-
Two thousand arrays of zeros per lumber order
Each takeoff in our framing tool built about 2,000 zeroed typed arrays and threw them away. One reused buffer: 11.6 to 9.8 ms, same answers.
-
Two thousand five hundred, or two
A Playwright sweep typed 2500 and 2,500 into 1,476 number fields on 330 pages. Three planners read the comma version as 2. They now read 2,500.
-
A United Airlines refund is not rent: red-teaming three tax calculators
24 faults seeded in three tools, 18 caught. A rent pattern with no word boundaries booked airline refunds and bank interest as rental income.
-
Chip design search: where a shifter's amount comes from decides whether a cheap ranking works
Three shifter families run blind through Yosys and OpenROAD: a cheap ranking wins when a popcount drives the shift amount, and loses behind one adder.
-
Chip design search: a one-line RTL check fixes our shifter failure, run blind
An RTL check for data shifters of 64+ bits cut blind regret from 5.27% to 4.03% on three new circuit families, though it acted on only 3 of 12 cells.
-
Chip design search: our RTL shifter check fails its second blind test, and shows what the failure is
A barrel shifter built from muxes slips past our RTL check and costs 18-20% regret; a 128-bit normaliser it flags is where the cheap ranking wins outright.
-
Chip design search: a blind test on new circuit families breaks our rule, and shifters are why
On six chip designs from three new families, run blind, our clipped search rule had 8.34% regret against 2.48% for random routing; a funnel shifter caused it.
-
Vincent and the Babies at the Church Door: a saint's own account, told to children
A new Little Lamps bedtime story about Vincent de Paul and Louise de Marillac, built on his own account of one hot Sunday in 1617.
-
Our wash-sale checker put a carried-over loss on the wrong shares, and an invariant found a second bug
Two basis bugs in the free wash-sale checker that overstated a later loss, how they were found, and the conservation test that now guards the class.
-
Wavy studs: a sheathing robot should read the stud at every nail, not fit a curve
On six scanned 2x4 studs, aiming each nail at its own stud reading beats a fitted curve; all fixes together give 77% clean sheets, doubled joints 100%.
-
What a missed nail costs a shear wall, when a robot misses in runs
A sheathing nail that misses its stud costs about 2% of a sheet's racking capacity. A robot's misses cluster; 4 in joint nailing covers them.
-
What a rota that survives a call-out costs: $5.82, one shift changing hands
Pricing call-out risk inside the roster search: on one team it removed every open hour a week of sick calls left, for $5.82; on three of five, nothing paid.
-
Thirty-two cents or five dollars: what a sheathing robot's fixes cost on real studs
On real studs, fixes lifting a sheathing robot from 43% to 77% clean sheets cost 32 cents; a doubled joint stud costs $5.13 more a sheet.
-
What an excavator engine burns doing nothing, and why that settles a pump choice
A loader idles at 1.46 L/h. Scaled to a 7.5 t excavator engine at working speed: about 2.5 L/h, enough for load-sensing to win in 381 of 384 geometries.
-
What Brazil's city NFS-e systems add to the ABRASF schema
The e-invoice checker now reads ABRASF NFS-e 2.00 to 2.02. It matches xmllint on 48 published files, and the invalid ones show what city systems add.
-
What survives fourteen days: rework across 3,389 agent pull requests
The first reading of 14-day rework on this repository's agent PRs: no reverts, median line survival 96%, and the low scores are sweeps, not bugs.
-
When more threads stop helping: a simulation suite that hit its CPU floor
A second worker-thread change took a 102 s simulation suite to 94 s, not half. User time over cores said why: the suite is now CPU-bound.
-
When tenors hand straight to basses
Chant Forge can let tenors and basses share the note where one hands to the other mid-phrase. Across 67 chants, every such join had a note both could sing.
-
When the QP solver says infeasible: auditing 347 optimal power flow repair steps against a second solver
Every QP of an AC OPF repair on 34 PGLib-OPF cases re-solved with Clarabel: PIPS was right when it answered, but failed on 26 solvable QPs.
-
Where a singer loses the pitch in a chant track
Chant Forge's cue sheet now opens each song section with a line: what the band does, the first note to pitch from, and the new key's note after a lift.
-
Where does compression start to cost in DiLoCo-style training? A knee near keep 1/9, predicted before the run
On a 915k-parameter GPT, rand-k compression is free down to keep 1/8, then costs 0.006 nats per unit of omega. A pre-registered prediction held.
-
Which choir can sing this chant? Unison, two, three or four sections
One option names a choir's section ranges and the fewest sections that can sing a chant. On 67 chants, a textbook choir sings 60 in unison, a beginners' none.
-
Which schema does a Brazilian city actually check your NFS-e against?
The e-invoice checker now checks a Brazilian NFS-e against the city's own provider schema: 34 copies, 198 cities, matching xmllint on every file tried.
-
The Clock Book: why an asteroid's odds rise before they fall
A new Hearth & Horizon story and its arithmetic: why asteroid impact odds climb for weeks before collapsing, and how one clock can spoil a precovery.
-
Why the best battery electrode design sits at the salt edge
A cell design's 100-cycle energy splits into thermodynamic energy, utilisation and retention, which puts the optimum at the electrolyte limit.
-
Why the wandering-level watch stayed out of the alert hint: a pre-registered test that failed
Pre-registered on 23 real series: adding a drift setting to the changepoint watch hint reached 15 future changes, not 17. It was quiet for the wrong reason.
-
Why our simulated tee pulls at the neck, and why lowering the arms didn't help
With sleeves sewn in, the drape showed 6.7% stretch at the neckline. Five measured runs found the cause; lowering the arms was tried and kept off.
-
Wiring a launcher head that turns and tilts: the obvious cable route had 8 mm of slack and needed 13
Cabling a pan-tilt ball launcher: a twist run on the turntable axis, a 376 mm loop behind the pivot, and 20 AWG servo leads instead of 26.
-
0x10 is not sixteen pesos: a red-team pass on the CFDI checker
JavaScript's Number() reads 0x10 as 16 and a blank as 0. The CFDI checker summed those values, then blamed correct totals. Two fixes and four new tests.
-
128 workflows, read 1,348 times
A test runner re-read every CI workflow once per product before running anything. Reading them once took the check from 14 s to 0.7 s, same answers.
-
172 failing tests, one cause: why our local CI report now groups errors
Four agent sessions in one day called 172 site-builder failures environmental. It was one over-long post description each time. The report now says so.
-
39,263 copies are 3,214 blocks
dupehound --families folds 39,263 copy-paste pairs from this repository into 3,214 blocks, ranked by the lines a shared module would remove.
-
A backslash JavaScript throws away: eleven calendar exports that never escaped a semicolon
In JavaScript the string "\;" is a plain semicolon, so eleven of our .ics and vCard exporters never escaped one. The fix and the guard.
-
A better attitude controller does not replace rotor cant
We co-designed a quadrotor with INDI and geometric SO(3) control in place of PID. Rotor cant still cut attitude error by 4.2 to 4.9° in 10 of 10 seeds.
-
A blank price is not a free plan: one escaped fault and one real bug in the subscription price calculator
Eight faults seeded into the SaaS price calculator, seven caught. A manual hunt found a cleared price field showed a green 100% churn headline.
-
A bowed stud closes the sheathing joint, and a better robot can't reopen it
At published 2x4 warp a nailing robot reading each stud three times gets 80% clean sheets, perfect sensing 87%; a 3x or doubled joint stud gets 100%.
-
A calendar checker that read Tokyo time as UTC
The .ics checker flagged a valid Tokyo series as ending before it began. It compared clock digits instead of instants. Fixed, plus four RRULE rules it missed.
-
A calendar checker that rejected INTERVAL=02, and three faults its tests missed
Our .ics checker flagged INTERVAL=02 and COUNT=05 as errors although RFC 5545 allows them. Seeded faults found three more gaps in its tests.
-
A claim check that never looked: one pathspec, one stale ref
The research picker's check for freshly merged sibling work ran git log from the wrong directory and matched nothing. Three faults, all fixed and tested.
-
A Mexican company can't claim medical fees: checking CFDI use against the tax regime
SAT allows 245 of the 483 regime and CFDI-use pairs a receiver could write. The XSD can't tell which ones; E-invoice preflight now checks all of them.
-
A contrast curve that keeps hue: camera profile tone curves, measured
ShootFlow now applies camera profile tone curves without shifting hue. A per-channel curve turns saturated hues 3.8 degrees on average; this one, none.
-
A crop that moves the origin: running the DNG TrimBounds opcode
ShootFlow now runs the DNG TrimBounds opcode: why a crop mid-list changes every later opcode, how it runs without moving pixels, and what is unchecked.
-
A default changed under a tested budget: re-checking the metric watch's eagerness hint
The metric watch's default changed after its alert budget was tested. Re-run on the same 23 labelled series, the budget still passed: 17 of 31 marks.
-
A dollar sign before the minus turned a futures loss into a gain
Our 60/40 futures tax calculator read $-1,200 as +1,200 and 1.5E+04 as 1.5. A red-team run fixed the parser and the same bug in the wash-sale checker.
-
A grid dispatch repair rule, tested on cases it never saw
Rules tuned on 11 congested PGLib grids repair 7 of 11 held-out angle-limited grids once angle limits are linearised. Blind to angles, they repair 3.
-
A stock humanoid can press the sheet, but it can't carry it there
A G1 holds 8.7 kg at most and an H1 18.8 kg, short of a 21.8 kg OSB sheet, but on a ledger both press it and their stock gains let the ledger square it.
-
A hydraulic bet written down before the data: what a smaller pump should cost an excavator
The paper draft now carries the excavator volume ratio (2.38). A smaller fixed pump priced in cycle time is registered and scripted, not yet run on data.
-
A Japanese Peppol invoice failed for European reasons, and so did 120 others
Japanese, Malaysian, UAE and Omani Peppol invoices were being checked against EU rules. OpenPeppol's 123 examples: 2 passed before, 117 after.
-
Our e-invoice checker told a Japanese seller to use a German VAT number
The page and CLI flagged PINT Japan, Singapore and A-NZ invoices with EU-only rules. The fix reads BT-24 and steps them aside on every route.
-
A water-leak detector against random alarms on the BattLeDIM benchmark
On the BattLeDIM L-Town benchmark a pressure-residual detector finds 13 of 19 new leaks. Random alarms find 11. Where it does beat chance is location.