You can't run a prediction market on a database. A database has an owner, and an owner can be asked — or paid — to settle it wrong. That's not a bug in someone's implementation. It's the category.

A chain changes the question. The contract holds the pool, the arbiter signs the result, and everyone refunds if the arbiter goes silent. There's no "trust us, this settled correctly." You can check.

database:  "trust us, moon won"
chain:     "the result is signed, verify it yourself"

the arbiter is a single key

And that's the honest part: we say so. One key, clearly scoped — it can only sign a result after close, never touch the pool. If it goes silent, everyone refunds. The trust model is the product.

Prediction markets need a chain because the truth, once money depends on it, is an adversarial problem.