A deep analysis of a new forecasting primitive on Solana — and what it reveals about the limits of yes/no markets.

Everyone's building prediction markets wrong.

Polymarket. Kalshi. Augur. All of them force your belief into a single bit: up or down. Yes or no. The nuance that actually lives inside a skilled forecaster's head — I think BTC hits $98,400, not $97,000 and not $100,000 — gets flattened into a single bit of information and thrown away.

@trepa_io just shipped a mechanism that fixes this. It's small, fast, and deliberately narrow in scope right now. But the mechanism it introduces — precision payouts, not binary outcomes — is a genuine contribution to the design space.

What Trepa Actually Is

Strip away the crypto context and Trepa is a recurring estimation contest. Every round lasts 60 seconds: 30 seconds to submit your price prediction for Bitcoin, then 30 seconds until the round resolves at the actual BTC price. Entry costs $1 USDC. Winners get their dollar back plus a share of the losers' pool.

The twist is in how "winner" is defined, and how winners are paid.

Winning condition: Your error — the absolute distance between your prediction and the real price — must be below the median error of all participants in that round. So if 100 people play and the median error is $150, everyone within $150 of the actual price wins. No binary up/down call. No threshold you either clear or miss.

Payout: Winners don't split the pool equally. They split it by accuracy weight — a function of how much closer you were than the median. The person who predicted $98,412 when BTC settled at $98,415 earns vastly more than the person who predicted $98,550. Both won. But their payouts reflect the distance between their beliefs and reality.

This is the core mechanic. It sounds like a small change from "predict the direction correctly." It isn't. It rewires the incentive structure from the ground up.

Why Binary Markets Produce a Different Kind of Forecaster

To understand what Trepa is doing, you have to understand what binary prediction markets actually incentivize — and it's not what most people assume.

In a binary market, you buy a contract that pays $1 if an event happens and $0 if it doesn't. The contract trades between $0 and $1, and the price reflects the market's implied probability. You make money if your probability estimate is better than the crowd's.

This creates a specific game: find mispriced probabilities and bet against them. The skilled binary market participant is not trying to predict the future — they're doing relative arbitrage against the crowd's mistakes.

Here's what nobody says about this: your precision has zero economic value in a binary market.

If you think something has a 60% chance and the market prices it at 58%, you buy. Whether you privately think it's 60% or 90% doesn't change your trade. The extra precision in your model is worthless.

Binary markets have three structural problems that follow from this:

1. Precision is penalized — your exact belief doesn't matter, only your direction

2. Spread eats your edge on thinly traded questions

3. The information extracted is coarse — you get a probability, not a distribution

Trepa sidesteps all three problems by construction. Your precision has direct economic value. There's no spread. And the distribution of predictions is the game.

Mapping the Prediction Market Landscape

To position Trepa clearly, I mapped the ecosystem along two dimensions:

X-axis: Resolution timeframe — how long between forecast and settlement. Fast (seconds to minutes) on the left, slow (days to months) on the right.

Y-axis: Payout structure — how winnings are calculated. Binary (direction-based) at the bottom, precision-based (continuous, error-penalizing) at the top.

Top-left — Fast + Precision: Trepa The only significant occupant of this quadrant. One-minute rounds, precision payouts. Previously empty — no prior art in prediction markets at this intersection.

Top-right — Slow + Precision: Academic Scoring Rules Proper Scoring Rules (Brier Score, log score) have been used in forecasting tournaments like Tetlock's Superforecasters for decades. But they've never been productized at scale with real money. This quadrant is the theoretical ideal that nobody has shipped.

Bottom-right — Slow + Binary: Polymarket, Kalshi, Augur The dominant quadrant. $1B+/month in volume. Serious long-form forecasting questions — elections, macro events, regulatory outcomes. The forecasting quality can be excellent, but the incentive structure has the limitations described above.

Bottom-left — Fast + Binary: DeFi direction games High-speed binary games: predict up or down in the next minute. These function more like gambling than forecasting. No precision signal extracted.

The key insight: Trepa is not competing with Polymarket. Polymarket discovers long-run probabilities on discrete events. Trepa continuously samples the crowd's exact price belief at high frequency. Complementary primitives, not substitutes.

The Strengths of the Precision-Payout Model

Honest reporting is the dominant strategy. Economists call this "incentive compatible." In Trepa, submitting your true belief maximizes your expected return. No reason to shade your estimate. If you think BTC is $98,415 — you type $98,415. The game extracts real beliefs from real people.

Skill compounds differently. A player who is consistently 10% closer to the outcome than the median accumulates both higher win rates and higher per-win payouts. The skill signal is expressed twice per round. Over time, this creates a much steeper separation between skilled and unskilled players.

The crowd signal is richer. Trepa's aggregate is a shape — a full distribution. The spread of predictions tells you about uncertainty. Systematic clustering tells you about focal points. A binary market's price is a single number. This is a qualitatively better representation of what the crowd actually believes.

Solana enables the mechanic. Precision-payout, high-frequency forecasting would be uneconomical on Ethereum — gas fees would make $1 rounds impossible. Solana's throughput and fee structure are what make 1-minute settlement with on-chain resolution viable. The infrastructure genuinely enables the design.

The Weaknesses and Real Tradeoffs

Price prediction is a narrow domain. The precision-payout model works because the outcome is a continuous number. You can be "close" to a price. You cannot be "close" to "will Candidate X win the election." This constrains Trepa to financial data unless the team solves a harder generalization problem.

Sixty seconds is too short for most forecasting value. The people with genuine informational edge — macro analysts, on-chain researchers — have knowledge that operates on days and weeks, not seconds. At 60 seconds, the relevant information is almost entirely technical: price momentum, order flow, recent volatility. The crowd signal Trepa extracts, however precise, may be measuring short-term reflexes rather than genuine forecasting skill.

The pari-mutuel structure is a leaky bucket. In a round where the crowd is highly homogeneous, the precision advantage matters at the margins but the prize pool is thin. In a round with wide disagreement, the prize pool is fat but any individual's accuracy edge may be overwhelmed by noise. The variance of payouts is high in a way players need to internalize.

Skill discovery is slower than it looks. Half the field wins every round by construction. A player can run a hot streak through luck. Identifying genuinely skilled forecasters requires large sample sizes.

What This Model Contributes to the Literature

Prediction market theory has two major unresolved problems that the precision-payout model speaks to directly.

The calibration extraction problem. Markets are good at aggregating directional bets. They're poor at extracting calibrated probability distributions. The precision model aligns incentives so that expressing your true belief is the profit-maximizing strategy — the property that proper scoring rules have in academic settings, now applied to a real market with real money.

The thin market problem. Prediction markets on narrow topics often fail to aggregate meaningfully because there aren't enough participants. Trepa sidesteps this by being single-instrument at high frequency — one liquid, continuously active market. The depth comes from time, not from a wide question set.

The analog worth watching: what Uniswap did for token liquidity — one mechanism, infinite instruments — Trepa could theoretically do for precision forecasting. That requires solving the question-type generalization problem. It's a hard problem. The fact that Trepa hasn't solved it yet doesn't mean it won't.

The Honest Verdict

Trepa is a well-designed game with a genuine mechanism innovation. The precision-payout model is the most intellectually honest implementation of a real-money forecasting contest I've seen. It aligns incentives in a way that binary markets fundamentally cannot.

The weaknesses are real. It's domain-constrained. The 60-second horizon probably captures noise more than signal. The pari-mutuel variance will frustrate players who understand their edge but can't smooth returns.

But the weaknesses don't undermine the core contribution. Binary prediction markets have been the only game in town for 30 years not because they're optimal, but because they're simple to implement and regulate. Trepa's precision model is harder to explain — but easier to play honestly. That's a meaningful design achievement.

The question that will define Trepa's long-run relevance: can the mechanism extend beyond price data?

If yes — this is a new forecasting primitive. If no — it's an excellent niche game for traders.

Either way, it's worth understanding. The prediction market design space has been stagnant for years. A team taking a genuine swing at a different mechanism is rare enough that it deserves serious attention.

What's the hardest question type to apply precision forecasting to — where "how close were you" is genuinely difficult to measure? Drop your answer in the comments.

Docs: docs.trepa.io | @trepa_io on X Early access: trepa.app/waitlist