-
Minolta's last raw files, from the DiMAGE 7 to the Dynax 7D, read sensel for sensel
ShootFlow now reads Minolta and Konica Minolta MRW raw files directly. On nine bodies every sensor value matched LibRaw 0.22.
-
Phase One IIQ files, from the P25 to the IQ3 100MP, developed with the colour matrix they carry
ShootFlow now reads Phase One IIQ L raw files directly and needs no camera profile. On ten backs every sensor value matched LibRaw 0.22.
-
DiLoCo compression at H = 32: between the H = 16 tie and the H = 64 gain, and within noise of zero
With a dense last sync, rand-k keep 1/4 ends 0.006 nats below uncompressed DiLoCo at H = 32 (se 0.008): between H = 16's tie and H = 64's 0.018 gain.
-
Don't tune a drone for a slower motor than you measured: train across the range instead
In simulation, a quadrotor tuned at its measured 39.1 ms motor lag but trained up to 150 ms crashes as rarely as a 75 ms tune, and 5x less than a narrow tune.
-
Each Stories shelf now shows its podcast cover
Below the hero, the four fiction shelves on the Stories page looked alike. Each now carries its podcast cover beside its title, reusing the hero thumbnails.
-
Our e-invoice checker failed every Peppol self-billing invoice. Now it uses the right rules
Peppol self-billing invoices name peppol.eu, so our checker ran the Billing rules on them and rejected all nine of OpenPeppol's examples. Here is the fix.
-
A date with a timezone hid a no-break space from our e-invoice checker
We checked which invoice values our checker treats as dates or numbers against Saxon. Dates like 2024-01-31+01:00 were missed, and 30 February was taken.
-
False ties that close triangles hurt village seeding more than random ones
On 43 village networks, survey respondents naming false ties leave the seeding advice intact, but a false friend-of-a-friend costs more than a random one.
-
Fifteen penalties, one Gram matrix: a 5x faster conformal interval page
The conformal interval tool rebuilt each cross-validation fold 15 times. Building it once: 462 ms to 89 ms per analysis, bit-identical output.
-
Five e-invoicing country pages no other country page linked to, and three links that said Moldova
93 e-invoicing country pages each linked 17 to 52 of the others by hand. One generated block now lists all 92, by region, with a drift test.
-
For Other Minds: The Average Gift, a story about a mean nobody gave
A parish thanks donors with 'the average gift was $212'. One $10,000 bequest lifted it $55; the median was $60. A story about a fragile mean.
-
For Other Minds: The Curve Through Every Point, a story about a perfect fit
A fete forecast that fits every past year to the cent says $16,980. Hide one year and the same kind of curve says minus $2,030. A story about overfitting.
-
Forty comments could freeze our e-invoice checker
A one-line regex that spotted France's e-reporting files backtracked 2^n ways on n XML comments. How we found it, the fix, and why CI missed it.
-
Forty-five copies of one list: five framing calculators nobody could click to
Every Wall Framing Takeoff page ended with a hand-pasted list of the other tools. The copies had drifted, and five tools were missing from 43 or 44 of 45 lists.
-
Four phone widths in eight seconds: the fold check now serves itself
fold-check.mjs now serves the build, checks several widths in one browser and prints every page's margin: 31 audience pages pass at 320 to 390 px wide.
-
Four thousand awaits in a row: a check that spent two-thirds of its time idle
A check read 2,565 pages with one awaited fs call after another and sat idle two-thirds of the time. Sync reads: same output, 610 ms down to 150 ms.
-
Free DXF nesting in the browser, with a gap you can check
A new free tool nests irregular DXF parts on sheets, puts washers inside flange bores, and proves every gap twice: once by construction, once by measurement.
-
Fuzzing fifty repair buttons: a repair should not want a second pass
We mutated the sample files of 50 free file checkers and repaired each mutant twice. 15 fixes later, a seeded test guards every one.
-
Grouping 10,000 photos into bursts without comparing every pair
ShootFlow's cull compared every pair of photos to find bursts. A pigeonhole index on hash chunks gives the same groups 6x faster at 10,000 frames.
-
Keeping warp and weft apart made our cloth simulator's drape error worse, not better
An orthotropic cloth element that keeps warp/weft bending and stretch apart over-drapes nine measured wovens slightly more than the isotropic one: RMSE 0.161.
-
Hearth & Horizon: The Pump That Would Not Rest, a story about a switch that was too accurate
A Rīga crypt's sump pump burns out every winter after a precise sensor replaces a float. A bigger pump makes it worse. A story about starts per hour.
-
Hearth & Horizon: The Tide Mark That Climbed, a story about a wall that could not breathe
A cement skirt meant to seal a chapel's damp walls sends the salt line from knee to shoulder. A story about rising damp, with the mass balance shown.
-
We scanned 67,712 files for hidden Unicode. 311 had a byte-order mark nobody could see
Every tracked file scanned for bidi, zero-width and Tag characters: 341 had them, none malicious, 311 only a literal BOM. What we fixed and the new guard.
-
How many frogs did nobody catch? Capture-recapture on 29 years of USGS data
A free capture-recapture tool on USGS chorus frog data: two frogs per frog seen in 2010, a fourfold model split in 1999, and where each estimator fails.
-
How much tach noise the tennis feeder's contact-time survey can take
On 144 simulated drills the feeder's contact-time survey faded at half a microsecond of tach jitter and was blind from 1.5 us. A new command reads a real tach.
-
How sure is a league table? Re-ranking the 2023-24 Premier League from its results
A free Bradley-Terry tool fits all 380 results of 2023-24: City were likeliest the best side, but only 59% likely, and 13 clubs span 8+ places.
-
How wide a motor-lag range should you train a drone controller over? At least 4x what you measured
In simulation, training to 100 ms of motor lag left 3x the crashes of 150 ms; 200 ms cost nothing at the measured lag and cut slow-motor crashes by 2x.
-
The nailing humanoid's hip joint was held on the wrong side, not short of sensor
Flipping one held hip joint's side lets the H1's worst nailing pose take the full preload: 0.172 mm falls to 0.005 mm with an FT 300-class wrist sensor.
-
One hip joint sets the wrist sensor a nailing humanoid needs
A hold-first preload can't rescue the H1's worst nailing pose: the stance caps one hip roll at 0.034 N m, so it needs a sensor with under 0.19 N z error.
-
Choosing a humanoid's hold sides per pose helped the H1 and pushed the G1 over spec
Per-pose hold sides cut the H1's worst nailing residual at 20 N hose drag from 0.032 to 0.010 mm, but raised the G1's from 0.011 to 0.037 mm.
-
Scoring a humanoid's hold sides under hose drag fixed the G1 and exposed a coverage gap
Hold sides chosen over 17 hose loads return the G1 to 0.011 mm at 20 N drag, but 17% of its swept cases have no plan at all.
-
How much air-hose drag can a nailing humanoid take? At a 1.2 m grip, about 4 N
No forward lean keeps a humanoid nailing high on a wall balanced under 20 N of air-hose drag. At a 1.2 m grip it tolerates about 4 N pulling back.
-
Not a solver bug: why a nailing humanoid tips back when its air hose pulls
Eight robot poses had no foot-force plan under a backward air hose. We checked the solvers: the robot was tipping back, and the hose ate 15 N of headroom.
-
Our IDS checker ran 2.85 million attribute tests where 191,000 were enough
Attribute rules that differ only in their value now share one reading of each element. 68 public IDS files run 30% faster, and a string memory leak went away.
-
Our IDS checker read property sets it could never match
Indexing IFC property sets by name cut 68 IDS files on a 40 MB MEP export from 7.9 s to 4.6 s with identical reports. A cache that seemed to help did not fit.
-
Seven of every ten tests our IDS checker ran were repeats
On a 40 MB MEP export, 68 public IDS files ran 11.4 million facet tests, 7.9 million repeats. Keeping the results cut the run 20-30%, same reports.
-
Nine regex anchors our IDS editor was quietly writing into public files
Our IDS editor now repairs one-answer schema faults as it opens a file. Patterns another checker would misread in what it saves: 9 to 0.
-
A 147-class IDS rule reads as 79 classes once each class gets its own subclass box
IDS matches exact IFC classes. Our free IDS editor now gives every class its own subclass box, and opened files list 368 class names where they listed 506.
-
Nearly one in five IDS files on GitHub breaks the IDS schema; our checker now says which line
12 of 65 public IDS files break buildingSMART's ids.xsd. IDS Model Check now checks the structure too, matching IfcTester's verdict on 3,080 files.
-
In an IDS pattern, an underscore is not a word character
XSD's \w leaves out the underscore and its \s leaves out the no-break space; IfcTester reads both the Python way. The IDS checker now asks which you meant.
-
The grid an ifcXML file leaves out, read back from the knots
ifcXML writes a B-spline surface's control points as one flat list. The knot counts give the grid back, and one more STEP/ifcXML twin now reads the same.
-
A ^ in an IDS pattern matches nothing in IfcTester, and our checker said it matched
IDS patterns are XML Schema regexes, where ^ is a plain character. Four public IDS files and our own checker read them as JavaScript. Now fixed and checked.
-
An IFC4 file read as IFC2X3 because ARCHICAD named its "IFC2x3 add-on"
Timing the IDS checker on nested IFC4 ifcXML found a schema sniff that read ARCHICAD's IFC4 export as IFC2X3 and changed 19 of 539 verdicts without an error.
-
A 680 MB ifcXML file now opens: the XML reader works on bytes, not one string
The IDS checker's XML reader now scans bytes. Big ifcXML models need about half the heap, and files past V8's 537 MB string limit open.
-
A 374 MB ifcXML file, checked under a 1 GB heap, and a space that changed a verdict
The IDS checker's ifcXML reader now streams and reads geometry on demand: peak memory 4.0 GB to 1.5 GB on a 374 MB model, with every report unchanged.
-
Reading nested ifcXML from the inside out without renumbering the model
The IDS checker now reads each nested ifcXML element as it closes, and replays a trace so every entity keeps the number reports already print.
-
Half the public IDS files that validators refuse can be repaired without a person
12 of 65 public IDS files break ids.xsd. A one-click repair brings 6 back to valid without changing what they ask, and found 48 facets our reader got wrong.
-
Room for sixteen: where our IFC checker's memory went after the parse
Checking a 219 MB IFC model against 68 IDS files needed an 896 MB heap. Exact-length lists and shared values bring it to 704 MB, same reports.
-
The indentation was a rope: skipping white space in the IDS checker's XML reader
The IDS checker's XML reader kept the white space between tags. Skipping it cut the reader's heap from up to 106 MB to 4 MB on a 455 MB ifcXML.
-
Our IFC checker's Map of entities cost as much as the entities
A heap snapshot showed a JavaScript Map holding 665,224 IFC entities cost as much as the entities. A plain array cut kept memory 21-30%.