Skip to content
CHRONOMANCY

The trust gap

Prediction markets have outgrown their toy phase. $13B per month in notional, 130× growth in 18 months, Fed-benchmarked, ICE-backed. And still, the ecosystem has three structural failures that the existing stack does not address.

1. No reputation layer

Section titled “1. No reputation layer”

Track records die with the platform. Nobody knows who’s skilled. The evidence says they should.

  • Top 1% capture 84% of gains — Akey et al. 2026 (N=1.4M users, $20B volume). Top 0.1% capture 58.5%. 70.8% lose money.
  • 3.1% of wallets take all $228M net profit — Yang 2026 (N=152.8M trades). Top-5% by rolling accuracy sustain 96.2% mean accuracy.
  • Only 7.3% of traders are “price-sensitive” — Bossaerts et al. 2022. Markets with 2–3 such traders had the best AUC.
  • Skill persists out of sample. Yang 2026: wallets classified as skilled in the first half achieve 89.4% accuracy in the second half vs 50.4% for experience-matched non-skilled. Atanasov 2024: test–retest reliability r=0.38 across seasons.
  • Selection outperforms aggregation by 21%. Atanasov 2024: 122 elite forecasters beat 404 sub-elite even with full algorithmic optimization. Simple median of 120 elites beats optimized aggregation of 400+ sub-elites. “Who forecasts matters more than what system they use.”
  • Capital is anti-correlated with edge. Deleep 2026: edge decays monotonically with participant size. Vocality has zero correlation with accuracy. Whales are not the elite.

The elite exist, they’re measurable, and the infrastructure to identify them across platforms does not exist.

Section titled “2. Resolution is the weakest link”

Every major PM scandal is a resolution failure. The Oracle Trilemma formally proves this is structural, not incidental.

  • Oracle Trilemma (Cong 2025): No blockchain oracle can simultaneously achieve decentralization, aggregate truthfulness, and scalability. This is an impossibility result, not a missing feature.
  • Scale of losses. Oracle manipulation attacks cost $403M in 2022 alone (Duley, via Sanjabi). MakerDAO Black Thursday: $8.32M liquidated at zero DAI. Compound malfunction: ~$89M bad debt. Pyth: 90% BTC/USD misreport.
  • Empirical risk is quantifiable. Ma 2026 (N=2,738 UMA events, 747 days): Adjudication Risk Index spikes correlate with voting-power concentration. When error probability exceeds 1/3, no bond parameter can restore deterrence.
  • Semantic risk is predictable from contract text alone. Sanjabi 2026 (N=804 UMA disputes): Semantic Risk Score predicts disputes (Spearman ρ=0.157, p<0.001). A 0.1 SRS increase reduces trading volume by ~60%.
  • Liquidity flight, not risk-adjusted pricing. Markets respond to ambiguity with withdrawal, not higher spreads. This is an Akerlof lemons dynamic — rational traders flee rather than demanding compensation. Enforcement gap is vast: zero CFTC insider-trading cases on event contracts as of 2024 (Ashar 2025).

Resolution risk is inherent, currently unpriced, and extensively quantified. It is the largest unaddressed category risk in the sector.

3. 72% lose money and leave

Section titled “3. 72% lose money and leave”

The optimal strategy is mathematically terrifying. The retention curve reflects it.

  • Full Kelly drawdown math (Osband 2025): 75% chance of a 25% drawdown, 25% chance of a 75% drawdown. Half-Kelly cuts growth 25% but drops the 75%-drawdown tail probability below 2%. Quarter-Kelly cuts growth 56%.
  • 72% of Polymarket traders lose money (Reichenbach 2025, N=124.5M trades). The loss share is increasing over time as new entrants arrive (60% → 72% from 2023 → Sept 2025).
  • Losses concentrate at extremes. 63% of the average user’s trades happen at prices below 10¢ or above 90¢ (Akey 2026). Kalshi contracts priced <10¢ lose >60% on average (Burgi 2025).
  • The barrier is multi-dimensional — loss aversion (Kahneman/Tversky; confirmed by Osband math), representativeness bias (Vaze: +10.43pp NBA dynasty mispricing), “yes-bias” / conviction bias (Deleep), AI lemons (He 2026), and fee friction (Burgi: 3.54% effective Kalshi fee).

A single-product response doesn’t solve this. A reputation layer + insurance + structured exit together can.

The white space

Section titled “The white space”

Zero of 264+ tracked prediction market projects build insurance, derivatives, or protocol-level reputation — PredictionIndex 2025–2026 annual report. The closest analogues are adjacent categories:

  • Gondor — lending on PM positions, $2.5M seed. Different layer.
  • Causal Markets / Sozu — information layers. Complementary, not competitive.
  • functionSPACE — resolution primitives. The closest in-thesis competitor; we monitor.
  • Chainlink — reputation infrastructure (oracle node reputation), but not forecaster reputation.

The financial-infrastructure layer is empty. Chronomancy is the first and currently only team building into it.

Next: the three products →