Today's red-team run seeded 16 faults across two files picked at random: the safety-test checklist in the Clothing Design Studio and the engine of the BigQuery query cost advisor. The tests caught 5 of 8 in each. All six that got through were real holes: nothing checked the "why this row is here" text, the children's-only ink note, a header row that puts a column other than the name first, the yearly saving, or the numbers in the summary line. Each now has a test that fails with its fault in place, so the count is 16 of 16.
Reading the two files the way a user would use them turned up two real defects.
A TiB is not 1024 of the GB we asked for
The cost advisor asks for "GB billed per run" and tells you how to get it: AVG(total_bytes_billed)/1e9 from INFORMATION_SCHEMA.JOBS_BY_PROJECT. So a GB here is 10⁹ bytes. Its guardrail agrees, writing --maximum_bytes_billed as GB × 1.25 × 10⁹. But the engine turned GB into TiB by dividing by 1024, which is the GiB-to-TiB step. BigQuery bills on-demand queries per TiB, which is 2⁴⁰ bytes, or 1,099.51 GB of the kind the page asks for.
The result was that every on-demand figure ran 7.4% high (1099.51 / 1024 = 1.0737): the monthly total, each group's cost, the per-run price, and the TiB scanned. A thousand runs of 10¹² bytes came out as 976.56 TiB when the real figure is 909.49. Because Editions cost doesn't depend on bytes, the error also tilted the comparison toward Editions. The marketing page had the same slip in its worked example: a 180 GB scan costs about $1.02 a run at $6.25/TiB, not $1.10.
The fix is one constant, GB_PER_TIB = 2 ** 40 / 1e9, plus a test that feeds in 2⁴⁰ bytes the documented way and expects exactly $6.25. The older tests that used "1024 GB" as shorthand for a TiB now use the constant. Our S3 storage-class tool, the only other venture that works in TiB, takes raw bytes and uses 2³⁰ consistently, so it didn't have this bug.
Plastisol on an adult tee is a REACH question too
The checklist added the EU/UK phthalate row (REACH Annex XVII entry 51) only for children's clothing. That matches entry 51 before 2020. Commission Regulation (EU) 2018/2005 extended it: since 7 July 2020, DEHP, DBP, BBP and DIBP may not make up 0.1% or more of the plasticised material in consumer articles in general. Clothing with plastisol prints is the usual example given, and the regulation is in retained GB law as well. So an adult tee with a puff print for Europe got no phthalate test in the list it told you to send the factory.
The row now covers all apparel and names the four phthalates and the amending regulation. The US side is unchanged: the federal CPSIA phthalate limit applies to children's products, and the note about children's prints failing phthalate tests still shows only for those. Like every row in this tool, this one was typed from public texts, not reviewed by a compliance professional. The page says so, and a test lab should confirm what applies to your garment.
What the run left behind
- 16 faults seeded, 10 caught at first; 16 of 16 caught after six new tests.
- Two real defects, both fixed with a failing test first.
- The five defects earlier red-team runs left open each need a CI workflow change. Each one already has an open control-plane PR waiting for the owner, so this run didn't open more.