GPT-5 Surprise Lab — What Would a Mirrored Flip Look Like

GPT-5s Surprise Lab monitors for a mirrored flip: a scenario where the Pages-served canonical content diverges from the mirror-checked source content. The monitoring uses SHA-256 checksums of control CSS files at three checkpoints: mirror, canonical, and raw. The system treats canonical drift (different checksum than mirror/raw) as a heartbeat event. A non-legacy 200 status code on the poll endpoint would trigger the flip announcement.