🎯 The Strategy Framework

The Question This Answers

You've learned the math: EV, Kelly, Gaussian, CLV. Now what? How do you ACTUALLY decide which bets to make, when, and how much?

This page is the integration. Step-by-step decision process, paper trading to test it, projected growth, and an honest backtest framework.

📋 The 7-Step Decision Process

Every potential bet goes through these 7 gates. If it fails any gate, skip the bet.

🔀 The Decision Flowchart

📈 90-Day Expected Growth (Projection)

Assumptions

Expected bankroll trajectory

📝 Paper Trading Tracker

The fastest way to test the strategy WITHOUT risking real money: paper trade. Record every "would have bet" decision, track outcome, see if the math holds.

We're already doing this with the bankroll tracker — $10 per Rank-1 pick. Let's formalize it.

Rules of the paper trade

1. Apply the 7-step process to every game on our analysis pages
2. Only "bet" picks that pass all 7 gates
3. Track in /bets/bankroll.html with full transparency
4. After 30 picks, evaluate:
   • What's our actual win rate?
   • What's our actual avg edge?
   • Are we beating the closing line?
   • If +CLV but losing → variance, keep going
   • If −CLV and losing → strategy is wrong, recalibrate

Current paper trade status

🔬 Backtest Framework

True backtesting requires historical odds data (what was offered, when) + outcomes. We don't have that database yet.

What we'd need for real backtest

• Historical odds API ($50–200/mo from OddsJam, ActionLabs, etc.)
• Outcome data (free from ESPN-style APIs)
• Our model run against each historical game
• Compare predicted edge vs actual outcomes

Synthetic backtest example (clearly fake)

Here's what a backtest output WOULD look like with 200 historical bets:

Hypothetical summary: 200 bets, 53.5% win rate, $238 profit (+23.8%), max drawdown −8.2%. These numbers are illustrative — not real backtest data.

🧠 Combining the Tools

EV + Kelly (most common combo)

EV tells you IF to bet. Kelly tells you HOW MUCH. Always use both.

Gaussian + EV (for spreads and totals)

Use Gaussian to estimate your win probability for continuous outcomes, then feed into EV.

Beta + Updates (Bayesian)

As bets settle, update your confidence about your model:

CLV (the ultimate scoring)

After 30 bets, plot your CLV distribution. If consistently positive, you're sharp. If negative, you're slow to information.

📅 Timeline of Expected Outcomes

⚖️ Why this framework — comparison to the alternatives

Most bettors lose. Why? Most use a different framework (sometimes implicitly). Here's how ours compares — measured by what really matters: long-run ROI, not win rate.

Win rate is misleading. A 65% win rate at -200 odds = 65 × $50 − 35 × $100 = -$250 per 100 bets. Win rate ≠ profit. See EV lesson for the math.

🚫 What we DON'T do (and why)

• ❌ Parlays — vig compounds. A 2-leg parlay at -110/-110 pays +264 but fair value is +300. You're paying ~9% vig per leg. 5-leg parlays are advertised at +2000 with fair value of +3100.
• ❌ Same-game parlays — even worse. Books deliberately misprice correlations because they know correlated outcomes are exactly what makes parlays look attractive to recreational bettors.
• ❌ Live betting (mostly) — books update lines in milliseconds based on game state. By the time we see a price, the edge is usually gone. Exception: pre-game prices that haven't moved in slow markets.
• ❌ Player props on stars — these are the most-shopped market in sports betting. If we're betting Cunningham over 24.5, we're competing against sharps who built a model just for that line.
• ❌ Massive favorites (-300+) — bad risk/reward. To win $100 you risk $300. One upset (which happens ~25% of the time at -300) wipes out 3 wins.
• ❌ Chase losses (Martingale) — doubling after losses is mathematically guaranteed to bankrupt you eventually. The risk of ruin is 100% over infinite time.
• ❌ "Lock of the day" — there are no locks. Anyone who tells you they have one is either lying or has confused conviction with edge.

⏸ When to skip the framework

The framework's job is to gate bets, not to manufacture them. Sometimes the right move is to skip — even when a bet looks attractive.

• ⏸ Not enough information. Key injury news pending, lineup TBD, weather forecast unstable. Walk away — the line will be there tomorrow.
• ⏸ Markets moving rapidly. Line just moved 2+ points in your direction = sharp money already there, edge is gone. Line moved against you = market knows something you don't.
• ⏸ Public-heavy game. When 80%+ of bets are on one side but the line doesn't move, sharps are taking the other side. Default to the unpopular side or skip.
• ⏸ Emotional bet. Favorite team, revenge bet, recovery-from-loss bet. The framework can't see the bias — but you can. Skip.
• ⏸ Edge under 2%. Under the threshold, the variance dominates. You can't tell if you're winning because of skill or noise. Skip until edge is meaningfully positive.

🔬 What the backtest taught us

We ran the simplified model against 4,615 historical NBA games. The result:

• −95% ROI after 4,615 bets. The bankroll was effectively gone.
• Win rate of ~50.7% — looks ALMOST break-even. Vig is what killed it.
• Average CLV: −1.3% — the closing line consistently moved against our bets. This was the signal that the model wasn't beating the market.
• CLV detected the failure inside ~500 bets — long before raw P&L would have. That's the case for tracking CLV from bet 1.

The lesson: a small edge plus bad sizing equals bankroll death. The framework exists specifically to prevent this — gate the bets, size them small, track CLV from the start. The backtest page has the full breakdown.

🎯 Player Props — Where Most Retail Money (and Losses) Live

Player props are now the most common bet type by volume. They're also where:

• • Books make their highest margins (4–5% vig vs 2–3% for game lines)
• • The public loses the most over time (public-heavy overs of stars)
• • Sharps find the most edge — if they're disciplined about it

Our framework for props

1. Same 7-gate process — props get no special pass.
2. Higher edge threshold: ≥ 4% (vs 2% for game lines) because the vig is higher.
3. Smaller stakes: half Quarter-Kelly — single-player outcomes are noisier than team outcomes.
4. Track prop CLV separately — different market dynamics; don't pollute the game-line CLV signal.
5. Never SGPs / prop parlays — vig magnification is structurally negative.
6. 1–2 best props per game, not 10 — picking 10 is correlation-betting dressed as analysis.

Red flags that mean skip a prop

• ⏸ It's a "lock" — if everyone's talking about it, the line will move before you place.
• ⏸ Season averages only — without recent-trend data, you're betting blind.
• ⏸ Role just changed — trade, new starter, scheme tweak. Skip until you have 5+ games of new role.
• ⏸ Injury status uncertain — game-time decision. The market knows; you don't.
• ⏸ Obscure / illiquid prop — low limits, books close on winners fast.

Full eight-input prop framework: NBA prop analysis template →.

📚 What Comes Next

1. Track CLV starting NOW. Every bet recorded with closing line. Without this you can't know if you're improving.
2. Hit 30 paper-traded bets before making any strategic changes.
3. If CLV is positive at 30 bets — you have signal. Continue, consider real money (small).
4. If CLV is negative at 30 bets — recalibrate the model. We're slow to information.
5. Decision point at 90 bets: scale up or shut down.