A public paper fund vs. the prediction markets
When our calibrated v2 win-probability model diverges from a prediction market's price by at least 8%, this fund takes a simulated position on the model's side — sized by quarter-Kelly, capped at 10% of bankroll, with every venue fee modeled. Positions are logged the moment they open, before the game resolves.
Equity curve — from $50
No positions have settled yet — the curve begins at the $50.00 starting bankroll and will extend as games resolve.
Open positions — logged before they resolve
No open positions right now.
Settled positions
Nothing settled yet.
Market coverage — today's 5-game slate
| Venue | Markets found | Liquid enough |
|---|---|---|
| Kalshi | 5/5 (100%) | 5/5 (100%) |
| Polymarket | 5/5 (100%) | 5/5 (100%) |
"Liquid enough" applies a modest bar (Kalshi: any resting open interest; Polymarket: ≥ $5k resting liquidity on the moneyline). Both venues list per-game MLB moneyline markets; depth concentrates in the hours around first pitch.
Backtest — the strategy over our graded log (fees included)
Backtest universe is our forward-graded pregame_close log (no free historical Kalshi/Polymarket price archive exists), priced off the de-vigged odds-API line widened to a conservative taker ask, charged Kalshi's real fee schedule. Small samples are not evidence of an edge either way.
| Bar | Positions | Hit rate | ROI (post-fee) | Max DD |
|---|---|---|---|---|
| 5% | 24 | 46% | 24.1% | 13.8% |
| 8% | 9 | 33% | 18.2% | 11.1% |
| 10% | 6 | 33% | 21.5% | 8.0% |
| 15% | 4 | 50% | 32.1% | 0.0% |
Methodology & honesty
Entry rule. Buy the side where our calibrated v2 model prices the outcome above the market by at least 8% — the same divergence bar the model's "notable calls" use. No discretion; the rule is a constant in the code.
Sizing. Fractional Kelly. For a binary contract bought at price q with model prob p, full Kelly is (p − q) / (1 − q); we take 25% of that, capped at 10% of bankroll per position, and floor to whole contracts.
Fees, modeled explicitly. Kalshi taker fee ceil(0.07·C·P·(1−P)) per fill (settlement free); Polymarket sports taker fee shares·0.05·P·(1−P). A strategy that's positive pre-fee and negative post-fee is the most common self-deception here, so both numbers are always shown.
Conservative fills. We fill at the price available at decision time — the venue ask, never a better mid — and never assume size a thin book couldn't absorb. In the backtest, where only a de-vigged closing line exists, we widen it toward a taker ask before charging fees.
Settlement. Positions settle on the same game outcomes we grade on /model. Bankroll starts at $50.00.
Phase 1 is paper only. No trading credentials, no order placement. The architecture is built so a future execution module could subscribe to the same signals — but nothing here places an order.