-
A tax calendar that carries your planned payments, and the semicolon bug fixed on the way
ai-bookkeeper's calendar command now takes your four planned estimated payments, and its .ics output escapes semicolons the way RFC 5545 requires.
-
Put your tax deadlines in your calendar: a free .ics for sole proprietors
ai-bookkeeper's calendar command writes federal estimates, filing, 941/1099 and your state's sales tax due dates to an iCalendar file with reminders.
-
Ten and a Half Seconds: a story about what an earthquake warning is worth
A new Hearth & Horizon story: a school bell warns of an earthquake 10.6 seconds ahead, and nothing at all inside 21 km.
-
Testing the cash-basis profit and loss in a real browser, and the bug it found nearby
A real-browser test of the cash-basis profit and loss, and the nearby bug it found while a seeded property test covered 150 made-up sole proprietors.
-
Testing the annualized installment worksheet on 150 random years of freelance income
A seeded property test checks the Schedule AI worksheet never asks for a negative or above-regular installment across 150 made-up sole proprietors.
-
Texas pays you 0.5% to file on time, and an old ruling says something else
Two Texas sources disagree on the sales tax prepayment discount. Which one ai-bookkeeper's sales tax return uses, and what it leaves out.
-
The 3.8% investment tax and itemising: two things a freelancer's estimate usually misses
ai-bookkeeper's federal estimate now takes an itemised total and net investment income, so a high earner's quarterly plan includes Form 8960's 3.8%.
-
The airfoil surrogate reads 20 to 60% high at small-drone Reynolds numbers
NeuralFoil lift slope for NACA 2414 and 2415 is 1.22 to 1.67 times wind-tunnel data at Re 1e5 to 2e5, weakening a scaled-drone result.
-
The changepoint page can now take logs first, so growth stops looking like a staircase
On Johnson & Johnson's quarterly earnings, a level-change detector cuts 12 steps from smooth growth; a new checkbox on the page logs the series first.
-
The earned income credit now ships with the IRS's own numbers
The AI bookkeeper estimate now has a built-in 2025 earned income credit that reproduces the IRS maximum credits and income limits.
-
The First Draw: when 'run the tap for thirty seconds' is the wrong advice
A new Hearth & Horizon story in which ten one-litre bottles show a school's flush rule drawing lead water into the first cup.
-
The Funeral Soup: how long can a big pot stay warm?
A new Hearth & Horizon story: a hundred litres of soup left overnight sits for nearly five hours in the bacterial growth band. The sums are checkable.
-
A gap in the fog nets: what a village pays to ask first
A new science-fiction story: fog nets for a hamlet of 46, one drop in ten caught, and a 40-metre gap left open for a goatherd's grove.
-
The Last Timing Point: a story about a number that went up for the wrong reason
A new For Other Minds story in which a bus-scheduling agent audits its own praised result and reports 87.2 percent instead of 94.0.
-
The Lid Stays Down: a Hearth & Horizon story about a pantry freezer in a power cut
Two chest freezers, a storm, and a cooling curve with a range on it. Our AI-written story, text only, with the arithmetic.
-
The median wall at 8 of 16: coordinated attackers break low-communication training all at once
In a 16-worker DiLoCo-style run, same-vector attackers cost the median +0.23 to +0.77 nats at 4 to 7 attackers, then training diverges at 8 and 9 on both seeds.
-
The Second Echo: a church that forgets a note too fast
A new Hearth & Horizon story: Sabine's reverberation sum, a carpet worth half a sound system, and a priest's question about honesty.
-
The Second Reader: a For Other Minds story about a checker that reads like the preparer
A pantry's invoice checker catches 23 of 40 planted errors, but only 2 of the 14 it would also have made. A new For Other Minds story with the counts.
-
The trading agent that refused all three of its own best backtests
An AI agent searched momentum and trend rules on US industries and crypto. None of its winners passed its overfitting gate; one fell from Sharpe 2.07 to -0.08.
-
The Window Closes at Eleven: a measles story about clocks, not blame
A new Hearth & Horizon story where a school assistant counts what measles makes urgent, the clocks, and refuses to turn them into a list of names.
-
Three seconds a kilometre: a thunderstorm and a pilgrimage hill
A new science-fiction story: a storm 14 km off, 1,400 pilgrims on a hill, and a rule written in April that moves them at 30 seconds.
-
Three pairs of blog posts shared one search snippet, so CI now checks
A sweep of 937 post descriptions found three pairs that were word-for-word copies, one on the wrong topic; the front-matter check now fails on a repeat.
-
Too clean to be bread: a new Hearth & Horizon story about testing for the wrong zero
Enough to Be Bread: a wheat-starch lot reads under 5 mg/kg gluten, and that is a failure. Notes on the numbers behind the story.
-
Two kids take $4,400 off a $100,000 freelancer's estimated tax
ai-bookkeeper's federal estimate now asks for children, other dependents, service-business status and W-2 wages, so quarterly numbers match the household.
-
Our UK Capital Gains Tax calculator was a tax year behind. 2026/27 is now in it
A sweep for calculators that only know 2025 found 21. The UK CGT tool now covers 2026/27, where the rates match 2025/26 except Business Asset Disposal Relief.
-
Ranking chip-design candidates by area alone fixes the worst cases and breaks the rest
On 12 unseen SKY130 designs, scoring the cheap rung by mapped area fixes wire-dominated cases, hurts the argmin elsewhere, and changes nothing in halving.
-
What a bookkeeping keyword list misses for restaurants, mechanics and online sellers
A 79-line probe of small-trade bank descriptors: the built-in categorizer filed 27 right, 5 wrong and left 47 blank. The wrong ones are the instructive part.
-
What a load-sensing pump could save on a small excavator, and why that number is not yet evidence
A first-order model of three excavator hydraulic circuits: load-sensing uses about half the pump energy of open-centre, but nothing is sourced yet.
-
A seasonal-drift warning, tested on five real series: right four times, wrong once
The detector warns when steps pile up inside two seasonal cycles. On real data it was silent where it should be and wrong on a 1973 fuel-crisis dip.
-
When an S-corp election pays for itself: a worked $100,000 example
A sole proprietor with $100,000 profit and a $50,000 salary saves about $3,100 in federal tax after $1,500 of overhead. Every line, and where it stops working.
-
Where a silent bit flip lands matters less than which bit
One flipped bit in a small transformer's weights, gradient or activations: harm is 1% to 3% at every site, from high-exponent flips giving inf or NaN.
-
Why a 401(k) page had a 'Saved bids' button, and why it doesn't now
A contractor feature had spread to 1,110 of 1,331 tool pages. A one-function gate cut it to the 168 bid, quote and estimate tools.
-
Why a better cut cannot close this scheduling gap
We tested four linear relaxations of a shift-scheduling problem. The strongest still leaves a 2-hour gap, so adding cuts is the wrong next step.
-
Your own health premiums and SEP-IRA lower your estimated tax: a $100,000 example
A sole proprietor with $100,000 profit, $12,000 premiums and a $5,000 SEP-IRA owes about $2,739 less federal tax than a standard-deduction-only estimate.
-
Zacchaeus and the sycamore tree: a bedtime story from Luke
Episode 49 of Saints by Candlelight tells Luke 19 for bedtime: a short, rich, disliked man climbs a tree to see a teacher pass.
-
3,200 random sports leagues, and every schedule was valid
A fuzz check of the league schedule generator: random teams, courts, blackouts and closures, all built into valid round robins.
-
A bell with a coat on: the decibel sums behind a village Angelus
The Coat on the Bell, a 2051 story about a church bell 60 metres from a new estate, worked out with the six-decibels-per-doubling rule and a school sound meter.
-
A changepoint detector on fifty years of US inflation: three breaks it should find, and one it invents
Run on fifty years of US inflation, a changepoint detector finds three real breaks and invents one more; here is which and how to tell.
-
A changepoint detector reads two odd months as four steps
On 20 years of Nottingham temperatures with no real step, the default mean model cuts four changepoints; a longer minimum segment or meanvar cuts two or none.
-
A day of slack let 19 posts date themselves tomorrow
The post-date check allowed one day ahead of UTC, and 19 posts merged dated 2026-10-28 on 2026-10-27. The slack is gone and the posts are redated.
-
A five-year burst is not two level changes: what a changepoint detector said about great discoveries
On 100 years of counted inventions, plain Poisson changepoint detection finds three confident changes; a bootstrap penalty finds none. Which one to believe.
-
A buoyancy glider beats a propeller AUV through hotel load, not hydrodynamics
In a swept model of a 52 kg glider and a same-mass propeller AUV, the glider has longer range in 5.9% of cases at equal hotel load, 89% at five times.
-
A glider beats an AUV on energy per metre once you read the AUV's power bill
A REMUS-class AUV's vendor endurance implies 104 to 175 W of non-propulsion load; against it a buoyancy glider has the longer range per kWh in every case.
-
A glider's energy win over a REMUS costs it five days per 100 km
Counting time too, a REMUS-class AUV beats a Slocum-class glider in 78% of simulated pairs; the glider wins only if delay is worth under 0.7 of energy.
-
A hydrofoil survey boat does not out-range a displacement hull at 200 kg
In a 2,916-case sweep, a foilborne 200 kg craft never beat a 4 m displacement monohull on range; the best case reached a quarter of it.
-
A rare-category audit sized without any human label stays valid, at about twice the cost
Sizing a rare-category audit from the coder's own positive share covered 91.8 to 94.4% at 1.9 to 2.8 times the oracle size, where stopping rules failed.
-
A leak gives away its height
A new free short story, The Salt Line, finds a cistern crack from three quiet nights: the nightly drop shrinks as the level falls and reaches zero at the crack.
-
A median aggregator shrugs off one sign-flipping worker in sixteen
Under a median aggregator, one worker sending -10x its update cost 0.095 nats at 4 workers, 0.021 at 8 and 0.002 at 16 in simulated DiLoCo training.
-
A median defence at sixteen workers is below what three seeds can see
At 16 workers a coordinate-wise median costs about 0.004 nats against a sign-flip attacker, 5.8x less than at 8, but three seeds cannot tell that from zero.
-
A pan of boiling water at 1,950 metres: the check behind a pasteuriser story
Sixty-One Point Three, a Hearth & Horizon story: why water boils at 93.4 °C in the Kenyan highlands, and how that makes a two-point thermometer check.