Accuracy.Fun

0%

How It Works

Precizion rewards precision, not just correctness. Here's how the system works end to end.

Continuous Outcomes

Traditional prediction markets are binary — you pick yes or no, and you either win or lose everything. Precizion is different.

You predict a single numeric value — like the price of Bitcoin at a future date, or the total volume of a token on a given day. The closer your prediction is to the actual outcome, the larger your share of the reward pool.

This means there are no “losers” in the traditional sense. Everyone who participates receives a share of the pool, weighted by their accuracy.

Accuracy-Weighted Scoring

Rewards are calculated using an inverse-distance decay function. The math is straightforward:

Step 1 — Distance
Distance(i) = |Predicted Score − Actual Score|
Step 2 — Accuracy
Accuracy(i) = max(0, 1 − (Distance(i) / maxError))
Step 3 — Weighted Score
Weighted(i) = Accuracy(i) × Stake(i)
Step 4 — Reward
Reward(i) = (Weighted(i) / Total Weighted) × Pool

The maxError parameter defines the boundary — predictions beyond this distance from the actual value receive zero accuracy (and therefore zero reward).

Pool Distribution

The entire reward pool is distributed among all participants. A small platform fee (2%) is deducted from the total pool before distribution.

Pool Split
98% — Reward Pool
2% — Platform

After the platform fee, the reward pool is split proportionally based on each participant's weighted accuracy score. There is no “house edge” beyond the 2% fee — the pool is fully distributed to participants.

Market Lifecycle

From creation to reward claim — every market follows this flow.

01

Market Created

Anyone can create a market with a target metric, end time, and min/max bet limits.

02

Betting Period

Participants submit numeric predictions with NIGHT token stakes. The pool grows with each bet.

03

Market Closes

No more bets accepted after the end time. The pool is finalized.

04

Resolution

The actual outcome is determined (e.g., BTC price from an oracle). The final value is recorded on-chain.

05

Reward Distribution

Rewards are computed off-chain, organized into a Merkle tree, and the root is submitted on-chain.

06

Claim

Participants claim their share by presenting a Merkle proof with a ZK proof — no private data is revealed.

Worked Example

Market resolves to 850. Pool is 400 tokens. Four participants bet 100 tokens each:

Alice
Predicted: 820
Distance: 30
Accuracy: 0.429
57.53 tokens
Bob
Predicted: 850
Distance: 0
Accuracy: 1
134.23 tokens
Charlie
Predicted: 880
Distance: 30
Accuracy: 0.429
57.53 tokens
Diana
Predicted: 830
Distance: 20
Accuracy: 0.714
96.13 tokens

Max error = 50% of 850 = 425. All predictions fall within range. 100% of pool distributed (after 2% fee).