🎯 What This Hub Actually Does
Most "betting picks" sites tell you who to bet on. This hub tells you WHY, shows the math, and tracks results honestly — including losses. The goal isn't to win every bet (impossible). The goal is to find positive expected value opportunities and size them correctly.
Core principle: Even high-confidence bets lose ~40% of the time. The skill is finding situations where the sportsbook line undervalues a side.
📊 Expected Value (EV) — The Core Math
Every bet has an expected value — the average profit/loss if you placed the same bet many times.
EV = (Win Probability × Profit) − (Loss Probability × Stake)
Example: Pistons -110 spread bet at $10 stake. Your model says 55% chance to cover.
EV = (0.55 × $9.09) − (0.45 × $10) = $5.00 − $4.50 = +$0.50 expected profit per bet
Long-term, this bet wins. Doesn't matter if THIS bet loses — over 100 bets like this, you profit.
The rule: Only bet when EV > 0. If your win probability is below the "break-even" implied by the odds, don't bet.
Break-even probability table
| American Odds | Break-even % needed | What it means |
|---|
| -110 | 52.4% | Standard spread/total |
| -150 | 60.0% | Modest favorite |
| -200 | 66.7% | Heavy favorite |
| +100 | 50.0% | Pick'em |
| +150 | 40.0% | Modest underdog |
| +200 | 33.3% | Underdog |
💰 Kelly Criterion — How Much to Bet
Developed by John L. Kelly Jr. at Bell Labs in 1956. Tells you the optimal fraction of your bankroll to wager.
f* = (b × p − q) / b
b = decimal odds − 1 · p = your win probability · q = 1 − p
Example: $1,000 bankroll. Pistons -110 spread (decimal 1.91, b=0.91). You estimate 55% win probability.
f* = (0.91 × 0.55 − 0.45) / 0.91 = (0.5005 − 0.45) / 0.91 = 0.0555
Full Kelly: Bet $55.55 (5.5% of bankroll)
Quarter Kelly (recommended): Bet $13.88 (1.4% of bankroll)
Why fractional Kelly: Wharton research (2023) showed full Kelly leads to bankruptcy in 100% of simulations. Half Kelly: also ruin. Quarter Kelly: profitable. Eighth Kelly: even better growth rate over time.
Default for this hub: $10 flat stake (~1% of $1,000 reference bankroll). This is roughly 1/4 Kelly for typical Rank-1 picks.
📈 Closing Line Value (CLV) — The True Skill Metric
Win/loss tells you nothing about skill in small samples. CLV tells you everything.
The closing line is the most accurate odds. By game time, sportsbooks have processed all news, sharp action, and information. The "true probability" of an outcome closely matches the closing line.
CLV % = (Closing Odds / Your Bet Odds) − 1
Example: You bet Pistons -3.5 at -110. By game time, the line moves to -2.5 at -115.
You got a BETTER line than closing. You "beat the closing line by 1 point."
If you consistently do this, you'll profit even when individual bets lose.
Pinnacle's research (R² = 0.997): Closing odds at sharp sportsbooks correlate with true probabilities at 99.7%. This is why sharp bettors track CLV obsessively — it predicts long-term profit better than win/loss record.
🔍 Vig Removal — Finding True Probability
Sportsbooks bake in a margin (the "vig" or "juice") to ensure profit. To find true probability, you must remove it.
True Prob = Market Implied Prob / (1 + vig)
Example: Pistons -110 (52.4% implied) vs Cavs -110 (52.4% implied)
Total = 104.8%. The 4.8% is the vig.
True probabilities: Pistons 50.0%, Cavs 50.0% (the market thinks it's a coin flip)
If your model says Pistons 55%, you have a 5% edge over the market.
🎲 Boris Chen Tier Method — Applied to Bets
Boris Chen revolutionized fantasy football rankings by clustering experts into tiers. Within a tier, picks are statistically interchangeable.
Same principle here: A "62% confidence" bet and a "65% confidence" bet are the same tier. The model can't reliably distinguish between them. Focus on tiers, not point estimates.
Our tiers:
🟢 Tier 1 (60%+): Take it — clear edge, ~5%+ above break-even
🟡 Tier 2 (55-59%): Selective — only if odds compensate (look for +110 or better)
🟠 Tier 3 (52-54%): Pass usually — barely above break-even, no margin for error
🔴 Tier 4 (<52%): Never bet
❓ "Should I Fade Myself?"
This is a great question and the answer is no — at least not yet.
Why fading doesn't work:
- Casual bettors hit ~50-52% on spreads (need 52.4% to break even). Inverting noise doesn't help.
- Sample size matters: you need 30-50 bets minimum to know if you're losing systematically.
- The bet WHO loses is uncorrelated to your model's edge. You're not "wrong" in a pattern — you're losing one bet.
Better question: Am I beating the closing line? If yes, keep betting. If no, the market is faster than you. Stop or adjust.
📚 What Real Sharps Do
1. Line Shop
Have 4+ sportsbooks. The half-point difference between DraftKings -3.5 and FanDuel -3 means $$$$ over time.
2. Track CLV religiously
Pros pay more attention to line moves than game results. If you bet at -3 and it closes at -4, you got value regardless of outcome.
3. Use Quarter Kelly
Or flat 1% of bankroll. Full Kelly is for casino owners and lunatics.
4. Avoid parlays for value
Parlays have built-in extra vig. Use only for entertainment moonshots with small stakes.
5. Avoid recency bias
"He's been hot lately" is already priced in. Looking back 5 games is amateur. Looking at season-long indicators (xG, eFG%, BABIP) finds value.
6. Specialize
Pros bet 1-2 sports max. Knowing one sport deeply > knowing all sports surface-level.
🗓 How Future Games Are Covered
Game analyses are built on current odds for the actual upcoming game. We do not:
- Pre-write conditional scenarios ("if Team A wins tonight, here's tomorrow's pick").
- Build "Scenario A / Scenario B" branches for games not yet scheduled with confirmed lines.
- Forecast bets for matchups before the market has set a number.
We do:
- Build full analysis once odds are available — typically 24–48 hours before tip.
- Update analysis as new information emerges (injury news, line moves, lineup changes).
- Add a Result section after the game with bet outcome (win / loss / push) and any lessons learned.
- Track running performance on /bets/performance.html.
Pre-writing conditional branches looks like covering all bases — but it dilutes the actual call. A pick has to commit
to one number against one line for one game. Anything else is hedging in advance, which is just slow indecision.
📖 Sources & Further Reading
- Kelly, J.L. (1956). "A New Interpretation of Information Rate" — Bell Labs original paper
- Wharton Beggy (2023). "Investigation of Sports Betting Selection and Sizing"
- OddsJam — Closing Line Value & vig removal calculators
- Pinnacle Sports — Sharp-side odds reference
- Boris Chen — Gaussian Mixture Model tier clustering (borischen.co)
- Action Network — CLV tracking education
📐 Strategy v2 — sport-specific framework
Backtest reality: after 40 games tested (20 NBA + 20 MLB), my analysis has not demonstrated edge over blind favorite-betting.
v2 strategy at /bets/strategy/ documents the sport-specific framework + the confidence-calibration finding (STRONG bets actually underperformed MEDIUM bets in the MLB test).
🎰 SameSHOT Framework
"SameSHOT" = a low-cost bet where the payoff is large relative to the stake. They fail most of the time. The framework below identifies when factors stack so that, across many bets, the math works out.
Individual SameSHOTs fail. The framework is about the long-run aggregate.
The stack-factors (7 in golf, similar lists in other sports)
- Course/matchup fits the underdog's specific game — narrow course favors accuracy specialist over bomber; physical game favors specific archetypes
- Past data hides indicators — new venue, returning from injury, recent role change
- Recent form trumps reputation — books slow to update odds for hot streaks
- Pressure asymmetry — favorites carry expectations, underdogs have nothing to lose
- Conditions windows — weather, tee times, scheduling quirks favor specific players/teams
- Historical matchup type — known weaknesses for specific course/opponent types
- Randomness boost from novelty — first-time-at-venue, expansion team, unfamiliar opponent — local knowledge advantage disappears
Rule of thumb: when 3+ factors stack for a 50:1+ underdog, that's a real value SameSHOT. When only 1 stacks, it's just gambling.
How to stack SameSHOTs in a portfolio
- Diversify by outcome type — outright + top-N + make/miss-cut + parlay. Don't load all eggs on "X wins outright."
- Cap per-ticket spend — $1-5 each. Five $5 tickets > one $25 outright if you want the long-tail upside.
- Mix safe + speculative — pair a favorite outright (high hit rate, modest payout) with 2-3 long shots (low hit rate, huge payout if any single one hits)
- Aim for combined parlay odds 15:1 to 50:1 — below 15:1 too little payoff, above 50:1 too many things must go right
- Live betting after R1/G1 — odds reflect reality, books often slow to fully reprice
Worked example
See the PGA Championship analysis — Section 5 lays out 5 SameSHOTs with stack-factor counts (3-4 favorable factors each) plus a $25 portfolio in Section 8 that diversifies across outcome types.
Honest reality: Even a 4-factor stack on a 50:1 player still implies the bet fails ~95% of the time. The framework just tilts the math from "blind hope" to "calculated long-shot." Bet sizes should reflect that — $1-5 per ticket, never the rent money.
This methodology page is for education. Bet responsibly. Even +EV bets lose 40-45% of the time.