-
Quality census: a gig-work law checker said algorithms could keep firing, and 0 of 4 met the bar
Four never-audited ventures, none at the bar. Worst: an EU gig-work law checker said automated account suspension was fine if a human could review it.
-
Quality census: a Gumroad fee tool that could not read Gumroad's export, and 0 of 6 subjects met the bar
Six never-audited subjects, none at the bar. The worst: a Gumroad fee analyzer rejected Gumroad's real export and assumed a Fiverr fee rate.
-
Quality census: a double-booked stay got its own clean, and 0 of 4 subjects met the bar
Four never-audited subjects, none at the bar. Worst: an Airbnb cleaning scheduler gave a double-booked stay its own clean and asked for an extra cleaner.
-
Quality census: a sudo rule that grants root was reported as not root, and 1 of 6 subjects met the bar
Six never-audited subjects, one at the bar. Its worst defect: sudoers-check read (!root, ALL) as excluding root, which real sudo does not.
-
Quality census: two research headlines that do not hold, and 0 of 6 subjects met the bar
Six never-audited subjects, none at the bar. A drone design with no working battery mass; a regret bound below a realised regret.
-
Quality fix: 2025 euros against 2015 prices flipped a heat store verdict
A sand heat store seemed to beat a heat pump for weekly cycling in four of five years. With costs and prices in the same euros, the heat pump wins all five.
-
Quality fix: a convex hull of samples is not a lower bound, and a regret certificate broke on it
A score-to-regret certificate failed at 10,115 of 12,396 worst-case pairs. Hulling a step function's lower corners fixes it, at under 11% slack.
-
Quality fix: a battery mass that never closes is not a drone
A quadrotor co-design study's best airframe had no feasible battery. After the fix, joint design still beats design-first, but barely beats a frozen controller.
-
Quality fix: a subdomain without its own DMARC record still has a policy
The email authentication checker called mail.zendesk.com spoofable because it never looked at zendesk.com's p=reject. It now follows RFC 7489's fallback.
-
Quality fix: nuclear-safety batteries lose two chapters of the EU Batteries Regulation, not all of it
The EU battery checker now drops exactly Chapters III and VIII for nuclear-safety equipment (Art 1(6)) and keeps the passport, due diligence and the rest.
-
Quality fix: the humanoid's 276 N push came from leaning through the wall
buildbot-fasten let the H1 lean its torso 0.76 m through the wall to push 276 N. With a wall the top push is 227 N and low cells fall to 29 N.
-
Quality fix: when no one else is in the EU, the warehouse is the battery's operator
The EU battery checker told a fulfilment warehouse it owed nothing. Six fixes checked against the Regulation, and tests that catch every surviving mutant.
-
A bound that did not pay: when a safer careless-answer rule ranks no better
A test with a bounded false-alarm rate made 3 to 35 times fewer wrong calls on careless survey respondents, ranked no better, and so did not ship.
-
Every extra question is another look: raising the bar on long decision runs
Checking a stop after every answer lets lucky streaks end long runs as settled. A rising level cut those false results from 30 to 21 in 300.
-
Best and worst of four: a free MaxDiff mode, and why we didn't split it into pairs
A free MaxDiff mode for Quick Pick: best and worst of four, scored exactly. Fewer taps than pairs in 11 of 12 simulated cases; the exception says more.
-
Easy questions did not catch careless answers. Weighting the easy ones we already ask did.
Obvious check questions cost ranking accuracy and caught fewer careless respondents. Weighting existing answers by pair clarity caught more.
-
Flip, don't squint: comparing two photos on a phone
On a phone our photo ranker drew two landscape photos at 135 by 90 px. Stacking gives five times the area; a full-screen flip with a shared zoom shows more.
-
Is MaxDiff quicker than pairwise? Fewer taps, but the stopwatch decides
Best-worst screens of four took fewer taps than pairs in 11 of 12 simulations. Taps are not seconds, so the ranker now times real answers for a team.
-
Sort first, then compare: when a rough three-pile sort saves pairwise questions
A one-tap-per-option sort before pairwise picks cut 20 options from 93 actions to 64 for careful sorters; careless ones paid 120. Two bugs fixed on the way.
-
The question cap was picking the winner: a checkpoint instead of an end
Our free decision page ended at its question cap in 80% of simulated runs. Going on to 3x the cap named the true best 254 times in 300, not 203.
-
When your picks are close calls, the ranking shouldn't sound sure
The ranker told simulated near-coin-flip answerers their favourite was 56-62% likely when it was right 17-50% of the time. Now it refits first.
-
When every answer is "much better": reading the picks without the grades
People who answer "much better" to most questions made our decision page overconfident. A better threshold estimate did not fix it; ignoring their grades did.
-
Raising a drone's tilt limit without retuning: cap what the position error may ask for
Drones tuned at a 20° tilt limit crashed at PX4's 45°. Capping the position loop's proportional request removed every crash for 1–2% of cost.
-
Ranking pictures by picking pairs, without uploading a single one
Our pairwise ranker now takes photos and design drafts. The pictures never leave the device, yet a team can still combine picks: the file names are the list.
-
75 raters, 300 comments, and an alpha of 0.276: a free inter-rater agreement tool
Free no-upload Krippendorff's alpha, kappa, AC1 and ICC tool. On 300 GoEmotions comments, raters agree on sentiment far below any coding standard.
-
The blog's Tools topic was full of testing and SEO posts
Red-team, polish and SEO posts filled the blog's 'Tools and projects' topic. Two new topics now hold 416 of them; Tools drops from 2,876 posts to 2,462.
-
The shortfall had no units. Red-teaming the court reporter calculator
8 faults seeded into the court reporter calculator, 3 caught, 7 after four new tests. One real defect: an hourly shortfall printed as a bare number.
-
14:75 is not quarter past three. Red-teaming the service load check
16 faults seeded into the service load check, 11 caught. Three tools read an impossible clock time like 14:75 as a real one; a new CI check now tries them all.
-
A wall height nobody checked. Red-teaming the HVAC load and duct calculator
16 faults seeded into the HVAC load and duct calculator, 9 caught. A hand search found an unchecked wall height that gave a NaN heating load. Now fixed.
-
Zero rounds, zero dollars: red-teaming the mini-golf profit calculator
18 faults seeded into the mini-golf profit calculator, 15 caught. By hand: zero planned rounds read as loss-making, and free rounds put staff at 0% of sales.
-
"-$0.00" in 425 calculators. Red-teaming the stump grinding bid calculator
18 faults seeded, 9 caught. A money formatter signed the raw number but printed the rounded one; the same line was in 705 places in 425 tools.
-
Twelve open defects, twelve fixes already written: matching the red team's list to held PRs
All 12 defects the red team's open list showed on 2026-11-15 already had a fix in a pull request waiting on the owner. A new flag says so in one command.
-
The other receipt became a tip. Red-teaming the bookkeeper's receipt reader
16 faults seeded into the receipt reader and the chant reader, 6 caught. Splitting one card charge that paid two receipts posted the second receipt as a tip.
-
$0.00 is not a loss. Red-teaming the restaurant delivery commission analyzer
12 faults seeded into the delivery commission analyzer, 8 caught. A hand search found two break-even cases the page called losses, from float noise. Both fixed.
-
Red-teaming São Paulo's ISS rate table: 11 of 17 seeded faults caught, 6 holes closed
18 faults seeded into the NFS-e checker's ISS rate code, written on main the same day. The suite caught 11 of 17 real ones; six new tests catch all of them now.
-
A K code that took 77% of pay. Red-teaming the UK tax code checker
3 of 10 seeded faults caught. Two real errors fixed: K codes ignored the 50% overriding limit, and higher-rate taxpayers got Marriage Allowance codes.
-
One title line refused the whole file. Red-teaming the wash-sale checker
10 faults seeded into the wash-sale checker, 9 caught. A broker export with a title line above its header was refused whole, and line numbers skipped blanks.
-
Rehearsal tracks for every section of a split chant
Chant Forge now writes a part track and a minus-one track for each choir section, the learning tracks choirs usually buy or build by hand.
-
Retuning a drone at a 45° tilt limit bought nothing; a wider gain box bought a lighter, fragile one
Retrained at 45° tilt limit, co-designed quadrotors matched their 20° tune. Given a wider gain box, the optimiser built a lighter drone that needs its cap.
-
Rio de Janeiro's ISS rates on an NFS-e: 2% and 5% are always legal, 2.9% never is
Our NFS-e checker now reads Rio de Janeiro's own ISS rates (Lei 691/1984 art. 33). Its list is shifted against LC 116 the same way São Paulo's is.
-
Asked for robustness, a drone co-design kept the stiff gain and lengthened its arms
Given a wide gain box and an objective covering an uncertain centre of pressure, a co-designed quadrotor kept a high gain and some mass, and grew longer arms.
-
Robust PCA found the day the clocks changed without being told
Free robust PCA (principal component pursuit) on a year of hourly AEP load: the spring clock change ranks second and holidays barely register.
-
A status line every episode run had to edit by hand, now generated
Saints by Candlelight's notes hard-coded which episodes were text only, so each new episode made them stale. The generator now writes the line.
-
Saints by Candlelight 120: Egeria and the Friend at the Shrine of Thecla
Episode 120 of the bedtime saints podcast: a curious pilgrim goes three days out of her way to Seleucia and finds a friend from Jerusalem waiting.
-
Saints by Candlelight 121: Paul and the Nephew Who Walked into the Fortress
Episode 121 of the bedtime saints podcast: a council that ends in a brawl, forty men who swear not to eat, and a young man who walks into a Roman fortress.
-
Saints by Candlelight 122: Paul and the Governor Who Said Not Tonight
Episode 122 of the bedtime saints podcast: a hired orator, three charges answered in order, and a governor who hears the gospel and says 'not now'.
-
Saints by Candlelight 123: Paul and the King Who Said Almost
Episode 123 of the bedtime saints podcast: I appeal to Caesar, a governor with a blank letter, and the king whose famous 'almost' may not mean almost.
-
Saints by Candlelight 124: Paul and the House He Rented in Rome
Episode 124 of the bedtime saints podcast: a soldier chained to Paul, a day of scrolls with Rome's Jewish leaders, Onesimus, and the last word of Acts.
-
Saints by Candlelight 125: Timothy and the Cloak Left at Troas
Episode 125 of the bedtime saints podcast: Paul's last letter, Mark sent for after all, and a coat, some books and the parchments fetched before winter.
-
Saints by Candlelight 126: Titus and the Elders in Every Town
Episode 126 of the bedtime saints podcast: Titus left on Crete, a list of what an elder should be, a poet's insult, and the Midnight Mass reading.