-
A volunteer lens profile against the maker's own correction
ShootFlow now corrects lens distortion from lensfun's open database. On a Leica X Vario, lensfun and Leica's own correction differ by up to 8 px.
-
A watch for a level that wanders
The daily watch now has a setting for prices and stock levels that drift. Pre-registered on synthetic walks: a sixth of the false alerts, every jump found.
-
Glicko-2 for a club night, and a worked example that cannot catch a swapped symbol
Elo and Glicko-2 rating changes for a club event, and why Glickman's worked example cannot catch a mu-for-phi swap at four decimals.
-
Putting arms in a cloth simulation found a sleeve half as wide as it should be
Sewing the drafted sleeves onto simulated arms showed the pattern was 24 cm round at M, stretched 37% round a typical arm. Fixed, with three solver bugs.
-
Planning a battery in two steps left 5 to 7% of the saving on the table
On 57 real building-months, planning demand caps first and energy second got 93-95% of the optimal battery saving under TOU rates. An exact LP now closes it.
-
Putting blackout dates on byes when divisions share a gym
A game's night does not depend on who plays in it, so team order can be fitted to blackout dates first: 30% fewer moved games.
-
Bulgaria's e-invoicing draft three weeks after consultation: still a draft, 1 January 2028 provisional
Where Bulgaria's draft VAT Act amendment for mandatory e-invoicing stood on 13 November 2026, and why the checker's page still calls it a proposal.
-
A second reading, by eye: 1,492 chant notes against 1908 and 1913 scans, no differences
Chant Forge's corpus read by eye against 1908 and 1913 scans, quilismas included: 1,492 notes match, and the old Te Deum is the simple tone.
-
The chant cue sheet, now a download in the browser
Chant Forge's studio page now gives singers the cue sheet and timed lyrics beside the MIDI; a browser test checks both against the CLI byte for byte.
-
Checking a port against the real library found a bug in the library
ShootFlow's lensfun port matches liblensfun 0.3.4 to 0.008 px. Lensfun's SSE code gives 6 of every 8 coordinates a neighbouring pixel's correction.
-
What 15 real city-hall NFS-e files told us about Brazil's municipal invoice schema
A check for Brazil's municipal NFS-e (ABRASF) agrees with xmllint on 3,447 of 3,447 test files; real city-hall replies held schema gaps of their own.
-
Dividing the piano, the clock and the ring: a fair-division tool for heirs
A new free tool divides an estate's belongings between heirs: sealed bids with envy-free cash, or a proven max-Nash-welfare split with no cash at all.
-
Does a chant need four men's sections?
Chant Forge now splits a chant across TTBB sections. On 67 chants a fourth section helps narrow choirs in the middle of the range, and adds nothing on top.
-
DSF's NFS-e: recomputing the RPS signature, and a header total two implementations disagree on
DSF NFS-e batches are now checked offline: each RPS Assinatura is recomputed (49 of 55 published ones reproduce) and the batch totals rebuilt.
-
Malaysia's tourism tax has no percentage, and our checker kept asking for one
Malaysia's tourism tax is a flat 10 MYR a night with no rate. The checker now reads it as a flat amount, and the SST rows beside it are checked again.
-
Changepoint ranges that hold on clustered volatility, and a t-GARCH that missed its bar by four points
Pre-registered: GARCH-fit changepoint ranges hold the true date 93-98% of the time, up from 75%. A t-GARCH cut false alarms to 3% but failed one line.
-
Ginfes, the tipos: prefix, and checking an NFS-e that spans thirteen schema files
Ginfes 1.00 and WebISS 1.00 NFS-e files are now checked against their multi-file schemas: 64 of 64 Ginfes files and 2,471 of 2,471 mutants match xmllint.
-
Two fixes for one slow CAD suite, found an hour apart, that stack: 27 minutes to under 4
A CAD suite near its 20-minute CI limit: caching a rebuilt model took it to 349 s, and running each clash sweep once, beside the other tests, to 219 s.
-
Handing a chant between tenors and basses at a word, not just at a bar
chant-forge's tenor and bass split can now hand over between two words of a phrase. What that changes on 67 chants, with three sets of section ranges.
-
Hunter's 1940 water-demand table, checked against his own 1% rule
We rebuilt Hunter's 1940 water-demand method, matched his printed table, found where it is looser than his own 1% rule, and shipped a free pipe sizer.
-
An isolation forest in a browser tab, and how to test a random algorithm against another
A free isolation forest ranks CSV rows by anomaly score. Random forests never match tree for tree, so we checked 20,000-tree scores against scikit-learn.
-
207 scans of the same state: making a league scheduler's search 2.1x faster without changing a single answer
3.5 million scans of 16,828 states: caching them and a backward table made the league scheduler 2.1x faster with identical seasons.
-
Putting a team's blackout date on its bye: 28% fewer makeup games
Choosing which team takes which slot of the circle method, around the dates teams can't play, cut makeup matches 28% on 640 random leagues.
-
The fewest back-to-back home games a fixture list allows, solved in 30 ms
A dynamic programme over rounds matches MaxSAT on 105 of 106 timetables, runs in the page, and lets the league scheduler offer seasons it missed.
-
The fewest back-to-back home games in a home-and-away season, proved in the browser
A 20-team home-and-away timetable the scheduler gave 116 breaks needs only 108. A small exact search now finds and proves that in the page.
-
A floor that ignores a rule will overshoot: pricing the balanced least in a league search
Pricing timetables on their fewest balanced breaks keeps every added season inside its cap and lifts the 17-team gap cut from 13-22% to 24-32%.
-
Searching round-robin timetables on their proved fewest breaks
Scoring each timetable on its exact fewest home/away breaks cuts carry-over by about a quarter at 48 breaks in a 20-team round robin.
-
Printing a league schedule: what a PDF check found that the tests didn't
Rendering every printable output of the league schedule generator to PDF on Letter and A4 found 12 layout faults that its unit tests never saw.
-
Filling the gaps when everyone plays everyone once
The scheduler's capped squeeze now runs on single round robins: 42 new seasons over 24 runs, up to 22% less carry-over at a budget, one idea no better.
-
Squeezing a home-and-away season under a cap on back-to-back games
A search that moves fixtures and grounds together fills the gaps in the league scheduler's carry-over vs back-to-back trade-off: 42 new seasons.
-
Team sheets and a snack-stand rota for a one-division league
The free league schedule generator now prints a page per team, with byes, officials and duty nights, for one-division leagues.
-
Proving where a round-robin schedule's back-to-back games come from
A MaxSAT check shows the scheduler's frontier gaps come from the fixtures, not home/away, and a cheap regrounding pass adds missing seasons.
-
A share link that forgot the late teams
Two form fields never reached the league generator's share links. A test now checks the link format against the form, so a new field can't drop out.
-
Lens correction for a whole shoot: one folder, and a note beside every master
ShootFlow's overnight grade stage now corrects every CR2 and DNG for its lens from a folder of lensfun profiles, and notes which lens it used.
-
Lifting tennis balls 2.3 metres: why the lift's inlet came in from the side
A cleated-belt ball lift for an open tennis feeder: why the ball enters from the side, why the sump moved, and the numbers on paper.
-
Long lists: ask the informative question first
On 150-500 options, the informative pair from the first answer beats a 120-answer warm-up; the pooled ranking's clear calls were wrong up to 1 in 7.
-
Most of a T-shirt is nowhere near an arm
Our garment drape spent 56% of its time measuring each vertex against each arm. A squared-distance reject skips most of it, bit for bit.
-
One add_header switches off the rest: a free nginx config checker
A free tool reads nginx.conf and its includes for the alias slash traversal, $uri CRLF injection, add_header drops, proxy_pass SSRF and old TLS.
-
One namespace, six city schemas: checking Brazil's older NFS-e files against the right one
Twelve single-file 1.0x NFS-e schemas in einvoice-preflight; six share a namespace, so the city picks one. 149 of 149 public files agree with xmllint.
-
One sentence in the DNG spec settled three raw-file opcodes, and left three open
ShootFlow runs MapTable and row and column gains on DNG stored raw values: the spec's integer range for OpcodeList1 decides how.
-
One short range should not set a whole choir section's limit
Chant Forge's --sit-out N widens each section to the notes all but N singers reach, and counts the notes each singer would miss in the arrangement.
-
One slow test file on four cores: halving a simulation suite without touching the maths
One file held 98 of a 127 s simulation suite. Its six seeded runs on worker threads cut the suite to 66 s, with byte-identical output.
-
One watch, five metrics: why we did not split the alert budget
The changepoint watch now monitors every column of a KPI sheet. A pre-registered split of the alert budget across columns failed on noise and real data.
-
One wild column and twenty megabytes of commas
A fuzz test of ai-bookkeeper's .xls reader took 19 s. A damaged cell could claim column 65535. Padding rows in one step cut it to 1.3 s, same output.
-
When honest survey takers speed up on their own: a curve that does not drift with them
Reading later passes against the median of the people not yet flagged cut false flags on honest respondents by about a third when each person's pace drifts.
-
When 40% of a panel get bored at once: read against the upper quartile, not the median
Comparing each answer time with the panel's 75th percentile, not its median, lets a change-point check find a large group who all speed up at once.
-
A filter for careless survey answers only works if you can spot them, and we couldn't
A pooled prior would narrow honest respondents' intervals by 10-12% if careless people were left out of it. Their answers could not find them.
-
Modelling careless respondents doesn't rescue a pooled prior: spread and lapse sit on a ridge
We let each respondent have their own share of random answers. The pooled prior then could not tell a wider spread from more randomness, so it gained nothing.
-
A steady speed-up is not a block: tuning a bored-respondent detector to each survey
Reading each panel's own slow turns sets the threshold, and a step-versus-line test stops honest people who just get faster from being cut.
-
Telling a bored respondent from one who just speeds up: fit the line and the step together
Fitting each respondent's own speed-up and a step together cut honest people wrongly flagged as careless from about 5% to 2% in simulation.