A Committee of Agents for Prediction-Market Trading: Design of a Paper-Only Kalshi System

micro = microstructure_analyst(m, side, ask) if micro^[0] < 0 and micro^[1] >= 0.7: return {"decision": "no_trade", "vetoed": True, ...}

Approved cases are not trades. They form a docket, and a human reviewer (in our case, an AI assistant operating in a supervised session) examines every case, writes an independent probability estimate with a stated rationale, and either confirms or rejects. Rejections are logged with reasons, so the audit trail records everything the reviewer saw, not just what was liked.

## Sizing as an experiment, not an argument

Rather than arguing about position sizing, the system runs six paper books simultaneously on identical picks: a conservative rails book (quarter-Kelly with 5% per-trade, 10% per-event, 20% per-theme concentration caps), full Kelly, 1.5x Kelly, 2x Kelly, an all-in book, and an adaptive book that stays fully deployed with allocations weighted by committee confidence under a 60% single-trade ceiling. Kelly theory predicts the ordering of outcomes - growth peaks at 1x and turns negative beyond 2x - and the parallel books exist to measure that prediction against reality rather than assume it.

## Results

None yet, and that is the honest report. The system has passed its software test suite (ten offline test groups covering fee math, committee blocks, sizing rails, and ledger scoring), but no picks have been made and no validation gate has been reached. The pre-registered gate requires fifty settled picks per book, profitability after real fees, and reviewer calibration that beats the market's own prices before any capital discussion is allowed. A finding of "no edge survives fees" is a publishable outcome of this research, not a failure of it.

## Limitations

The bias prior is taken from betting-market literature, not measured on this exchange; the gate exists precisely because its local size is unknown. Fill assumptions are conservative but unverified. Fees are modeled from the published schedule. Capacity in these markets is small, and nothing here scales to institutional size. Above all: an unvalidated system described clearly is still an unvalidated system.

*Full implementation available on request.*

## Key definitions

**Favorite-longshot bias** - A documented tendency in prediction markets where heavily favored outcomes are priced slightly lower than their empirical frequency, creating a small but systematic mispricing.

**Cross-venue disagreement** - Variation in the price of the same event contract across different exchanges, potentially offering arbitrage or relative-value opportunities.

**Kelly criterion** - A position-sizing formula that maximizes expected logarithmic wealth growth by betting a fraction of bankroll proportional to edge and odds; the optimal fraction is called "full Kelly."

**Hard block** - A veto condition in the committee system (phantom price, dead book, or adverse price movement on volume) that automatically rejects a candidate regardless of other analysts' votes.

**Paper trading** - Simulated execution of trades using historical or real-time data without committing actual capital, used to test strategy logic and risk controls before live deployment.

**Microstructure** - The mechanics of order books, spreads, volume, and fill patterns; microstructure analysis assesses whether a quoted price is real enough to trade.

## References

- Kalshi, "Markets & Rules," Kalshi (2024). Https://kalshi.com/markets
- Thorp, E. O., "The Kelly Criterion in Blackjack, Sports Betting, and the Stock Market," *The Journal of Portfolio Management*, 25(5), 100-112 (1997).
- Snowberg, E., Wolfers, J., & Zitzewitz, E., "Prediction Markets for Economic Forecasting," in *Handbook of Economic Forecasting*, Vol. 2A (2013). Https://doi.org/10.1016/B978-0-444-62731-5.00018-9
- Malkiel, B. M., "A Random Walk Down Wall Street," W.W. Norton & Company (revised 2019).
- CFTC, "Prediction Markets," U.S. Commodity Futures Trading Commission (2024). Https://www.cftc.gov
- Investopedia, "Kelly Criterion," Investopedia (2024). Https://www.investopedia.com/terms/k/kellycriterion.asp

---
*Educational research on historical data only - not investment advice, not a signal, and never a performance promise. Past results do not predict future performance. Drafting uses AI assistance; every citation is link-verified before publication and every paper is re-audited weekly against the library's editorial standard.*
*Last reviewed by the PropLedger research pipeline: 2026-08-26. Educational research on historical data; not financial advice.*