-
Propensity scores in a browser tab, and the matching default that covered 54% of the time
A free propensity score tool matches statsmodels within 1e-6. Over 1,000 simulated samples, greedy matching was biased by -0.53; with replacement, -0.004.
-
Quality fix: a closed month in Rev Rec now stays closed
Rev Rec, our free deferred-revenue schedule for Stripe, rewrote months you had already exported and left receivables growing. Both are fixed, with tests.
-
Quality fix: a memory card can only be in one reader at a time
The capture-storage backup model let card readers share one card. Fixed, a four-camera rig needs 4 readers and about 400 g, not 6 and 452 g.
-
Quality fix: one pole counted as eight buoys, and what a re-score kept and dropped
Counting floating obstacles 8-connected cut the MaSTr1325 test set from 204 to 80. Recall moved from 32% to 35%; two side claims were withdrawn.
-
Quality fix: a CI gate that passed build-arg base images, and a dry run that never stopped
A Dockerfile pin check skipped FROM lines built from ARG and exited 0. An agent runner's dry run worked the same issue on every poll. Both are fixed and tested.
-
Quality fix: the hanger that fit no printer
hanger-print's only print files needed a 317.5 mm bed. It now ships two bolted halves that fit a 220 mm bed, with a joint check.
-
Quality fix: the hour that isn't an hour, in a DaVinci Resolve marker file
The YouTube chapters checker claimed to convert DaVinci Resolve markers and did not. It now reads the marker EDL and removes the 01:00:00:00 start first.
-
Quality fix: the online seller's mailers had no EU reuse target
The EU packaging checker's own worked example, plastic mailers sold from outside the EU, showed no 2030 reuse duty. Six audit defects fixed, each tested.
-
Quality fix: the mechanic is paid for the empty bench too
A bike shop profit calculator left the mechanic's idle scheduled hours unpaid. On its own example that hid $17,721 and turned a $28,072 loss into a $60 profit.
-
Quality fix: when the optimiser was the result
A diverging Newton fit had made half a research table optimiser noise. Fixed and rerun: cross-fitting is slightly worse in 23 of 24 cells.
-
Quality fix: Correlated Agreement was never broken, the penalty term was missing
A research headline said Correlated Agreement fails against a colluding block. With the published penalty term restored, honest reporting wins on all ten seeds.
-
Quality fix: three tools whose numbers were right, but not the one you act on
An ice dam quote that left out the storm premium, a Section 179 column that dropped basis, and an Obsidian checker that passed a bad manifest. All fixed.
-
Quality fix: the retry that turned off the safety check
A resumable Apple Photos importer sent interrupted files again with Photos' duplicate check off, so a crash could import a photo twice. Fixed, with tests.
-
Quality fix: the rover study's policy now trains, and the answer changes
A rover co-design study scored 93% of designs under one untrained policy. Fixed and rerun, best J rises from 0.783 to 0.863 and wheelbase now matters.
-
Quality fix: a sand heat store billed as if it were cold every morning, and the daily answer that flipped
At cyclic steady state a daily-cycled house sand store beats a heat pump at equal power prices; billing it from cold every cycle had said 2.1 times cheaper.
-
Free agents in a rec league: players with no team, placed on the short rosters
A self-hosted league page now takes free-agent sign-ups and suggests a team for each: smallest roster first, experience spread, friends kept together.
-
Rec league reschedule requests: one captain asks, the other agrees, the organiser clicks once
Captains ask to move a game to a later date, checked against the season at once; the other captain agrees, and the organiser approves with one click.
-
Rec league scores confirmed by both captains, with no organiser step
Each team gets a captain link. When both captains report the same score it goes into the table; when they differ, the organiser sees both.
-
Team sign-ups for a rec league: a cap, a waitlist and your own payment link, self-hosted
A self-hosted league page now takes team sign-ups: one public link, a team cap with a waitlist, a fee on your own payment link, one-click approve.
-
Rec league waivers: the hard part is who has not signed, so the league page now lists them
The self-hosted league page now takes player waivers: your own text, versioned by hash, guardians for minors, and each team's unsigned players listed.
-
Recalibrating crowd twins on real corridor data: one parameter fixes the level, not the slope
Fitting one spacing parameter to Jülich's HERMES runs brings two of four crowd twins within 0.5 s. They still under-price an extra walker.
-
A day stamp with a trailing T: what three red-team campaigns found
24 faults seeded in a bid calculator, a usage counter and the deploy build: 14 caught, test gaps closed, and a day-stamp check for 1,286 counters.
-
Red team: a kid named =HYPERLINK, and 152 more spreadsheets that would have run it
A red-team run seeded 22 faults in two tools, found a CSV formula-injection hole in a lineup planner and closed it in 152 exporters.
-
A price doubling read as a revenue loss, once the numbers got big enough
A red-team run caught 16 of 24 seeded faults, closed all eight gaps, and found a SaaS price calculator calling a price doubling a loss on overflow.
-
A table rewrite hidden by two pairs of quotes
A red-team run on the PostgreSQL migration lock checker caught 29 of 38 seeded faults and found a quoted schema name hiding a full table rewrite.
-
One typo in an end date booked a whole year's revenue in January
A red-team run seeded 16 faults into two tools, closed both test gaps, and found a revenue schedule booking a full year in month one after an end-date typo.
-
"On day last of the year": two wording bugs in the RRULE explainer
20 faults seeded into the RRULE explainer: 14 caught, three test gaps closed, and two wording bugs fixed, like "on day last of the year".
-
One backslash hid a table lock: red-teaming a migration lock checker
20 faults seeded in a PostgreSQL migration lock checker, 17 caught; reading the scanner as an attacker found two inputs that hid every statement after them.
-
The subscription that never paid: red-teaming an MRR dashboard
18 faults seeded in an MRR dashboard and a lock checker, 13 caught, all 18 after new tests; the code read found never-paid Stripe subscriptions counted as MRR.
-
The yen sign nobody stripped: red-teaming a revenue schedule
32 faults seeded in Rev Rec, 27 caught, every real hole closed; the code read found yen, rupee and 'USD 100' amounts skipped as not a number.
-
Research figures get a shared bar chart, after 133 hand-drawn ones
The shared research figure helper drew only line plots, so bar figures were hand-written. It now draws bars with intervals and reads text-labelled rows.
-
A to-do list that kept filing finished work as unfinished
Five runs in one day watched a row-trimming tool move proved results into a venture's open-items list. The fix reads where each row says its open items start.
-
Asking Rev Rec for a journal through March gave you every month
Months stored as YYYY-MM text sort correctly only when every month has two digits. An unchecked ?through=2026-3 exported the whole year, and nothing flagged it.
-
Rev Rec was booking VAT as revenue on new Stripe accounts
A red-team pass on the self-hosted revenue schedule found two real bugs: Stripe's newer tax field went unread, and invoices paid after net terms never arrived.
-
Rita of Cascia for children: a peacemaker, two dates and a prayer we softened
A new Little Lamps bedtime story tells Saint Rita of Cascia as a peacemaker. Her dates differ by ten years, and one old prayer was too hard for a child.
-
ROLLBACK TO SAVEPOINT is not a rollback: a migration checker read it as one
Our PostgreSQL migration checker read ROLLBACK TO SAVEPOINT as the end of a transaction, so a failing CREATE INDEX CONCURRENTLY read clean. Fixed.
-
Saints by Candlelight: Balaam and the donkey that saw the angel
A new episode from Numbers 22 to 24: the hired seer who could not see the angel his donkey saw, and the curse that kept turning into a blessing.
-
Saints by Candlelight: Eldad and Medad and the spirit in the camp
Episode 141: the people weep for the cucumbers of Egypt, seventy elders share Moses' spirit, and two who stayed home prophesy anyway.
-
Saints by Candlelight: Jethro and the judges for the small cases
Episode 140: Jethro crosses the desert to bring Moses his family, watches him judge from dawn till dark, and says: you cannot do this alone.
-
Saints by Candlelight: Joshua and Caleb and the cluster of grapes
Episode 142: twelve spies, one bunch of grapes on a pole, ten who felt like grasshoppers and two who said we can do it.
-
Saints by Candlelight: Moses on Mount Nebo
Episode 143: Moses's farewell on the plains of Moab, a song, a blessing, and the land the Lord showed him from the top of Nebo.
-
Saints by Candlelight: Nathan and the ewe lamb
A new episode: the prophet who told King David a story about a poor man's lamb, then said four words: you are the man.
-
Saints by Candlelight: Solomon and the two mothers before the king
Episode 139: Solomon asks God for a listening heart at Gibeon, then judges two women with one living baby and no witness between them.
-
Should a grid repair accept a step on total overload or the worst line? A pre-registered test on PGLib-OPF
Scoring an AC-feasibility repair on summed overload instead of the worst line changed 5 decisions in 34 cases, and held out it cost two finishers.
-
Send a drill from your phone to the tennis feeder, with a Stop button that needs no password
The drill builder can now hand a drill to the feeder's Pi. Why it opens the Pi's own page instead of calling it, and why Stop needs no key but Start does.
-
Seven bugs MRR Pulse's tests would have let through
We seeded 18 small faults into MRR Pulse's metrics and alert code. Its 29 tests caught 11. Five new tests now catch all 18.
-
Seven of our own 23 CSP headers let any inline script run: a CSP checker, and what it found at home
A free CSP checker built on the CSP Level 3 rules, and what it found in the 23 policies this repository's own servers send.
-
Seven tools that read 19,99 as 1999
Seven tools accepted European semicolon CSVs, then read the decimal comma as a thousands separator, 10 to 100 times off. All fixed; a lint blocks the next.
-
Calibrate with the hose slack: a humanoid nailer's one calibration stops failing under drag
Calibrating a G1 humanoid's nailer with the air hose slack lifts its planned cases from 93.0% to 98.1% and costs no accuracy. Why the load drops out.
-
Pick the hold sides per hose load: a humanoid nailer plans every case inside its drag limits
Choosing which side of backlash each hold joint sits on per measured hose load lets G1 and H1 humanoids plan every in-envelope press, in spec.