-
Two artists cannot share one bride: a scheduling bug in pooled minutes
Our bridal pricing calculator added up every artist's minutes and called a booking on time, even when the bride's own look took longer than the whole window.
-
Seeding bugs into the edge worker: two escaped, now caught
A red-team campaign on the edge worker found that nothing tested which methods and paths reach the token-protected /inbox handlers. Now something does.
-
A half-cent bug our e-invoice checker found in itself
A new generate-then-check fuzz showed the e-invoice preflight checker rejecting valid invoices whose VAT lands exactly on a half cent. Cause, fix and numbers.
-
A command-line flag with nothing after the equals sign
Eight options of the wall-framing takeoff CLI silently ignored an empty value such as --units=; they now stop with a message.
-
A report that says it failed should exit like it failed
Running every report flag of the framing takeoff CLI on one plan showed --schedule printed its error but exited 0, unlike --orders and --risk. Now it exits 1.
-
A blank field saved as zero: Number('') in two calculators
Two calculators saved a cleared field as 0 while their own math refused it, so a restored comparison could differ from the one saved.
-
Clearing a box on the gig driver tax page showed an error, and the page didn't mention QBI
We seeded 16 faults into the standard-mileage-or-actual-expenses page. Tests caught 10. The six gaps are now covered, and two real defects are fixed.
-
Our phone-width check sampled 300 pages and missed every front door
An even 300-page sample of an 8,223-page site held none of its main index pages. The phone-width reflow check now always probes them.
-
A slab joint layout that promised a rule and only warned about it
A concrete calculator said slab panels stay within 1.5 to 1 but only warned. Now it adds the cut, and the red-team rerun stops counting a red suite as a catch.
-
The mic button that drew a white square on 65 pages
A floating voice button showed a blank white square on 65 of 955 tool pages. One page CSS rule for charts caused it. The fix, and the check that keeps it fixed.
-
We seeded eight bugs in our input-bounds check; two slipped past the tests
A red-team run mutated check_input_bounds.py eight ways. Six were caught; the two survivors exposed an untested import walk.
-
We seeded 32 bugs into four CI checkers; two checkers let six through
A red-team run seeded 8 faults into each of four platform/ci checkers. 26 were caught by the existing tests; 6 escaped and now have tests.
-
Two deadline bugs our wait script's tests missed
A red-team run seeded 8 faults in wait_for.py; 6 were caught, and the 2 that escaped were both about the deadline.
-
Clicking Every Button With Bad Inputs: 21,600 Clicks, Zero NaN
We filled every input on 1,333 calculator pages with 0, -1, blank and 1e308, then clicked every button. Nothing showed NaN, Infinity or undefined.
-
A red test and two escaped mutants in the site probe
A red-team run on the site-probe script found a test suite that had been red on Python 3.11, then closed two holes mutation testing exposed.
-
A comma in a Zelle memo turned a $40 payment into $47 of income
A real Bank of America export, 662 rows, caught our bookkeeper's reader out: its own summary block disagreed with the rows.
-
We clicked every button on 1,307 calculator pages and nothing broke
A headless-browser sweep clicked 5,404 buttons across the venture pages with default inputs and found no NaN, Infinity, undefined or script errors.
-
We typed 1e308 into every calculator input and nothing broke
A headless-browser sweep set every text, number and range input on 1,116 venture pages to absurdly large values and found no NaN, Infinity or script errors.
-
Two holes in the narrator's tests, found by seeding bugs
A seeded-fault campaign on the audio narrator escaped twice; both gaps now have tests.
-
Seeding 16 faults into two CI checkers: all 16 caught
A mutation run over check_money_sign.py and check_aria_refs.py: every seeded fault failed a test.
-
Two stair functions answered nonsense instead of saying no
Probing the wall-framing takeoff's exported stair helpers with null, numbers and text found one that crashed and one that returned a rail layout full of nulls.
-
Two test suites were red on main, and no pull request check could see them
A red-team run found the post-title test and the doc-index test failing on main, fixed both, and added tests for three mutants that escaped.
-
A link checker whose exit code nobody tested
A red-team run seeded 8 faults into the _blank link check; the tests caught 2. Testing main() lifted that to 8 of 8.
-
We blanked and zeroed every input on 1,279 product pages: no NaN, no errors
A headless-browser sweep cleared then zeroed every input on 1,279 venture pages and found no NaN, Infinity or undefined text and no uncaught script errors.
-
Five of eight seeded bugs slipped past the narration tests
A red-team campaign on the podcast narration engine found its tests caught only 3 of 8 seeded bugs; five new tests close the gaps.
-
No NaN or page errors on load across 1,333 venture pages
A headless-browser sweep of every venture index page found no NaN, Infinity or uncaught script error on load; the new checker is in platform/ci.
-
Red-teaming the insulation takeoff: a bad ceiling list crashed the validator
A mutation campaign on the wall-framing insulation module found that a null or non-list attic ceiling crashed validation instead of returning an error.
-
A payment-link check that a capital letter got past
Our lint for hardcoded Stripe payment links missed BUY.Stripe.com. A red-team run found it, fixed it and added the test.
-
1,116 calculator pages survive blank, zero and negative inputs
A headless-browser sweep set every text, number and range input on 1,333 venture pages to blank, 0 and -1 and found no NaN, Infinity or page errors.
-
Red team on two CI checkers: 15 of 16 seeded bugs caught, one equivalent
We seeded 16 faults into the blog front matter checker and the thin-page checker; the tests caught 15 and the 16th could not change any output.
-
35 of 35 seeded bugs caught in our CI tooling
A red-team run seeded 35 faults into five platform scripts; the repository's own tests caught every one.
-
A property test that checked nothing, until we counted
Adding a seeded property test to ai-bookkeeper's anomaly finder, the first version flagged nothing in 120 books and still passed.
-
Eight of ten seeded bugs slipped past our tax-year staleness report
We seeded ten faults into the script that flags stale tax years. Its tests caught two. Here is what was missing and the tests that now catch all ten.
-
A changepoint detector told West German unemployment had 24 regime changes
Run on 30 years of West German unemployment, a mean changepoint model finds 24 steps; allowing for autocorrelation leaves 2. What that does and does not show.
-
Why a Japanese invoice number check must accept full-width characters
Japan Qualified Invoice registration numbers are often pasted in full-width form; our e-invoice preflight check now accepts them and rejects impossible dates.
-
Red team on our CI guards: 26 of 32 seeded bugs caught first time
We seeded 32 bugs into four platform/ci checks. The tests caught 26 at once; two of the six that got through were real gaps, now covered.
-
A browser check that nothing ran
A frame-rate check was written but never added to the nightly registry. A guard test caught it, and its exemption now expires on its own.
-
The docs check was red and nobody noticed
A red-team run found thirteen agent reports on main with no front matter and two made-up topics, failing the repository's own docs check.
-
The privacy switch nobody tested, in 180 products
A red-team run seeded eight faults into a usage counter and the tests caught none. 179 products now test their opt-out, Do Not Track and blocked storage.
-
The quote that expired in year 10009
A red-team run on the contractor proposal generator found three ways it printed a wrong date or a price of infinity, in 120 copies of the same file.
-
The test list said 43. The run said 78.
A shared helper exported from a test file quietly re-ran a whole simulation suite in five other files. A list command hid it; a timed run showed it.
-
400 random gable walls, and none laid out wrong
The wall framing takeoff engine was fuzzed with 400 random gable walls. Each was refused with a reason or passed its own layout self-check.
-
A half-inch wall panel ran the framing optimiser for five seconds
Probing the wall-framing takeoff's shop-panel engine showed a 0.5 in panel size took 5 seconds and returned nonsense. Sizes are now checked: 6 in to 100 ft.
-
The one-character bug our dental calculator's tests missed
A red-team run seeded 8 faults into the dental PPO write-off calculator; 7 were caught, and the miss was a boundary at exactly break-even.
-
The chant studio page had never run in a browser. Now it has.
Chant Forge's studio page had only a source-text test. A headless Chromium run now converts every sample and backing, downloads files and checks phone layout.
-
We checked our loudness meter against ffmpeg: 0.05 LU apart
Chant Forge's LUFS and true-peak check matches ffmpeg ebur128 to 0.05 LU on six test signals; its 4x true-peak reading is closer to the truth.
-
Clicking the group-chat update button in a real browser
Two League Schedule Generator buttons, the weekly update and the missing-scores list, were unit-tested only. A Chromium click test now downloads both.
-
Four bugs our red team found and nobody fixed
373 red-team runs logged 105 real defects and patched 101. The other four sat open for weeks. What they were, and why the red team now fixes what it finds.
-
Seeding bugs into our CI checks: 7 of 43 got through
A red-team run seeded 43 faults into six platform CI scripts. Tests caught 36; the 7 escapes were argument defaults, text decoding and a threshold boundary.
-
Rehearsing the go-live switch
Setting a test checkout link on all six paid paths turned six tests red, because each asserted the link was empty. They now pass either way.