Source checks
Provably fair is not a feed of every Stake bet
What Stake’s fairness verification proves, what it cannot prove about global betting activity, and why that matters for prediction markets.
Verify a known result
Stake’s published implementation combines a server seed, a client seed, a nonce and a cursor to generate a verifiable result. The server seed is committed before play and revealed after the seed pair is rotated. With the required inputs, a player can check the calculation.
Stake Engine also publishes a verification interface for compatible games. This is useful evidence about a particular result. It is a different kind of evidence from a complete, timestamped record of all betting activity.
Stake’s published fairness implementation ↗One verified bet cannot answer every question
Suppose a prediction asks whether anyone wins more than a threshold before midnight. One valid, timestamped qualifying bet can establish that the event happened, provided it satisfies the published conditions. One hundred non-qualifying examples cannot establish that nobody else won.
A NO answer requires evidence covering the relevant activity and period. A clip, a screenshot or a sample of large bets can miss the very event the market asks about. Absence from a sample is not proof of absence from the platform.
Match the question to the available proof
Before publishing, separate the event you want to predict from the evidence you can actually collect. A prediction about a named, accessible result has different requirements from a claim about every player on Stake. Private account data also cannot be assumed to be publicly available.
For automated resolution, define the source, identifiers, timestamps, coverage and failure rule in advance. Prefer a precise observable over an exciting claim that cannot be checked. The payoff is a result people can understand and independently inspect.