Match probabilities.
A price worth taking.
A likely winner can still be an overpriced bet. Our CS2 model compares its estimated win probability with the price available to buy, including trading fees and available liquidity. Favorites and underdogs face the same value checks.
How the probability is formed
Pinnacle prices provide a market baseline after removing the bookmaker margin. A team rating built from completed series supplies a second estimate. The model fits their combination on verified historical prices and results, with limited departures from the market when the evidence is weak.
Team identity and series format matter. Academy, junior, and women's rosters keep separate identities. Rankings and recent match coverage determine eligibility and data quality. Map vetoes and player statistics are not fitted inputs in this version.
When a match becomes a pick
| Check | Current rule |
|---|---|
| Forecast window | New forecasts are generated 1 to 6 hours before match start. |
| Team coverage | Both teams must be ranked within the top 100 and have at least 8 recent matches. |
| Model edge | At least 8 percentage points above the market probability after removing the margin. |
| Expected return | At least 8% at the reference odds, and at least 8% after estimated Polymarket fees at executable asks. |
| Reference odds | The current exposure limits are -150 through +150. |
| Stake and volume | One model unit per pick, up to 3 picks per match day. No automatic stake increases. |
| Approval | An operator must approve publication. Approval refreshes the probability, price, fees, and available size. |
Read the maximum buy price
Each approved Polymarket pick shows a maximum buy price. This is the highest quoted ask that still meets the model's minimum expected return after estimated fees. A price above it no longer qualifies, even if the team is still likely to win.
Available size matters too: a small amount at an attractive ask may not cover your whole stake. Where automatic execution is enabled, it checks the order book at your actual stake and submits a protective limit. Prices and depth can change before a fill.
How results are evaluated
The repaired model starts a separate forward record. We compare its probabilities with subsequent outcomes and the market baseline, then measure returns at recorded executable quotes after estimated fees. Closing-line comparisons use verified quotes captured strictly before the match starts; missing quotes remain missing.
Historical records contain affected price mappings and cannot establish this version's profitability. Quote-based returns also differ from actual account profit, which depends on confirmed fills and paid fees.
A higher win rate alone does not establish an advantage. Profitability requires returns that cover the purchase price and costs, supported by enough new results to distinguish an advantage from variance. The repaired model has not yet established profitable forward performance.