Why a second vendor reviews every build
A model asked to check its own work tends to agree with itself. Here is what changes when the reviewer comes from another company, and what it still can't catch.
Ask a model to review the code it just wrote and it will usually find it reasonable. That isn’t laziness. It reads the change through the same assumptions that produced it.
The bug that started it
A card asked for sessions that last the number of days set in Settings. Claude wrote:
maxAge: days * 24 * 60 * 60,The tests passed. Express, though, reads maxAge in milliseconds: a 7-day session would end after about ten minutes. Codex, reviewing read-only, put the finding under line 42: F1 · P1 — maxAge is in milliseconds in Express. Claude fixed it, Codex read the fix again, and round 2 found nothing blocking.
What a different vendor brings
- Different blind spots. Two models trained apart rarely miss the same thing in the same way.
- No reason to agree. The Challenger didn’t write the code and can’t change it. Its only job is to find what is wrong.
- A record. Every round stays on the card: what was raised, closed or left open.
What it doesn’t replace
Your tests, your checks and your judgement. Plyloft runs your project’s checks at every Build done, and you decide when a card ships.
Draft for the launch. To review before publishing.