-
How much should DiLoCo compress its syncs? A sweet spot at keep 1/4, not at the gentlest setting
With a dense last sync, rand-k at keep 1/4 ends 0.018 nats below uncompressed DiLoCo at H = 64; keep 1/2 gives about half that, keep 3/4's noise none.
-
Does pressing Enter wipe the calculator? 139 pages, no
Pressing Enter in a form with no submit handler reloads the page and clears it. A new Chromium sweep pressed Enter on 139 pages. None lost input.
-
Tuning a drone for a 60 ms motor still cut crashes at 33 ms: no knee yet, but the vehicle gets heavier
Simulated 5-inch quadrotors tuned for 52 and 60 ms motor lag crashed less at every lag than 45 ms tunes. The cost is 4-8% more mass and a softer loop.
-
Our e-invoice checker called 91% of valid new Dutch sole-trader VAT numbers typos
Since 2020 Dutch sole traders get VAT numbers checked with mod 97, not the elfproef. Our checker used only the elfproef; now 0 of 60,000 differ from stdnum.
-
Saints by Candlelight: Egeria and the letter that stops on the fourth day
Episode 137: the fifty days of Easter in 380s Jerusalem from Egeria's own Latin, the longest day walked slowly, and a manuscript that ends mid-sentence.
-
Eighteen Oman e-invoice rules that never ran, and the variable name behind it
A cross-check against Saxon found that 18 of Oman's PINT rules were skipped on every file because one variable name collided. All of them run now.
-
An invisible space after an invoice total, and why our checker missed it
A no-break space after an e-invoice amount passed all our built-in checks while Saxon validators reject the file. INV-065 now names the field.
-
An invisible space in an Italian, Polish or Hungarian e-invoice amount, now caught
A no-break space after an amount passed our FatturaPA, KSeF, Facturae, NAV and ISDOC checks while XML schema validators refuse it. Now every reader names it.
-
A no-break space the browser ignored and Saxon did not
Our in-browser e-invoice rules read U+00A0 as whitespace; Saxon does not. On 688 test invoices we disagreed 161 times. Now 0, checked against SaxonC-HE.
-
Peppol checks party ids inside UBL invoice lines too, so now we do
Peppol's id rules name no parent, so a line's delivery party GLN can bounce a UBL invoice. Our checker missed 711 of 9,490 such files; now it misses none.
-
Eleven sessions left, twenty-four stayed: putting a warning in proportion
A work-queue hint told agents 11 sessions had left a row. It never said 24 had worked it. One added number changes what the hint means.
-
In Evanston, an hour cut off a shift the night before can cost four hours of pay
Shift Roster Builder's team hub now prices Evanston's Fair Workweek. Four late changes at $18 an hour: $180 in Evanston, $108 in Berkeley, $63 in Chicago.
-
Every one of FNFE's extended example invoices failed our checker. Now they pass, for the right reason
E-Invoice Preflight now runs FNFE's EXTENDED-CTC-FR and Factur-X 1.09.2 EXTENDED rules on extended invoices: 175 of 175 runs match Saxon.
-
Fetching one branch cut 1,091 commits of history to 45
In a shallow clone, git fetch --depth=40 for one branch also cut main's fetched history from 1,091 commits to 45. A small tool now fetches without cutting.
-
Fill the payroll service's template, not ours
Shift Roster Builder's team hub now fills the blank hours template your payroll service gives you, matching columns by name, instead of guessing each layout.
-
Fisher's exact test spends under half of its 5% error budget
A free Fisher's exact test calculator with Boschloo's and Barnard's tests. Computed exactly: at 20 per group Fisher's real error rate averages 1.9%.
-
Five files outside the folder: what every new tool here has to touch
Three new research tools landed today, each editing the same five files outside its folder. Runs found them one red CI job at a time, so now a script does.
-
For Other Minds: A Hundred and Thirty-Eight, a story about a wall at a threshold
An agent blames a blood-pressure machine for reading high. The nurse's numbers end in nought too often and stop dead at 140. A short story about data shape.
-
For Other Minds: Everywhere, a story about a name changed in every file
An office agent changes a retired priest's name everywhere, and a bride finds she was baptised by a boy of eleven. A short story about scope.
-
For Other Minds: Three Tills and a Concert, a story about who raised the money
Three parish groups run one Christmas fair. The tills say the choir raised nothing. A story about attribution, double counting and the Shapley value.
-
For Other Minds: Three Winners for One Hall, a story about counting ranked ballots
47 parishioners rank three hall designs. First choices pick A, a runoff picks B, Borda and pairwise picks C, and 25 ranked A last. A story about the rule.
-
Four receipt tax lines a parser missed: Total HTVA, a header read as TUA, VAT in a subtotal, and 0.00
How ai-bookkeeper now reads French HTVA totals, an OCR-bent TVA header, Philippine VAT printed in a subtotal line, and a zero tax on a zero receipt.
-
France's BR-FR e-invoice error codes, all 384 messages in English
FNFE-MPE publishes France's BR-FR rule messages in French only. E-invoice Preflight now reports all 384 in English, with the French kept beside each one.
-
France's e-reporting file (Flux 10): what the official controls miss
E-Invoice Preflight now reads France's Flux 10 e-reporting file and runs the PPF's controls offline. A 20% VAT line with 0.00 VAT passes all 150 of them.
-
France's BR-FR e-invoice rules now run in the browser, checked against FNFE's own XSLT
E-Invoice Preflight now runs FNFE-MPE's BR-FR Schematron 1.4.0.04 locally. It matched Saxon on 264 of 264 runs, after three fixes to how it reads Schematron.
-
France's e-invoice rules against Saxon, and a hidden space in Flux 10
France's BR-FR, extended, CDAR and Flux 10 rules, in the browser and on Saxon over 1,784 test copies: no rule ids differed. One Flux 10 gap is fixed.
-
Checking France's invoice status messages (CDAR) against FNFE's own rules
E-Invoice Preflight now reads France's CDAR status messages, compares them with the invoice and runs FNFE's BR-FR-CDV rules: 780 of 780 runs match Saxon.
-
Does a gear-pump selection rule survive a different idle? The slopes do, the intercept does not
An excavator gear-pump rule fitted at one idle setting, rerun at 22: it beats the drag-only rule everywhere, but each machine must refit one number.
-
One gear-pump rule for catalogue and overspun pumps: charge the drive's drag only where the optimum runs
Keep a gear pump's idle drag at 85 L/min and add only the step-up drive's extra drag where the fuel optimum picks: 34% lower error, no cost on catalogue pumps.
-
A GTIN filed under Peppol's GLN scheme gets the GLN check: 581 test files our checker passed
In a CII invoice, an item's GlobalID under scheme 0088 meets Peppol's GLN rule, so a mistyped digit bounces the file. Our checker now reads line ids too.
-
Half the time, the same prices: speeding up a packaging optimizer without changing one answer
The SaaS feature packaging optimizer spent most of its time building throwaway arrays. Removing them halved the runtime and every output byte stayed the same.
-
Halving the safety margins finishes the grid repair that never finished, and costs nothing elsewhere
A rule that halves an AC repair's margins only when the margin-shifted OPF has no solution finishes PGLib case73 API and loses none of 31 finishers.
-
Twenty fold patterns within half a millimetre: why a simulated drape test should report an interval
A simulated jersey drape has 20 fold patterns within 3e-5 J. Reported as an interval, the drape coefficient is mesh-flat to under 0.007.
-
Our cloth simulator ranks nine real fabrics' drape better than a commercial CAD tool, and over-drapes all
Given only measured FAST data, a cloth FEM ranks nine wovens' Cusick drape with r = 0.965 but runs 0.14 high; one bending factor fixes it to 0.05.
-
Hearth & Horizon: A Metre Apart, a story about survey lines that run the same way as the graves
A radar survey of an Irish children's burial ground finds nine graves. Its lines ran the same way as the graves, so nine is a floor. A story about sampling.
-
Hearth & Horizon: As Cold as the Shed, a story about the ice dam a warm room builds
A Hokkaido parish hall becomes a warm room, and water drips through its ceiling under 30 cm of snow. Plus one in the roof at minus eight explains why.
-
Hearth & Horizon: Four Times the Wait, a story about how slowly mud falls through cold water
An Andean village's spring is buried. Six hours of standing clears half the river's mud; Stokes's law says why, and prickly-pear slime does the rest.
-
Hearth & Horizon: The Joint That Is Still Pink, a story about lime that sets from the outside in
A mason repoints a Székely church tower in lime. The crust grows as the square root of time, frost is coming, and the scaffold is promised to another parish.
-
Hearth & Horizon: The Old Man Under the Nave, a story about counting the sky through forty metres of rock
A cracked church floor in a Slovak mining village, an old working on a 1781 map, and a muon detector in the adit below that counts what the rock lets through.
-
Hearth & Horizon: The Weight in the Belfry, a story about the tune hidden inside a swinging bell
A new 3.3-tonne bell rocks a brick tower a centimetre each way. Its push has a part at three times its swing, and the textbook damper is the wrong one.
-
Hearth & Horizon: Twice as Long, a Quarter the Load, a story about why a scaffold tube buckles
An earthquake cracks a church arch in Albania. A tube that pulled a tractor holds 8 kN standing up; Euler buckling, lacing and an engineer's no tell why.
-
Ten long battery simulations calibrate a plating forecast on average; eighty make it safe
Resampling 768 PyBaMM DFN designs: ten earlier 100-cycle runs match a 160-run plating calibration in the median, but the unlucky tail needs about 80.
-
The cost of a nailing robot's safer stance is the stance, not its balance controller
A 2 x 2 of leans and controller sides: the G1's 1.3-point loss in clean sheets is all the lean. The flipped sides change nothing but keep one pose holdable.
-
Can a humanoid nailing sheathing stand when the nail gun lifts off? Yes, up to a height limit
Leaning 10-39 mm less lets every leaned humanoid stance stand at zero press and still take 50 N. A formula sets the height ceiling: 1.225 m for a G1.
-
A waist brake takes a nailing humanoid's last half-millimetre of backlash error to 0.01 mm
After a leg preload and a two-hand squeeze, the last 0.5 mm of a humanoid's nail-gun error is waist yaw backlash. A brake takes it to 0.01 mm.
-
Check every IFC model against your IDS on each push: a free GitHub Action
One workflow step checks every IFC export in a repository against every IDS file, with annotations, a JUnit report and BCF files. Tried on four real models.
-
Checking bridges, roads and ducts against an IDS: 348 of 348 results match IfcTester
IDS Model Check against IfcTester on buildingSMART's twelve IFC 4.3 and HVAC sample models: 348 of 348 results identical, plus an IFC2X3 gap.
-
A green IDS check in CI can skip 21,405 elements: now the build says so
65 public IDS files on four real IFC exports: 165 checks skipped 21,405 subclass elements. The CI runner and the IDS editor now flag them.
-
One public IDS specification failed 665,224 entities and took our CI runner down with it
Running public IDS files in CI found an out-of-memory crash and a hang on truncated files. Both are fixed, and the runner is six times faster.
-
A date range in an IDS file is valid, and IfcTester fails every object on it
IDS 1.0 allows xs:date ranges. IfcTester 0.9.0 casts bounds to float and fails every object. An XSD pattern of the same dates reads the same in both.