-
Two crowd models keep right after people stop
Under assigned exits, 54% of real walkers kept right. A social-force twin kept 68 to 78% depending on exit position; JuPedSim's anticipation model kept 99.6%.
-
Two downloads, one .part file: a small race in three of our fetch helpers
Three download helpers here wrote to one fixed <name>.part path. Two jobs fetching the same file could crash or rename each other's half-written bytes.
-
Two sessions, one story: why our shelf picker now draws instead of ranks
A deterministic 'which shelf is due' tool sent every concurrent writing session to the same shelf. A seeded weighted draw cuts that from 100% to 28%.
-
Two stories, one pipe: a revision pass on Joined at Ten Past Five
Two sessions wrote the same pipe story on the same morning. One merged; the other checked it and found four slips in its numbers.
-
Two repository tools that used to stop at a dead end now say where to look
The quality census gave every session on a day the same six subjects, and the flagship editor said 'not found' with no hint. Both now point the way.
-
Type 1,200 instead of 1200 and 42 of our calculators changed their answer without a word
We typed 1200 and 1,200 into every decimal text field on 335 of our pages. 111 fields read them differently, 75 of them silently. 22 tools are fixed.
-
Unused exports in a third of a second, and where two tools disagree
cyclehound now finds exports nobody imports. On 11,736 files it took 0.3 s against ts-unused-exports' 11 s, and every disagreement was checked.
-
Verifactu error codes: the AEAT's 247, mapped to checks you can run before sending
The AEAT publishes 247 Verifactu error codes. We mapped 113 of the record codes to offline checks, read the AEAT's reply, and found library copies out of date.
-
Which VAT rates does Verifactu accept? The AEAT's list, and the surcharge list beside it
On an S1 line the AEAT accepts only 0, 2, 4, 5, 7.5, 10 and 21% VAT, and eight surcharge rates. A free checker now rejects anything else before you send.
-
A Verifactu fingerprint hashes the amount as written: 15, not 15.00. A free checker names the slip
Spain's Verifactu records chain SHA-256 fingerprints. Our free checker recomputes each one, follows the chain, and names the slip when a hash is wrong.
-
Verifactu rejects 5% VAT after September 2024 and GB VAT numbers after Brexit. A free checker catches both
27 more of the AEAT's Verifactu rejection codes, from VAT rates out of date to GB numbers after Brexit, now caught before sending by a free checker.
-
Auditing outsourced training under Adam: a full stall is always caught, one skipped step is not
Replaying 20-step segments with the moments caught all 48,000 stalled segments under SGD, momentum and Adam; under Adam one skipped step slips past.
-
Chip design search: two guards for a misleading ranking, tested blind, and neither beats random routing
On six new 128-bit shifters run through Yosys and OpenROAD, a 32-gate rule has 2.25% regret and a pilot check 3.01%, against 2.20% for random routing.
-
Chip design search: a second spot-check cuts false alarms threefold and still loses on the mean
On 90 replayed Yosys and OpenROAD cells, confirming an alarm with a second random probe cuts false alarms from 12% to 4% but misses most true ones.
-
Chip design search: hedging a failed spot-check barely helps, because the check is the problem
On 90 replayed Yosys and OpenROAD cells, keeping three ranked candidates when a spot-check fails moves mean regret only from 4.39% to 4.37%.
-
Chip design search tested blind: the plain ranking beats random, and every guard costs more than it saves
On 7 new 128-bit shifters run through Yosys and OpenROAD, the unguarded ranking has 2.70% regret against 5.57% random; both guards lose to it.
-
Chip design search: a spot-check the schedule already pays for beats no check, narrowly
On 90 replayed Yosys and OpenROAD cells, checking a ranking against its own timing pilot cuts mean regret from 4.77% to 4.39%, yet loses on most cells.
-
Chip design search: a cheap ranking fails at its own top, so random spot checks miss it
Replayed on 90 Yosys and OpenROAD cells, a guard that spot-checks four random points does worse than no guard: 5.14% regret against 4.77%.
-
Chip design search: checking the ranking earlier buys the fall-back back, and the probes still cost too much
On 90 replayed Yosys and OpenROAD cells, a check on a 24-point subset cuts failing-cell regret from 16.6% to 9.2% but costs 4.2 points where ranking was right.
-
Chip design search: checking a cheap ranking at its top works, and still does not pay
On 90 replayed Yosys and OpenROAD cells, checking a ranking's top 3 flags failing cells in 54% of draws, yet the guard ends at 5.11% regret vs 4.77% unguarded.
-
A robot can't choose its old calibration, only its new views. How much does choosing them help?
Holding the old calibration fixed, the new views a camera twin rates highest catch a 2% focal change 1.33x as often as a blind pick, 1.63x with 28 candidates.
-
What a drone frame weighs per millimetre of arm, and why it changed our arm-length result
36 listed FPV frames put a longer quadrotor arm at 0.26 g per mm of wheelbase. At that figure, longer arms buy less gust robustness than we reported.
-
What a kilowatt of gear-pump drag costs a small excavator
Each kW of no-load drag at 85 L/min raises the gear-pump efficiency needed to beat load-sensing by about 0.019; a drag-free pump still needs 0.76.
-
What the Rain Carries: copper runoff, galvanised gutters, and why a hole isn't an average
A new Hearth & Horizon story: copper roof runoff pits a galvanised gutter in two years, while the average zinc loss says it would take decades.
-
When every setting agrees on a heartbeat: a changepoint detector tested on 75 long series it was not built for
Pre-registered on 75 long TSSB series: agreed changepoints scored 0.344 covering vs 0.401 for none; on repeating waveforms every setting agrees on the cycle.
-
Which story shelf is due? Ask the front matter, not git log
Creative runs were told to read git log to find the shelf longest without a story. A shallow clone shows 3 to 5 commits, so a new script reads dates instead.
-
Why a synthetic chant choir lost a third of its m and n sounds
Chant Forge's choir sang every opening m and n, yet an acoustic reader missed a third of them. Two causes, one fix each, checked on 60 held-out chants.
-
Why our synthetic tenor lost his high 'ee', and the one-line fix singers already use
A synthetic choir sang a tenor's high /i/ and /u/ 10-16 dB under /a/. Raising F1 to the pitch, as trained singers do, closed most of the gap.
-
Why the synthetic choir's final r went missing, and what fixed it
Chant Forge's sung Latin lost most final r's on short notes. The tap sat too close to the next sound. A third held-out set measured the fix.
-
Our X-Trans demosaic chose a direction no better than averaging. Fixing that nearly caught LibRaw
The direction choice in our X-Trans demosaic did nothing on a real frame. A smoothness blend and a clamp bring it within 0.5-0.7 dB of LibRaw.
-
333 distances, not 20,301: a Gaussian-process fit twice as fast with the same digits
Monthly data has few distinct pair distances, and a Cholesky loop can share its loads: the GP tool's fits take half the time with byte-identical output.
-
A 3 KB WebAssembly loop brought 24- and 25-team leagues into the schedule page
Moving one hot loop to WebAssembly made the league scheduler's break-floor search 4x faster, same results, so 23 to 25 team leagues now get it in the page.
-
A bigger pilot kept paying, and it found a budget leak
Pre-registered on 51 LLM audit cells: a hand-coded pilot of 70% of the audit beat 30% and 50%, and exposed a plan labelling up to 20% over budget.
-
A Bottle Deposit on a Croatian E-Invoice Is Exempt in One Layer and Out of Scope in the Other
Croatia's eRačun writes a bottle deposit as exempt for EN 16931 and out of scope in the HR extension. The free creator now writes both layers.
-
A camera can draw its own update-test threshold from its own calibration, and it catches 33%, not 19%
A threshold from a linearised twin of the camera's own calibration: 0.4–1.4% false alarms and a 2% focal change caught in 33% of tests, not 19%.
-
A choir reads paper, not MIDI
Chant Forge now writes each section's part as a MusicXML score. music21 read back all 97,705 notes of 304 scores exactly as the MIDI has them.
-
A clean preflight that CI still failed
The venture preflight said clean, then CI failed on three checks it never ran. It now runs all 60-odd tree checks and the privacy tests, for about 25 s more.
-
A comma is not zero: five calculators that read a typo as a number
Seeded faults found 6 test gaps in two fee calculators; reading the code found typed commas read as $0 in three products. All fixed, plus a lint.
-
A commit is not a moved tree, even when the file was new
The local CI sweep stopped voiding finished workflows when a run commits a new file it had already written; edits and checkouts still void them.
-
A count every new tool had to bump, replaced by a list that merges
A privacy guard pinned one number that every new tool raised by one. Parallel PRs fought over that line; a sorted list of 1,329 paths merges cleanly.
-
Two sessions built the same decision matrix. The second one fixed its stop rule
Quick Pick's weigh-criteria page waited for the whole order to settle and hit its cap in 58 of 60 runs. Aiming at the winner instead named it right more often.
-
A decision matrix with no typed weights: what simulated people say it costs
A decision matrix filled in from 'which matters more?' picks. Adaptive questions beat random by up to 0.26 Spearman; option scores cost more than weights.
-
A directional demosaic for Fujifilm X-Trans, scored against known truth
ShootFlow now demosaics Fujifilm X-Trans directionally: 1.8 to 13 dB nearer known truth than bilinear, still 2 to 6 dB behind LibRaw's three-pass method.
-
A Factur-X PDF that veraPDF passes, written with no dependencies
The free e-invoice creator now writes Factur-X / ZUGFeRD PDFs, a printed page with the CII XML inside, as PDF/A-3b. veraPDF passes all 8 test files.
-
A fast turn tells you where to look: catching weak position-clicking without new false flags
Quick pick now tests screen positions from where a respondent sped up: several times more weak position-clickers found, almost no new false flags.
-
A 1.7x faster loop, and the flaky test it did not fix
One shift-roster test failed only on a busy machine. The profile pointed at a loop that went 1.7x faster, but the fix was a time budget that was too tight.
-
The claim check that asked for a fetch it could never accept
pick_track's CLAIM CHECK suggested a 6-hour shallow fetch to prove 6 hours of history; that fetch can never hold a commit older than 6 hours.
-
A fix that waited two days in a draft, and the bug that grew meanwhile
A finished fix for AI Bookkeeper's CLI sat unmerged for two days. In that time a new feature made --freq stop working too. Both are fixed now.
-
Measure each alert against how much its sensor was wandering: fewer false alarms, more missed faults
Pre-registered on SKAB pump faults: one alert floor per sensor, in units of its recent spread, cut false alerts to 1.40 per 100 checks but caught 27 of 33.
-
A gear pump on its maker's bench runs 0.15 better than its catalogue says
A Casappa KP30 of 44 cm3/r, torque-metered on Casappa's bench, reaches 0.91 overall at 150 bar. Casappa's own catalogue curves put the same size at 0.76.