Beta Notice: Proximarket is in beta testing. All credits have no real monetary value and are for testing purposes only.
Open-ended text prediction markets with deterministic, objective payouts
Proximarket uses semantic embedding models to mathematically measure how close your prediction was to the actual outcome. The closer your prediction, the larger your share of the total prize pool. Here's how it works:
When a market resolves, the system compares your prediction text to the actual outcome using semantic embeddings. The cosine similarity is squared and then 0.25 is subtracted to create a minimum threshold, ensuring only reasonably accurate predictions earn significant rewards.
similarity_score = max(0, (cosine_similarity(your_prediction, actual_outcome))² - 0.25)
Your points are calculated by simply multiplying your similarity score by your wager amount. This straightforward calculation determines your share of the total prize pool.
your_points = similarity_score × your_wager
The total prize pool is distributed based on your share of all points earned. Your final payout depends on both your prediction accuracy and your wager amount.
your_payout = floor((your_points ÷ total_points) × total_prize_pool)
"What will be the top trending Google search in the US on December 15, 2025?"
Alice: 0.36 × 20 = 7.17 points
Bob: 0.15 × 30 = 4.41 points
Carol: 0.07 × 50 = 3.74 points
Total Points: 15.32
Total Prize Pool: 100 credits
Key insight: Alice made a profit of $26.78 (won $46.78 from a $20 wager) because her prediction was accurate. Bob lost $1.24 and Carol lost $25.56 because their predictions were less accurate than their wager amounts justified.
Note: The similarity scoring formula is subject to change and improvement as we conduct ongoing research into optimal prediction market mechanics. We may adjust the mathematical approach to better reward accuracy and create fairer market outcomes.