Yesterday’s results
Every qualifying pick. Original lines. Final game stats.
By default, results show all saved picks with a projected edge of 4% or more. A higher edge means our model sees more value at the listed odds. Where we show a suggested stake, half Kelly uses the edge, odds and bankroll to size it. At the same odds and bankroll, a higher edge suggests a larger stake.
Overall since November 5, 2026
Every qualifying pick from our current model. Your book, category, type, player and edge filters apply to this total.
Selected day
0 offered picks at 4%+ saved edge. Hit rate excludes ties and pending results. Alternate lines and picks across books can share the same outcome.
Every edge: stakes based on a $1,000 bankroll
Hypothetical RockBet single-bet results for your current filters. PrizePicks entries are excluded.
Select RockBet or Both books to see this example. PrizePicks requires entry composition and payout details.
How Kelly and half-Kelly betting work
1. What Kelly measures
Kelly chooses a bankroll fraction to maximize expected long-run logarithmic growth under the assumed outcome probabilities and payout. It is a sizing rule, not a prediction model, and it does not guarantee a profit on a day or season.
2. The inputs
The inputs are your estimated chance of winning and the offered odds. A projection must first become a probability: this report uses a Poisson model with the saved full-precision pregame projection for RockBet’s count markets. A saved edge percentage is not the win probability. Whole-number lines have separate win, loss, and refund probabilities.
3. Convert the odds
Net odds b are profit per $1 staked: +150 gives b = 1.50; −150 gives b = 100 ÷ 150 ≈ 0.6667. Break-even probability for a bet with no push is 1 ÷ (b + 1).
4. Calculate the bankroll fraction
With win probability p and loss probability q, full Kelly is max(0, (b × p − q) ÷ (b × (p + q))). For bets without a push, q = 1 − p and p + q = 1. Half Kelly uses half of the full-Kelly fraction. A nonpositive estimated advantage means a $0 target stake.
5. A worked example
At +100 with a 55% estimated chance to win, b = 1 and q = 45%. Full Kelly is (1 × 0.55 − 0.45) ÷ 1 = 10%. Half Kelly is 5%, or $50 from $1,000. A win produces $50 net profit; a loss costs $50; a push refunds the stake. This is an illustration, not a pick from this slate.
6. Why use half Kelly?
Half Kelly reduces the full-Kelly stake and typically reduces volatility and drawdowns, at the cost of lower modeled growth. Overstated win probabilities can still lead to oversized bets. Several props on one player or game can lose together; independent single-bet Kelly is not a joint portfolio optimization.
7. How this example handles a whole slate
Every stake is half Kelly multiplied by the same hypothetical $1,000 bankroll, rounded to cents. No proportional scaling or slate-level funding cap is applied. We show what the results would be if every qualifying RockBet pick were taken at its suggested stake, even when the total requires more than $1,000. This is an uncapped results illustration, not a fundable bankroll path. It does not model settlement times, sequential bankroll changes, limits, fees, taxes, or injury-protection rules. Net profit uses the original offered odds and the rounded stake; returned principal is not counted as profit.
Further reading: Kelly’s original paper (1956) and Kelly criterion formulas.
Results by category
No picks match these filters.
Every pick, with the result
| Player | Book / type | Category | Direction | Offered line | Book odds | Actual | Result | Saved edge | $1k example stake | Net P/L |
|---|---|---|---|---|---|---|---|---|---|---|
| No matching picks. | ||||||||||
How results are graded
Only picks with a saved edge of at least 4% are included. Outcomes compare the original offered line and direction with final NHL game statistics. Player IDs distinguish players sharing a name. A tie means the actual equals the line. Missing or ambiguous results stay pending. These are statistical grades; sportsbook injury protection, voids, entry payouts, and settlement adjustments are not applied. Time on ice is shown in decimal minutes. Saved edge definitions differ between books. Model-snapshot dates are labeled separately from confirmed pregame publications.