The Bad Beat Trap
A bet at plus 110 with a true seventy percent win probability is plus forty seven percent expected value, which is GEM-tier edge, and yet thirty percent of the time it loses regardless of how correct the decision was. The probability of three consecutive losses on three independent plus EV bets at this edge is one in thirty seven, which means a disciplined bettor will hit a three-loss streak roughly thirteen times across a five hundred bet season, and confirming a real four percent edge at ninety five percent confidence requires about twenty five hundred bets while the bettor's intuition evaluates strategy on five to twenty. The discipline is to grade the decision, not the outcome.
Episode 15 of the WagerBird Methodology series. Watch on YouTube →
What Is Resulting
Resulting is Annie Duke's term (Thinking in Bets, 2018) for grading the quality of a decision by the outcome it produced. A +EV bet that lost feels like a bad decision. A −EV bet that won feels like a great decision. The decisions are identical in quality. The outcomes are variance. Resulting conflates signal (decision quality) with noise (single-bet outcome).
The Worked Example
A bet at +110 American odds (decimal 2.10, implied 47.6%). Model says true probability is 70%. EV per $100 staked = 0.70 × $110 − 0.30 × $100 = $77 − $30 = +$47 = +47% EV. GEM-tier edge. The decision is clearly correct. The variance: 30% of the time, the bet loses regardless of how correct the decision was. Probability of 3 consecutive losses on 3 +EV bets at this edge = 0.30³ = 2.7% = 1-in-37. Across a 500-bet season, expected number of 3-loss streaks ≈ 13. The streak is built into the math.
The Signal Detection Math
To confirm a 4% edge against the null hypothesis of zero edge at 95% confidence: approximately 2,500 bets. The bettor evaluates strategy intuitively on 5-20 bets. The sample-size gap is roughly 125×. The bettor is reading noise as signal in every direction. Bad beats feel like a broken system. Lucky wins feel like skill. Both are the same variance running in different directions.
The Asymmetry
The bettor and the book have the same information. Same prices. Same model. Same outcomes. The bettor grades the decision by the outcome (resulting). The book grades the decision by the EV (process). The book is indifferent to which direction the bettor flips: tilt after losses sends the bettor into chase mode at higher stakes on lower confidence; confidence after wins sends the bettor into size-up mode at higher stakes on the same lower confidence. Either flip puts the book on the right side of the next bet.
How WagerBird Prices It
WagerBird picks are graded by the model on decision quality, not by results. Confidence-scored from 25 to 100. The signal sets the stake. The result of any single pick is one realization from a probability distribution. A 96-confidence pick that lost is still a 96-confidence decision.
The Transcript
Machine transcript of the narration, lightly cleaned. It reads as spoken word rather than authored prose.
Result lies, process tells, the bet that won, the bet that lost. Same decision. Two grades. This is the bad beat trap.
Here is what the better says after three losses on three plus EV bets in a row. My system is broken, the model is off. Time to chase, time to switch strategies. Here is what three losses on plus EV bets actually is.
Variance. The probability is 2.7% one in 37 built into the math will happen roughly 13 times across a 500 bet season. The streak is not a malfunction, the streak is the dice. Run the math on a representative ticket.
A plus EV bet at +110. American odds +110 implies 47.6%. The model says the true win probability is 70%. Expected value per 100 staked equals 0.7 times $110 equals $77 less 0.3 times $100 equals $30.
Net plus $47 per 100 staked. Plus 47% expected value. Gem tier confidence. The decision is clearly correct.
Now the variance. The same bet loses 30% of the time regardless of how correct the decision was. Three losses in a row on three independent plus EV bets at this edge has probability 0.3 cubed equals 0.027 equals 2.7%. One in 37 sequences of three.
Across a 500 bet season at this edge. The expected number of three loss streaks is 13.5. The streak is the math, not the malfunction. Now the conflation.
The better reads the outcome, the better grades the decision by the outcome. A plus EV bet that won is a good bet. A plus EV bet that lost is a bad bet. Same decision, same +47 expected value.
Two grades. This is resulting. Annie Duke's term for the formal error of grading process by result. The signal versus noise math.
To confirm a 4% edge against the null hypothesis of zero edge at 95% confidence requires approximately 2,500 bets. The better evaluates strategy intuitively on five to 20 bets. The sample size gap is 125 times. The bettor is reading noise as signal in every direction.
Bad beats look like a broken system. Lucky wins look like skill. Both are the same variance running in different directions. This is the asymmetry the bad beat trap depends on.
The better and the book have the same information. Same prices, same models, same outcomes. The better grades the decision by the outcome. The book grades the decision by the expected value.
The better abandons plus EV process after a bad beat streak. The book continues to price the next bet at the same model probability. The book is indifferent to which direction the better flips. Tilt after losses sends the better into chase mode at higher stakes on lower confidence bets.
Confidence after wins sends the better into size up mode at higher stakes on the same lower confidence bets. Either flip moves the better off the optimal sizing curve. Either flip puts the book on the right side of the next bet. This is where WagerBird lives.
Each WagerBird pick is graded by the model on decision quality, not by the result. The confidence score from 25 to 100 is the decision quality input. The result of any single pick is one realization from a probability distribution. We do not flinch when a 96 confidence pick loses.
We do not size up when a 50 confidence pick wins. The signal sets the stake. The result does not change the signal. This is the pattern.
The better grades the decision by the outcome and abandons plus EV process after a bad beat streak that was variance not malfunction. The book takes the post tilt bets at unfavorable prices. The trade is to grade the decision by the expected value of the call at the time the bet was placed. The rule is one sentence.
Wins fool, losses fool. Grade the decision. Every WagerBird pick is published with the confidence score alongside the pick. The Hotsheet delivers our 76 to 95 rated picks.
The Terminal carries the full pregame board including every gem. We grade the call not the cover. See it, size it, send it. What does that mean?
See it. The pick lives in the Terminal on the full 25 to 100 confidence scale. Size it. Higher confidence, larger position.
Send it. Take the pick out of WagerBird. Your bet is placed at your book wherever you place your action. WagerBird is the analytics platform that prices the market itself.
No handicappers, no personalities, real traders, real models, real edge. As we say here at WagerBird, a see it, size it, send it.
// MORE BOOKMAKER TRAPS
