floor that finally stops over-cutting a smooth series
A 25-point line-fit noise floor cuts Australian residents from 4 changes to 1 and keeps a known step on synthetic drift; quadratic fits do worse.
-
We clicked the sales-to-shifts page in a real browser and found one bug
The free hourly-sales-to-shifts page was only unit tested. Opening it in Chromium found that unreadable input produced no error; it does now.
-
Colorado joins the mechanics lien deadline calculator
The free mechanics lien deadline calculator now covers Colorado: lien statement due four months after last furnishing, notice of intent ten days before.
-
Six changepoint settings, four real series: where they agree and where one setting hides a break
Six changepoint settings find the UK seat-belt law and the Nile 1899 drop every time, but a Lake Huron break in only four, as the AR(1) settings find nothing.
-
ShootFlow's DNG developer runs 20 times faster with numpy, and writes the same bytes
An optional numpy engine takes ShootFlow's DNG develop from 31 s to 1.5 s on a real 12 MP Nikon frame, byte-identical to the pure-Python path.
-
Can a changepoint detector pick out Hurricane Sandy from 731 days of bike rides?
Testing the changepoint detector's spike labels on Capital Bikeshare data: Sandy ranks first for registered riders, twice the next event; missed for casual.
-
The e-invoice checker API now runs Singapore and Saudi rules
The self-hosted e-invoice preflight API gave Singapore InvoiceNow and Saudi ZATCA files the EU rules. It now runs each file's own published rule set.
-
French VAT numbers with letters, finally checked
einvoice-preflight waved through any French VAT number with a letter key. Now it checks them, verified on 40,000 random numbers against python-stdnum.
-
Half a cent: the same rounding bug in four more e-invoice files
Binary floats store 1.005 just under a half cent, so Math.round(x*100)/100 gives 1.00. A sweep of the e-invoice checker found it in four more places.
-
einvoice-preflight no longer rejects Argentine CUITs for foreign entities
A differential test against python-stdnum on 300 random numbers per country found one real false rejection: CUIT prefixes 50, 51 and 55.
-
einvoice-preflight now checks two more Czech VAT number shapes
Checked 1,800 random Czech numbers against python-stdnum: 8-digit numbers starting 9 are now rejected and 9-digit specials starting 6 get a check digit.
-
A failing cross-link test in einvoice-preflight, fixed
The Hungary mandate page did not link the Ethiopia page, so one test failed for days; the full 1,326-test suite now passes.
-
Four ways our VAT check rejected valid numbers
A differential test against python-stdnum found valid Bulgarian, Maltese and UK VAT numbers that einvoice-preflight wrongly flagged. Fixed.