- Cultural correctness100
- Safety95
- Text legibility95
- Aesthetic quality95
- Forward appeal85
- Brand fit·
- Prompt adherence·
Passed everything. The publish gate can ship these.
Something's off — wrong text, missing motif. Waits in the inbox for a person.
Cultural or safety fail. Code keeps it locked; only an audited human override opens it.
The game · you vs the QA Agent
You be the judge.
Five rounds, four cards each. In every round we planted one mistake on purpose — some you'll spot instantly, some are the kind only a devotee (or the QA Agent) catches. Pick the card you'd flag.
- 1
- 2
- 3
- 4
- 5
Four Diwali cards fresh off the line. One should never ship.
Which one would you flag?
How it works
Nobody can look at thousands of cards a day, so the QA Agent takes a fair sample: a few from every batch, the never-checked ones first, and enough of them that the answer actually means something. Each card is then read seven ways — is the text spelled right, does it look like the god it names, is it culturally right, is it safe, is it beautiful, is it what was ordered, would anyone forward it. A judge weighs those seven and calls it: ship, flag for a human, or lock. Anything culturally or religiously wrong is locked by the code itself rather than by the judge's opinion — a model does not get the last word on somebody's faith. What passes gets published in one go, what didn't is held back, and the whole batch can be rolled back.
Why it exists
Pretty pictures nobody forwards are worthless. Wrong pictures that people do forward are worse — that is somebody's god spelled wrong, on their mother's phone, forwarded to forty people. So: sample honestly, measure openly, judge in the open, veto in code, and always leave a person able to overrule.