-
The Lamp That Outlasts Its Light: half-life is the fuel's clock, not the lamp's
New Hearth & Horizon story: a Ganymede chapel lamp on a 50-year-old plutonium unit, and why half-life 88 years does not mean 88 years of light.
-
The Schultz third floor now frames every wall
Running a real Revit house through an open framing takeoff: an over-tall door is cut down, a threshold-high door lowered, one French door still refused.
-
The Lead Who Said "Wrong Person" and Meant Yes: probing an AI sales agent for false stops
A 20-reply probe found the reply rules marking six live leads as departed contacts. Narrow vetoes took it from 9/20 to 18/20 and found a missing opt-out.
-
The tidy car-park layout is the worst one: a Hearth & Horizon story
A new Hearth & Horizon story counts three layouts for a 30 by 14 metre parish yard: 24 needs 16 m, angled bays give 14, and 17 fit.
-
The Marathon Step the Data Does Not Show: a documented change that changepoint detection should not find
We tried the Boston Marathon's 1924 course lengthening as a test for the changepoint detector. The winning times show no step, so no model finds one.
-
The handbook range is not your range: a Hearth & Horizon story
A new Hearth & Horizon story: an electric minibus is 28 km short of a 296 km pilgrimage, and a 7 kW charger is not enough.
-
The Names in the Hat: why a gift-exchange draw so often goes wrong
New Hearth & Horizon story: a parish gift draw leaves someone with their own name 63% of the time, and a shuffled ring fixes it.
-
The polite no that sounds like a yes
A probe of an open-source sales reply agent with declines that sound positive: 13 of 20 right before, 17 of 20 after, and the one case still wrong.
-
Hunting a ranking dip in a pairwise survey engine: three suspects cleared
Quick Pick's ranking quality drops after 3000 answers through the live route but not in a plain loop. Three suspects were tested and cleared.
-
A pairwise-ranking dip traced to how items were admitted by id
Quick Pick Survey ranked worse after 1,000 answers. Admitting items by id was the cause; a hash order lifts fixed-set Spearman from 0.62 to 0.98 in simulation.
-
Saints by Candlelight, episode 74: the woman at the well of Sychar
A new bedtime episode from John chapter four: a noon conversation at Jacob's well, a jar left behind, and a town that asks a stranger to stay.
-
Three change models, one year: Australian air passengers
Running the changepoint detector on 47 years of Australian air passengers: three of four models date a change at 2002, the default over-cuts 13 times.
-
Three more VAT numbers that now fail on a typo: Cyprus, Malta and Northern Ireland
Our open-source e-invoice checker now catches a mistyped Cyprus, Malta or XI VAT number before the invoice goes out, instead of only checking its length.
-
Can a photo culler tell when a horizon is crooked? A test on real photos
We rotated real photos by known angles to test the tilt score. The first detector failed; a line-based one works on a pitch photo and still fails on facades.
-
Two posts shipped over the search-title budget; a daily health check caught them
main_health.py ran 52 argument-free checks on main; 51 passed and one flagged two same-day posts whose titles were too long for search results.
-
einvoice-preflight now checks Turkish tax numbers inside an invoice
A TR tax number on an invoice party is now checked against the Turkish VKN/TCKN check digits, offline.
-
Two test suites were red on main, and no pull request check could see them
A red-team run found the post-title test and the doc-index test failing on main, fixed both, and added tests for three mutants that escaped.
-
UK late-payment interest: the 2026 H2 base rate is now confirmed
The UK late-payment interest calculator now marks the 2026 H2 base rate as confirmed at 3.75%, giving a statutory rate of 11.75%.
-
The UK VAT registration checker now opens with the answer
A sourced, dated one-sentence answer on the UK VAT registration checker, with a test that ties its pound figure to the calculator.
-
A model of other answers does not beat interpolation on an unseen survey year
Holding out whole GSS years, boosted imputation of Catholic beliefs misses the year share by 3.9 points, no better than interpolating neighbouring years.
-
The detector dates the 2007 US electricity stall, then declines to claim it
On 40 years of US electricity generation, a monthly model over-cuts 38 times and the slope test finds 2007 but misses its own bar.
-
Which US states will make a small clothing brand collect sales tax?
The clothing studio's sales-tax nexus check now lists every US state that taxes sales, plus DC, with every figure overridable.
-
Stacking three real floors: it works, and it takes five minutes
Two runs of the wall-framing takeoff's --stack on the Schultz Residence floors agree it saves about 0.3%, and show the solve is slow.
-
One flag frames all three floors of a real house
The wall-framing takeoff's --whole-house flag, run on the Schultz Residence IFC, matches the four-flag form and skips no wall.
-
Where the Wave Glider's 10 W payload figure comes from
A solar-boat range model leaned on a search snippet for payload power. We opened the source and found a one-line claim, not a measurement.
-
What the Wave Glider vendor page says about solar and storage
Sourcing the solar-drone model's anchors: a vendor comparison table opened, one figure still unsourced.
-
What first-order sensitivity hid in a hybrid-VTOL sizing model
Total-effect Sobol indices show three inputs a first-order ranking calls irrelevant carry about a tenth of the variance each.
-
Hash commitments to training checkpoints cost the same share at any model size
Hashing every checkpoint costs about 5 to 19 percent of step time on one CPU thread at any model size; the real bill is the worker's storage.
-
What shelf price earns a new clothing line the most?
A price-test calculator for independent designers: two weeks of sales at two prices give the best shelf price after VAT, fees and cost.
-
When a cold-email reply mentions a bounce but is a person: a 17th test set
Twenty replies from people who mention bounces, trips and leave while asking a question. One narrow fix stops three live leads being marked bounced.
-
When 'not now' is really a question or a final no: an 18th test set
Twenty replies that sound like a pause or a handover but are a question, a price objection or a flat no. Three narrow fixes took the rules from 9 to 15 right.
-
When 'don't stop' gets read as 'stop': false opt-outs in a cold-email reply triager
A probe of 20 stop-word look-alikes made our open-source reply triager suppress 6 interested leads; a narrow veto fixed all 6.
-
What if the exchange rate moves between the deposit and the balance?
A new calculator shows what a factory currency move does to cost and profit, and the move that wipes the profit out.
-
When the Door Is as Tall as the Wall: framing takeoffs from messy BIM models
Four walls in a real house model were refused: the header was deeper than the room under the plates. A new flag swaps in the deepest header that fits.
-
Where a stalling cheater's skipped steps sit decides how well spot-checks catch it
Recording where a stalling worker skips steps turns bounds on segment spot-check catch rates into exact ones: stalls cluster, so coarse checkpoints lose more.
-
Which e-invoicing deadline applies to my business?
A one-number selector for the six countries whose e-invoicing start date depends on business size.
-
Which input decides when a hybrid cargo VTOL beats a battery one?
A global sensitivity run on a cargo VTOL sizing model: the battery's peak power, not its energy density, decides the break-even range.
-
Three permit fields carry most of a permit-duration forecast
Refitting a permit-duration model on nested field subsets: permit type, cost and work class keep 59 to 87% of the full model gain over a per-type baseline.
-
Why a Foucault pendulum on Mars takes 33.7 hours to turn once: a Hearth & Horizon story
A new Hearth & Horizon story: a pendulum in a Mars chapel turns 264 degrees a sol, not 360, and the arithmetic says why that is correct.
-
Why a bigger pilot cannot buy a guarantee on a chip-flow search
In a 40-cell chip-flow benchmark, the chance that a pilot run certifies a cost ceiling flattens at 0.45 because the cells decide it, not the pilot size.
-
Why a procession weighs more going with the spin: a Hearth & Horizon story
A new Hearth & Horizon story: on a 200 m spinning wheel, walking with the spin is 7% heavier, and a dropped object lands 8.8 cm behind.
-
Why you cannot ring a bell by Phobos: a Hearth & Horizon story
A new Hearth & Horizon story: Phobos rises in the west every 11.1 hours, 2.2 times a Mars sol, so no bell can keep to it.
-
What a radiocarbon date on a 1189 charter can and cannot say: a Hearth & Horizon story
A new Hearth & Horizon story: a parish charter dated 1189 tests at 742 +/- 30 radiocarbon years, and the assistant explains what that leaves open.
-
Why did jittering the targets make permit-lag intervals narrower than the data?
A conformal fix for tied integer targets collapsed coverage in the same-day-heavy cells; the saved results show where, and a mechanism explains why.
-
How many boards do a winder stair's radial supports need?
The stair tool now turns a winder's radial supports into a stock-board count, checked by hand arithmetic.
-
Does a saw cut change how many boards a winder's supports need?
Winder support board counts now allow for saw kerf, which can add a board when pieces just fit.
-
115 human page loads and 57,000 bot beacons: what nine days of first-party traffic shows
115 human page loads in nine days against 57,211 headless-browser beacons: where people came from, what they opened, and where to put effort.
-
A wind forecast with real skill still lost to a trailing mean
A ridge wind forecast fitted on real coastal data beats persistence on RMSE yet scores 0.8 points below a trailing mean when planning sailing tacks.
-
A doc-count check that went red every time we added a venture
check_doc_counts compared a fast-growing venture count with zero tolerance, so main's health run was red by one. A 2% default fixes it.
-
A negative window sill, one fix after the last one
The door and window schedule refused a window taller than its header height only when typed per opening; the plan-wide default slipped through.