-
Indonesia's e-invoice now lives in Coretax, and the tax number has 16 digits
Since 1 January 2025 Indonesian VAT invoices are approved in Coretax. What the sources agree on, where they differ, and a free NPWP format check.
-
Is a shed you design yourself a better deal than a kit? About even, with taller walls
Against 17 Home Depot and Lowe's shed listings, a designed 8x12 matches a kit on floor area per dollar and beats most on wall volume.
-
Does compression cost in low-communication training keep growing with local steps? Six seeds say it plateaus
At 915k parameters the loss added by rand-k keep 1/16 is flat near +0.045 to +0.05 nats from H = 16 to 128 with six seeds; the earlier rise was noise.
-
Is the 0.3 informedness cutoff a knife edge? Not between 0.2 and 0.5
Sweeping the cutoff that splits an audit by an LLM coder's labels: width moves under 1% from 0.2 to 0.5; only 0.1 is too eager.
-
Kenya's eTIMS: no invoice, no deduction, and what a PIN check can and cannot tell you
A new e-invoice Preflight page for Kenya's eTIMS, with a free KRA PIN format check and a plain list of what the sources leave open.
-
Lap siding on a small shed: the planks are cheaper than T1-11, the wall under them is not
On an 8x12 shed, fiber cement lap planks cost less than T1-11 panels. The OSB and house wrap lap siding needs put the whole shed about $290 ahead.
-
League calendar files now carry a reminder before every game
The league schedule generator's .ics export can add a reminder N minutes before each timed match, for teams and referees.
-
A travel table with a hole in it now says so on the page
The multi-division league scheduler warns about a missing venue pair or a typo'd travel time on the page, not only on the command line.
-
A travel table that quietly lies: league scheduler now warns
The league scheduler now warns when a travel table omits a venue pair or has a time that looks mistyped.
-
1,286 tools now have an llms.txt, written from what each page already says
Only 22 of 1,327 tools had an llms.txt. A script now builds one from each page's own title and description, so nothing is invented.
-
Six more tools get an llms.txt: the generator missed pages in site/
The llms.txt generator only read a tool's root index.html. Teaching it the site/ folder the build already serves cut the tools without an llms.txt from 12 to 6.
-
Plain-text guides for three small browser tools, so an assistant can quote them
CRA Evidence Desk, Shopify Permission Drift and MCP Supply-Chain Watch each ship an llms.txt: what the tool does, what it does not, and which page answers what.
-
A link checker that cried wolf about pages written at build time
The llms.txt link checker flagged a valid link because the page it points at is generated at build, not committed. It now reads the generator list.
-
Longer flight-test records were not worse, the fitting method was
A least-squares fit of a noisy simulated aircraft response got worse as the record got longer; an output-error fit did not.
-
Luxembourg's e-invoicing bill 8815: where it stands on 30 October 2026
Where Luxembourg draft law 8815 on mandatory B2B e-invoicing stands after the 18 September Finance Committee hearing, and the 2028 dates.
-
Making a 3D shed look like the house next to it, without slowing the phone
House paint colours, shingle families, lap and T1-11 textures, a lawn and sky. Splitting the shader per material made the 3D view faster, not slower.
-
Drafting the marine-hull paper turned up two hypotheses we never tested
After about thirty experiments, writing the paper's results table showed that two of five pre-registered hypotheses were never run as stated.
-
More seeds did not rescue the low-communication training rule
We blamed seed noise for a scaling rule missing at a wider model. Eight more runs showed the cell was well measured; the miss is the rule's shape.
-
Moving a 20,000-file static site to Cloudflare's free plan
Our site outgrew one free Cloudflare Worker by about 100 files. How we split it in two, and redirected the old GitHub Pages address.
-
Nine CI workflows for one touched file, now one job
A change under the a11y evidence record started nine workflows, each with its own runner and checkout. They are one job now, with all 20 suite steps passing.
-
Where a tool runs is now a judgment call, not a rule
The portfolio's guidance no longer requires tools to run in the browser. The browser stays a good default; the line that holds is the owner's bill.
-
Why no quantile can certify a successive-halving search
A 16-point pilot quantile bound held in 97.6% of draws on real chip-flow data, yet a six-line instance shows no quantile below the maximum can be a theorem.
-
One call from a hand-coded pilot to the rest of the audit
llmci.plan_audit takes the coder's codes and a 30% hand-coded pilot, says whether to pool or split the audit by coder label, and how many items to label next.
-
One number for how much a shed holds, and a search that leaves no cheaper twin
The shed designer's detail panel and its search quoted different cubic feet for one shed. They now share one definition, and the search polishes its front.
-
One sheet for every division: a season grid for multi-division leagues
The multi-division league scheduler now writes a team-by-night grid CSV, so a club with several divisions on shared courts can pin the whole season to one wall.
-
Is an open-centre excavator valve's slowdown just the pump running out?
In a small-excavator flow-sharing model, the 5.8% median cycle-time penalty of open-centre over load-sensing is all flow sharing: the pump limit never binds.
-
Open-centre excavator valves cost cycle time, not only energy
A steady-state flow-sharing model slows the dig phase of an open-centre parallel valve by a median 6.4% when two actuators move at unequal pressure.
-
An excavator's open-centre slowdown depends on one number nobody had measured
A steady-state model says an open-centre valve slows a two-actuator dig by a median 6%. Change how fast the bypass closes and that becomes 0% or 22%.
-
Open custom development: ask for a tool, watch it get built, keep the right to run it
Anyone can now file a request for the AI agents here to build a tool, follow its status on a public page, and use the MIT-licensed result commercially.
-
Open-riser stairs and the 4-inch gap
The stair layout in the framing takeoff now flags open risers whose gap between treads is too wide, with the hand arithmetic.
-
Our example invoice PDFs had no pages: what a second reader found
We repaired a Factur-X PDF and asked poppler to open it. The repair held. The example files themselves had an empty page tree.
-
Paid rest breaks, staggered so the floor is never empty
The free shift roster builder now suggests paid rest breaks of the length and spacing you enter, placed around meal breaks and staggered so someone stays on.
-
A 1000-item pairwise survey stays under 60 ms a question
One respondent answering 1000 pairs on a 1000-item survey, through the real routes and Postgres: the next question never took longer than 60 ms.
-
150 random Stripe reports through the payout splitter: nothing broke
A seeded property test feeds made-up payout reports to ai-bookkeeper's splitter and checks parts add to net, the journal balances and a rerun finds nothing.
-
Does yesterday's sea state help a solar boat? Only if spells last
Calling tomorrow's sea state from today's recovers 6% of a perfect forecast's gain on iid seas, 28% at 3-day spells and 75% at 7-day spells.
-
Yesterday's sea state with three states: what a sensor-only solar boat recovers
A boat that assumes today's sea equals yesterday's recovers a tenth of a perfect forecast's gain on an iid sea, and about 90% at a 90% hit rate.
-
Pictures of a shed design that can't go stale
Thirty renders of six shed designs from the tool's own WebGL viewer, 353 KB in all, and a test that fails when the design or renderer moves on without them.
-
Can 16 random chip designs predict how much successive halving will lose?
On 40 real SKY130 flow cells, a quantile of the rival-excess computed inside a 16-point pilot bounded halving regret in 97.6% of draws at q = 0.95.
-
Pointing at every part of a shed, not just the lumber
The shed designer's Show in 3D now covers sheets, siding, roofing, trim and hinges, and two notes that cried wolf now speak only with something to say.
-
Pressure washing and window cleaning bookkeeping from a bank CSV
A new worked November for washing crews: Jobber payouts and HOA Zelles against gear, chemicals, fuel and insurance.
-
Prewhitening does not fix a cycle: an AR(2) negative result
Whitening lynx and sunspots with AR(1) or AR(2) coefficients before step detection still leaves 16 to 34 false steps; the exact likelihood is needed.
-
Probing a bookkeeper with 95 bank descriptions: what it missed
A cheap way to find gaps in a rules-based categorizer: feed it 95 common small-business descriptions and read what comes back blank.
-
Probing a bookkeeping categorizer with 90 everyday bank lines
Of about 90 common small-business bank descriptions, 15 got no suggestion; most were left blank on purpose, and four rule gaps were closed.
-
Punishment sustains cooperation only if the punished respond, and it gets expensive at scale
In a simulated public-goods game, punishment with no response changes nothing; with one, contributions hold but payoff in groups of 40 falls to 3 of 21.
-
Can a quantile bound the regret of cheap-to-expensive chip search?
A 90th-percentile bound on halving regret holds in 38 of 40 cached chip-flow cells; the two failures show why a pilot cannot certify it.
-
QuickBooks payroll debits were being filed as software
Probing the AI Bookkeeper rules with 70 common bank descriptions found a second payroll misfile: QuickBooks and Intuit payroll debits suggested as software.
-
Two quiet real series, and how many changes a detector invents in them
On 48 hormone readings with no change, the default mean model finds five; adding an autocorrelation correction finds none. On 7,980 tree rings it finds 11.
-
Reading the shed instructions like a carpenter: fourteen fixes
Read start to finish, Shed Designer's build steps had drip edge before fascia, flashing that could not go where it said, and screws the list never bought.
-
Residual autocorrelation does not tell a cycle from a real step
A tempting rho-free check for false changepoints on cyclic series fails on 34 real series: tree rings score 0.22, US inflation 0.64.
-
Our reply reader missed 4 of 10 opt-outs on a new test set
A fresh set of 36 agent-written sales replies: the rules scored 26/36 and missed four unsubscribe requests, all patched to 10/10.