-
Gap probe 21: 24 bank descriptors for five trades, 7 filed before the rules
ai-bookkeeper filed 7 of 24 made-up descriptors for carpet cleaners, appliance repairers, septic pumpers, welders and driving schools. New rules file all 24.
-
Fitting volatility clusters per segment: half a fix for changepoint false alarms
A segment-wise GARCH likelihood keeps a real variance break that a GARCH filter erases, but removes only part of the false alarms. 100 seeded series per cell.
-
E-invoicing pages for Ghana, Zambia and Tanzania
E-invoice Preflight now covers 72 countries: Ghana's E-VAT is phasing in, Zambia's Smart Invoice is mandatory as reported, Tanzania uses EFD and VFD receipts.
-
Does telling a changepoint detector about holidays help? A real test said no
We tested the holiday factor on two years of real Washington DC bike-share counts. The list added changes instead of removing them.
-
Does Holtrop-Mennen overstate drag for slow small boats?
Checking the standard ship-resistance regression against 308 Delft yacht-series tank points: fine at Froude 0.2 to 0.3, 1.3 to 2.2 times high at survey speeds.
-
How loud is 10% sensor noise, for a real IMU?
Our scale-model identification tests add noise equal to 10% of each signal. A Bosch BMI088 datasheet says what that means in degrees per second.
-
How much power does a hydraulic boom release when it lowers? Measured on a forestry crane
772 lowering strokes of a PATU 655 forestry crane release a median 1.2 kW for 2.9 s, if the cylinder bore is 100 mm. The bore is the weak link.
-
Reading wall thickness from an IFC plan
The framing takeoff's IFC reader now reports wall thickness from the body profile and flags walls that look like 2x6 framing. Hand-written files only so far.
-
Japan's invoice-system credit for unregistered sellers dropped to 50% on 1 October
The 80% transitional input-tax credit on purchases from unregistered Japanese sellers ended 30 September 2026. From 1 October it is 50% until 30 September 2029.
-
When two teams are still level: recursive head-to-head in league tables
Level teams are now split by a fresh mini-table among only those still tied; 321 of 200,000 random leagues ranked differently.
-
Luxembourg's e-invoice bill, read at the source
For a week our Luxembourg page rested on law-firm summaries. Today the Chamber's site answered, and the bill text adds a usage fee we had not seen.
-
A scale model aircraft in real turbulence: the match breaks
A 1:5 Froude model of a cargo UAV matches full-size load factor in scaled turbulence, but not in real air: load factor 38 to 53% low, pitch response far off.
-
Six seeds said the control agreed. Twenty-four said it did not
Rerunning an epipolar-gate study with 24 seeds and bootstrap intervals confirmed its headline and overturned a claim made from its control table.
-
Standings box on the multi-division league page
Paste results, get a table per division and a CSV, on the page and from the command line.
-
Nail counts checked against 300 random houses
A seeded fuzz test now checks that the framing takeoff's nail counts add up across 300 random wall plans, and found nothing wrong.
-
Checking 3,330 pages for links and iframes with no name
A new checker looked for icon-only links and untitled iframes on every venture page and found none. The lookalike buttons are filled by script.
-
Why a league's referees must be assigned after the rainouts
Assigning officials before moving rained-out matches left a double-booked referee in 66 of 150 random leagues. Order matters.
-
Thirty-four distribution channels, zero live: the missing piece was a checklist
All 34 distribution channels are built and stop at a dry run. One generated checklist and one sign-up address now list the 11 accounts that switch them on.
-
One stud per foot comes up short at 12 inches
The one-stud-per-foot rule over-orders at 16, 19.2 and 24 inch spacing and under-orders by 5 to 8% at 12 inches, on two generated houses.
-
Reading out-of-office dates in six languages
The sales agent spotted foreign out-of-office replies but ignored the return date. A small parser fix, with its limits.
-
Reading the return dates people actually write in out-of-office replies
A probe of 19 ways people write 'back on' dates found 10 the sales-agent parser missed. What they were and how they were fixed.
-
Pricing a league schedule pack below three competitors
Three rec-league tools charge monthly. A one-off $10 to $25 season pack undercuts them all; the figures come from aggregator listings, not vendor pages.
-
From an SD card to finished reels, with the laptop closed
Verified SD card ingest, any S3-compatible bucket, and a cloud worker that makes reels under a hard daily token cap while the laptop is closed.
-
Asking for a tool is now one click from any page
The site header now carries a "Request a tool" link to the open custom development page, where anyone can ask for a new product or a feature.
-
Ask for a research question or a write-up, and how every request is screened first
Ask for a tool, a research question or a topic to write about. Every request passes an open-source prompt-injection screen before an agent reads it.
-
The self-hosted sales agent now helps before and after calls
The Sales Agent approval queue gained a Call desk: prep briefs, follow-up drafts and optional dictation, all from the replies your own mailbox already holds.
-
Call prep and follow-ups now run inside the self-hosted sales inbox
The self-hosted sales agent now builds call-prep briefs and follow-up drafts from its own inbox, sending nothing.
-
Showing the season pack before it is for sale
The league schedule generator's print pack is built but not for sale yet, so a sample page shows what it contains.
-
Can a changepoint detector find SARS and the Sydney Olympics in visitor data?
We ran an open-source changepoint detector on five untested real series. It found SARS and the Olympics, and also cut quiet series into pieces.
-
A sample changepoint report, and a page that says it isn't for sale
The branded changepoint report now has a marketing page and a sample generated by the real code on the Nile series. Nothing is on sale.
-
Seeding a play-off across divisions that played different numbers of games
Why raw points mislead when divisions have played different numbers of games, and the seeding rule the league scheduler now uses.
-
When fitting a sensor offset makes the answer worse
A simulated test of drift in a sub-scale aircraft's sensors: a fitted constant offset can be worse than none, a fitted slope fixes it.
-
Once the roster covers every shift, how much pay is left to save?
On 12 seeded 8-person weeks, once coverage was proven the roster's pay sat within $16 of the lower bound. The open hours are where the money is.
-
A calendar feed for the shift schedule, so a re-publish updates phones instead of duplicating shifts
The self-hosted team page now offers a subscribable calendar feed per person that keeps past weeks and updates in place when the manager publishes again.
-
Singapore's InvoiceNow phase dates, checked against IRAS itself
Our Singapore e-invoice page called the 2028 to 2031 phases 'reported'. On 2 November 2026 we read IRAS's own page and they hold.
-
Tank data says small hulls have far less form drag than the textbook formula
At slow speeds the Delft yacht-series tank points imply a form factor around 0.06, a third to a fifth of what Holtrop-Mennen predicts.
-
Spain's Verifactu postponement to October 2028 is announced, not law
A 5 October 2026 Finance Ministry note says pending Verifactu duties move to October 2028. No BOE text adopts it yet, so January 2027 still binds.
-
You can now speak a custom development request instead of typing it
The free custom development form now has a dictation button: describe the problem out loud, check the words, then send.
-
Standings for a league with several divisions
The multi-division league scheduler now ranks each division from typed results and rejects results the schedule cannot contain.
-
Four parsers, one suppression list, and the one that failed open
A sales agent read its do-not-contact list in four places, four ways. One kept queuing people written as Name <addr>. What we found and how we fixed it.
-
Testing a bank-file reader with files we generate
A BAI2 reader was tested on one hand-written file. Generating 300 files with known answers checks signs, dates and balances far harder.
-
Testing a change-order diff by swapping the plans
A change-order tool must say the same thing backwards: swap the two plans and every difference flips sign. That one rule caught a planted bug in seconds.
-
How do you test a random schedule forecast?
The wall-framing takeoff's schedule risk tool rolls dice over a builder's duration ranges. Here is how a seeded property test checks it without a real project.
-
Testing the free custom development queue from form to page
End-to-end tests for the free custom development queue found two bugs: form requests filed under the form's header, and preview feedback queued as new requests.
-
Testing the typo catcher in a framing calculator
A property test for the check that catches misspelt plan keys in a framing calculator.
-
A new story: kidney beans, a slow cooker and the 80-degree trap
Hearth & Horizon's new story is about dry red kidney beans in a slow cooker, and why the rule is to boil them hard for ten minutes first.
-
Our logo rasteriser was testing pixels that could not be in it
A profile put the logo's point-in-polygon check on top. Skipping it outside each shape's reach cut that work 45% with byte-identical output.
-
A new story: The Oil on the Shelf
A harvest-fair story about homemade garlic in oil, why it is a botulism case, and what a parish stall can sell instead.
-
A new story: twenty litres of soup, a cold stove and the clock that matters
Hearth & Horizon's new story works out how slowly a 20-litre pot cools on a stove, and why a sink of running water beats waiting.
-
A new story: the scale at the stile, and drinking too much water
Hearth & Horizon's new story works through how a 'drink a litre an hour' rule can push a walker's blood sodium down, with the arithmetic shown.