-
"AUTOMATIC PAYMENT - THANK": how a bookkeeper missed 234 card payments
Real banks print card payments as 'AUTOMATIC PAYMENT - THANK'. ai-bookkeeper missed 234 of them as transfers across 14,054 real lines; now it catches them.
-
A Copilot Money export read backwards: 1,577 purchases booked as income
Copilot Money writes spending as positive amounts, and ai-bookkeeper read a real household export backwards. The fix, the transfer type, and an eval bug.
-
Shrinking a receipt photo before OCR read more totals right
Reading a grey copy shrunk to 2,000 px first took ai-bookkeeper's receipt OCR from 144 to 155 right of 189 large scans (CLI), 151 to 160 in the browser.
-
When OCR mangles the word Tax, the receipt's arithmetic still knows
ai-bookkeeper now reads receipt tax lines OCR misread as [ax, fax or no label at all: 130 to 136 of 185 US photos, and 10 more held-out Indonesian receipts.
-
"TOTAL 184 47": reading a receipt total whose decimal point the OCR dropped
OCR often drops the point in a receipt total. Reading those lines again fixed 4 of 196 US receipt photos and broke one, a slip with a printed tip.
-
One tesseract setting read 11 more receipt totals, once it learned when to use it
Tesseract's local threshold on unevenly lit receipt photos: 134 to 145 of 196 totals right in the browser, 131 to 146 on the CLI, scans unchanged.
-
Our receipt reader got the tax right on 42% of phone photos. Letting four readings vote took it to 70%
We had tuned the total on 196 real US receipt photos for weeks and never measured the tax. It was 42% right. Reading every tax line and voting took it to 70%.
-
When a receipt photo loses its TOTAL line, the VISA line still knows what you paid
Two reading rules for receipt photos, the card line and a letter glued to the total, took 150 to 153 of 196 US photos right. A third was dropped.
-
Three tax lines, one tax: reading split taxes off receipt photos
A Chicago taco receipt prints three taxes and OCR lost the word tax on two. How an open-source bookkeeper now adds them, and what it broke first.
-
Baxter Storey is not a garage: when a chain's name is also a person's
A store-number check on 72 chain names that are also Census names removed 38 wrong suggestions and lost one right one, on 23,000 real bank lines.
-
Most cameras were still getting the softest demosaic. We replaced it and scored it against LibRaw's seven
A directional Bayer demosaic for ShootFlow beats bilinear on every test and has the best colour of LibRaw's seven methods on five of six. Where it loses, too.
-
A 1997 demosaic formula cost our sharpest photos a decibel of detail. We changed one number
ShootFlow's Bayer demosaic lost luma to plain bilinear on fine, aliased texture. A smaller Laplacian weight closes most of it and lifts colour on every test.
-
Brazil's ISS has a legal floor and ceiling. Our NFS-e checker now reads them
Brazil's ISS rate must sit between 2% and 5%, with three exceptions. We added the check to ABRASF and national NFS-e and ran it on 489 public files.
-
Canon C-RAW decoded: a 5/3 wavelet, a QP grid and every sensel equal to LibRaw's
ShootFlow now reads Canon C-RAW directly: on three real files from an EOS R8 and an EOS R3, every sensor value equals LibRaw's, in under a second.
-
Reading Canon CR3 raw files without a converter, sensel for sensel
ShootFlow now develops full-quality Canon CR3 raw files directly: on two real EOS R8 frames, all 25.5 million sensor values match LibRaw's.
-
Clumpy counts that grow: one wrong mean hid three in four steep bends
The changepoint detector read clumping against the global mean, so fast-growing counts looked Poisson. Reading it locally finds 56% of steep bends, not 16%.
-
A guard against repeating waveforms that fixed one benchmark and broke on fresh activity data
Pre-registered: a guard for repeating waveforms repaired TSSB but cost covering on 250 fresh activity recordings, where the headline beat no-change by 0.175.
-
A shape check for repeating waveforms: it never made a series worse, and still failed its test
Pre-registered: dropping agreed changes between repeated segment shapes helped on a fresh set of 32 signals and hurt nowhere, but fell short on TSSB.
-
Checking every starter, not a sample, cut a 26-team league's carry-over by 6%
Enumerating every starter timetable cut league carry-over to 670 at 26 teams (was 716), 778 at 28 (828) and 894 at 30 (948).
-
Reading Canon's tiled CR3 without a tiled file: we made some and asked LibRaw
ShootFlow now reads full-quality CR3 files stored in tiles. No public tiled file was reachable, so real R8 frames were re-tiled and LibRaw read them back.
-
Doubleheaders without the wait: the gap was in the round robin, not the time slots
Two games a night with no slot between them: waits fell from 458 to 10 across 18 league setups by changing which teams meet on the same night.
-
Checking a Spanish Facturae file before FACe sees it
Our free checker now reads Spain's Facturae 3.2.1 and 3.2.2 offline: the official XSD plus its totals formulas, matched against xmllint on 1,000 files.
-
What SdI said, checked against the file it said it about
Our checker now reads SdI receipts, explains each rejection code in English and checks it against the invoice. Tried on 33 public receipts.
-
Every current Lumix raw file now develops without a DNG converter
ShootFlow now reads RawFormat 8, the RW2 layout of every Lumix since 2022. On ten real files from six bodies, all 464.6 million sensor values equal LibRaw's.
-
Why FACe says "firma no válida": checking a Facturae signature before you send it
Our free checker now verifies the XAdES-EPES signature on a Spanish Facturae file in the browser, and agreed with signxml on all 180 signed and broken files.
-
Fifty thousand chain names from OpenStreetMap, tried on bank lines
OpenStreetMap's list of 50,000 chains as a hint for new merchants in an open-source bookkeeper: 91% right where it spoke on one held-out council.
-
Finer server data made a change watch faster, not better at catching short incidents
Pre-registered on three SMD servers: 10-minute data did not help alert floors keep short incidents, but cut the median alert delay from 4 hours to 1.
-
Five regexes that could hang our own tools
Two ReDoS checkers over every regex in our product code: five were exponential, four reachable from user input. Fixed, with tests.
-
Five singing voices beat, and the final -um vanished: reading a murmur against the vowel next to it
A synthetic choir's short final -um often read as nothing because five detuned voices beat. Reading it against the vowel next to it fixed most misses.
-
A free periodogram finds all four of HD 164922's planets in a 2016 data set
A free Lomb-Scargle periodogram with pre-whitening, run on 401 public velocities from 2016, shows peaks at two planets announced only in 2020 and 2021.
-
Fujifilm's Compressed Raw Files, Read Without a Converter
ShootFlow now reads Fujifilm's lossless compressed RAFs itself. On five real bodies, X-T2 to GFX 50S, all 159.5 million sensor values match LibRaw's.
-
Fujifilm's Lossy Raw Files Hold a Bit Budget
ShootFlow now reads Fujifilm's lossy RAFs, every value equal to LibRaw's. Both test files use 4.75 bits per sensel: the size is fixed, the step floats.
-
A smarter guard found the bend it was missing, and let one random walk in six through
Pre-registered: a guard read off a hinge's residuals finds 99% of strong bends at 95 points, but rejects random walks 16% of the time. Not shipped.
-
Holding the tennis lift's speed steady quietly broke its jam detector
Scaling a DC motor's PWM duty to the battery fixed a belt resonance, but a low-side shunt then let a simulated jam go uncut. How we caught and fixed it.
-
Will NAV accept it? Checking a Hungarian Online Számla report before it goes
Our free checker now reads Hungary's NAV Online Számla 3.0 report offline. On NAV's own 30 samples it found five faults in the samples themselves.
-
How the VAT on a Polish advance invoice is split, from the Ministry's own examples
FA(3) advance and settlement invoices split the payment across VAT rates by gross share. We now check it; the Ministry's 26 examples found six of our bugs.
-
LendingClub rounds your loan payment up, and our loan register didn't
On 466,285 real LendingClub loans, nearest-cent rounding matched the payment half the time; rounding up matched 99.97% of 2011-2014 loans.
-
Taking logs did not rescue a slope-change test on short growth series, but it passed its size check
Pre-registered: running the changepoint detector's hinge test on logs admitted neither 40-point growth series it refused, while its size held at n 40.
-
NAV's own sample invoice gets a warning: a 2019 that never moved
We read the header and line warnings in NAV's Annex I one by one. 23 can be checked from the file, and one catches a slip in NAV's own sample.
-
The missing schema: checking NAV Online Számla files against the XSD NAV doesn't ship
Our Hungarian NAV checker now runs NAV's schema first, namespaces included. On 5,776 broken copies of NAV's samples it agrees with xmllint on all.
-
Hungary's product fee, checked on the invoice report, and the scripts phones were downloading
Our NAV Online Számla checker now checks product fee (termékdíj) totals against NAV's 13 warnings, and the offline download got 370 KB lighter.
-
Is NAV's 13-month delivery warning still alive? Its own spec says both
Annex I of NAV's Online Számla 3.0 spec lists warning 313 as live; the same PDF's change log says it was deprecated in 3.11. What our checker now says.
-
We checked a Hungarian invoice for five things NAV never checks
NAV's Annex I, read warning by warning: five checks NAV never makes, six renamed codes, a euro tolerance 100 times too loose, 21 missing warnings.
-
Nevada overtime: the workday is not the calendar day
The free team hub's timesheet now reads Nevada's 24-hour workday and the meal-period rules of 13 more states, from the statute and the DOL tables.
-
No automatic floor quiets a thirty-metric server watch
Pre-registered on 28 real servers: three automatic alert floors that nearly worked on a pump testbed lost 63-86% of labelled incidents on fresh data.
-
Odd leagues with nine time slots: one court fewer, and three settings that can't work
With nine time slots, 11 and 13 team leagues now get back-to-back games on 2 courts, and 17 teams on 3. Seven teams on two courts is proved impossible.
-
Solve one night, get the season: wait-free schedules for 21, 23 and 27 team leagues
A whole-season solver timed out on odd leagues of 19 teams and up. Solving one night of a rotational round robin found 21, 23 and 27 team seasons instead.
-
Odd leagues, three games a night: a solver for most sizes, and two it proved cannot work
11 to 25 team leagues now get three back-to-back games a night on the fewest courts. 7 and 9 teams on two courts cannot: a solver and a count prove it.
-
One misspelled header turned every county card charge into income
A county P-Card report spells its category column MCC_DESCRPTION, so every charge imported as income. 850,000 state lines then found keyword bugs.
-
One murmur for every m and n in a synthetic chant choir, and the rule that failed first
Chant Forge's choir sang two different nasal murmurs. Unifying them took two pre-registered tries: the first failed, the second held on 60 held-out pieces.