Knowledge, made liquid.

Most quiz apps reward luck, persistence, or attention — you tap, you earn, and the payout has nothing to do with whether you knew anything. Geek Protocol makes the reward a direct function of demonstrated knowledge: correct answers, answered quickly, verified on a server the player does not control, settled on Kaspa.

Three properties carry the design. Answers never reach the browser in any mode — a choice is committed to the server before the correct one is revealed, and the commitment is write-once, so learning the answer costs you your guess. Scoring happens server-side against a signed attempt token, so a player cannot report their own result. And payouts are queued and idempotent, so the same attempt cannot be paid twice.

The numbers below are read live from the alpha database every 60 seconds. They are real aggregates, including when they are small — a project asking you to connect a wallet should not be inventing its own traction.

When a player starts a round, the server selects the questions, shuffles the options for that attempt specifically, and issues an attempt token — an HMAC-signed payload holding the question set, the post-shuffle correct indices, and the server's own issue timestamp.

The client receives the questions and options. It never receives the answers. On submission the server verifies the token signature in constant time, scores the answers against the copy inside the token, and validates the timing against its own clock rather than the client's claim.

Settlement runs through a sharded queue. Payouts are distributed across independent lanes, each with its own worker and concurrency, so throughput scales with lanes multiplied by per-lane concurrency rather than being pinned to one transaction at a time.

Lane assignment is deterministic on the payout key. The same attempt always lands on the same lane, so a retry can never race the original on a different worker, and payouts for one player stay ordered relative to each other.

Per-lane queue depth is exposed at /health/payouts for monitoring.

From Level 10, players can write questions and review other people's. Approved questions enter the live rotation and earn their author a share each time they are served; reviewers earn a smaller amount per review.

This is the part of the protocol with the most obvious attack surface, since it pays people to produce and approve content. The safeguards are described below rather than glossed over.

One honest caveat: canvas rendering and behavioural signals raise the cost of a naive bot, but neither is a wall. Canvas is still glyphs and OCR reads it; telemetry is client-reported and therefore forgeable. They are treated as signals that flag an attempt for review, never as gates that withhold a payout, because a false positive there means refusing to pay a real player. The load-bearing defences are the server-side ones.