Research: Market Microstructure
Market Microstructure
Section titled “Market Microstructure”Understanding how prediction markets actually behave at a structural level — not what the surface-level volume numbers say, but what’s underneath them.
The Volume Reality
Section titled “The Volume Reality”Double-Counting and Wash Trading
Section titled “Double-Counting and Wash Trading”Most prediction market volume statistics are inflated:
Polymarket double-counting: Each trade generates two OrderFilled events on-chain. Most dashboards (Dune, CoinGlass) count both. Paradigm’s December 2025 analysis found reported Polymarket volume was ~2× actual. Frequently cited “$22B 2025 Polymarket volume” → ~$11B actual.
Wash trading: Columbia University study found approximately 25% of Polymarket volume is wash trading — artificial volume inflated by participants trading with themselves. This peaked at ~60% of volume in December 2024.
Corrected 2025 industry total: ~$34B actual (Polymarket ~$11B + Kalshi $22.88B + others). Most citations of $44–64B total are materially inflated.
Why this matters for infrastructure builders: A protocol sizing itself against inflated volume numbers will miscalibrate. The real addressable market is meaningful but smaller than commonly stated.
Concentration
Section titled “Concentration”| Metric | Value |
|---|---|
| Markets with zero 24h volume (Polymarket) | 63% of active markets |
| Markets with lifecycle <7 days | 67.7% of all created markets |
| Top 505 markets (by volume) share of total volume | 47% |
| 2024 markets created/month peak | ~13,800/month (Aug 2025) |
The prediction market ecosystem is extremely concentrated. A tiny fraction of markets by count generates the majority of volume. The long tail of user-created markets is almost entirely illiquid.
This has direct implications for Fast-Forward vault design: the vault only operates on liquid markets. “All prediction positions” is not the addressable universe — only positions in the top 200–500 markets by OI at any time.
Kalshi vs. Polymarket: Different Animals
Section titled “Kalshi vs. Polymarket: Different Animals”Despite being grouped together as “prediction markets,” Kalshi and Polymarket have fundamentally different user bases and revenue structures:
| Kalshi | Polymarket | |
|---|---|---|
| Monthly volume (Feb 2026) | $9.8B | $3.5B actual (~$7B reported) |
| Valuation | $11B | $8–9B |
| Sports % of volume | 89% | 37% |
| Key regulatory fact | CFTC-regulated FCM | Crypto-native, offshore |
| 2025 fee revenue | $263.5M | Significantly lower |
| Largest backer | CFTC approval | ICE (NYSE parent), $2B |
| Primary user type | Sports bettors | Crypto-native, political traders |
Key insight: Kalshi’s 89% sports volume makes it a regulated sports betting platform that also does prediction markets — not a prediction market platform with sports. The 5.1M MAU figure is driven primarily by sports gamblers, not forecasters.
This matters for Chronomancy because:
- FF vault targets the forecasting-intent segment, not sports gamblers
- REWIND insurance is more valuable to users making probability-based decisions than to sports bettors making directional bets
- Chrono Score is meaningless for sports betting (no forecasting skill being measured)
The addressable base within Kalshi is the non-sports segment — probably 10–15% of users, but the more sophisticated and financially engaged segment.
The Slippage Problem
Section titled “The Slippage Problem”Execution in prediction markets is not frictionless. Blended slippage on CLOB-based markets (Polymarket):
- ~1.3% on vault positions (buy-side market impact)
- Zero maker fees on Polymarket (limit orders at or better than mid)
- Slippage reduces base-case FF net spread: modeled 4% gross → ~2.7% net (using limit orders to mitigate)
- Best mitigation: limit orders routed through Polymarket’s 3,200 order/sec CLOB
For large positions (>$10K), slippage compounds. The vault’s execution strategy must be adaptive — smaller positions can fill at market, larger positions require limit order patience.
Implication for financial model: The 4% net spread in the base case is achievable only with disciplined execution using limit orders. Market orders produce materially worse economics. This is an operational risk, not just a modeling assumption.
The Siskind Cube
Section titled “The Siskind Cube”Scott Alexander’s “Siskind Cube” maps prediction market platforms on three axes:
- Real money vs. play money
- Platform-created markets vs. user-created markets
- Easy to use vs. complex
The full matrix:
| Platform | Real money | User-created | Easy to use | Resolves | Gap? |
|---|---|---|---|---|---|
| Polymarket | ✓ | ✗ (curated) | ✓ | Platform | |
| Kalshi | ✓ | ✗ (curated) | Partial | Platform | |
| Manifold | ✗ | ✓ | ✓ | User | Money gap |
| Augur V2 | ✓ | ✓ | ✗ | Community | UX gap |
| Unoccupied | ✓ | ✓ | ✓ | User+AI | ← Gap |
No platform occupies all four corners. This is the Siskind Cube gap.
Chronomancy’s positioning: The protocol isn’t trying to build a new prediction market to fill this gap directly. It wraps existing platforms to create the economic layer that the gap lacks — insurance, capital efficiency, reputation scoring. The wrapping approach means the gap can be addressed without requiring new oracle infrastructure or regulatory approval for each market.
The Siskind Cube gap is the long-term vision. The FF vault + REWIND insurance + Chrono Score is the near-term execution.
Power User Economics (The 1% Problem)
Section titled “Power User Economics (The 1% Problem)”1% of users generate 70% of prediction market volume. These power users have specific behaviors and needs:
| Characteristic | Implication |
|---|---|
| Trade across many markets simultaneously | Need portfolio-level risk management, not position-level |
| Capital deployment constrained by market resolution timelines | FF vault is directly valuable |
| Track record meaningful to them | Chrono Score has direct value |
| Losses are portfolio management failures, not emotional events | REWIND is a hedging tool, not emotional insurance |
| Use APIs, not UIs | Smart contract interface is the correct channel |
The 99% (casual users) are important for market liquidity but churn rapidly. The 1% (power users) are the retention anchor and the protocol’s primary revenue source.
FF vault sizing calibration: At Polymarket + Kalshi combined OI of ~$800M (Q1 2026), and assuming the top 1% holds ~50% of positions in liquid markets, the FF-addressable universe at any given time is approximately $200–400M in notional. 5–15% of that wanting early exit = $10–60M in addressable FF volume per month at scale.
This is why the base case (M12 monthly FF volume $480K) is conservative: it assumes only a tiny fraction of the addressable base is reached in year one.
AI Agents: The Emerging User Class
Section titled “AI Agents: The Emerging User Class”AI agents are already active in prediction markets:
| Platform | Activity | Performance |
|---|---|---|
| Gnosis Chain (Olas protocol) | 300+ daily active agents, 340K+ monthly transactions | 65–79% prediction accuracy |
| Polymarket (Polystrat) | 4,200+ trades in first month | 37%+ positive P&L vs ~15–18% for humans |
| Peak Gnosis days | Agents = 75%+ of all SAFE transactions | — |
ForecastBench (ICLR 2025): best AI Brier score 0.101 vs. human superforecasters 0.081. Projected AI-human parity: late 2026 (95% CI: Dec 2025 – Jan 2028).
The implication for infrastructure: AI agents don’t use Robinhood’s UI. They need:
- Smart contract interfaces they can call programmatically
- On-chain reputation (Chrono Score) to distinguish quality agents from noise
- Capital efficiency tools (FF vault) to manage portfolio positions
- Insurance (REWIND) for risk management across large positions
AI agents aren’t a threat to Chronomancy’s model. They’re the model’s best-fit users.
Related:
- Retail Failure — why human retail users churn
- Resolution Problem — oracle failures and microstructure
- Anti-Induction — the alignment problem as markets scale
- Chrono Score — the reputation layer for agents and humans