1. Strategy v1 — "Line Agreement + Talent Gap"
After backtesting 20 NBA playoff games, the data showed:
- My analysis returned +39% ROI but underperformed blind favorite-betting (+43.5%)
- The 4 losses all came from overriding the line with narrative factors
- The wins came from agreeing with the line on clear favorites
Core rules of v1
- Only bet when analysis AGREES with the line direction (no contrarian picks)
- Heavier weight on talent / seed / record gaps (these predict)
- Lower weight on narrative factors (must-win, pivotal home, etc.)
- SKIP coin-flip games (50-55% probability range)
- Stake by confidence: STRONG $8 · MEDIUM $5 · SKIP $0
- Track outcomes religiously to validate / invalidate
What actually happened in MLB
The v1 strategy applied to 20 MLB games (May 12-14): lost money. -11.7% ROI. 7-6 record.
Worse: STRONG-confidence bets went 4-4 (-19.4% ROI). MEDIUM bets went 3-2 (+8% ROI). When I was "most certain," I was wrong more often.
2. Why a "strategy" alone isn't enough
A betting strategy without sport-specific data is incomplete.
What works in NBA (favorites winning ~65-70%) does NOT work in MLB (favorites winning ~55-58%) because the vig kills you at lower hit rates. A -180 MLB favorite needs to win 64% of the time just to break even — that's an unrealistic bar for daily baseball.
Each sport needs:
- Sport-specific factors that actually predict
- Data sources beyond season records (especially MLB starting pitchers)
- Calibrated confidence levels validated by results
The strategy framework needs sport-specific models, not one universal approach.
3. Strategy v2 framework
Universal rules (apply to all sports)
- Track every prediction with explicit reasoning (in /bets/prediction-log/)
- Score outcomes honestly — not just W/L but why
- Don't trust any "edge" until 100+ samples
- Lower stake sizes until proven track record
- Calibrate confidence by actual results, not feel
🏀 NBA Playoffs
$5-8 per bet · target: 65%+ hit rate
What works:
- Favorites with talent / seed gaps win consistently
- Heavy home favorites in elimination games
- Road favorites in dominant series (talent travels)
- Series-closing G7 home court
Avoid: pivotal-G5 narratives (priced in), "must-win" home underdogs, overriding talent gap with situational stories.
⚾ MLB Regular Season
$3-5 per bet · target: 56%+ hit rate
What works (with caveats):
- MUST include daily starting pitcher matchup — season records alone aren't enough
- Bullpen quality for late-game scenarios
- Travel / rest patterns affect performance
- Bounce-back at home after upset loss (small sample, validate)
Avoid: heavy MLB favorites (-150+) without clear pitcher advantage. Daily variance and shutouts are real. Stakes smaller until pitching-data pipeline is built.
⛳ Golf Majors
$2-5 per long-shot · payout 50:1+ matters more than hit rate
What works:
- Course fit > player ranking for specific events
- Recent form matters more than career stats
- Use the SameSHOT Framework for value (3+ factor stack)
- Diversified portfolios across outcome types (outright + top-10 + cut)
Avoid: single big stake on one outright. Diversify across 5+ small tickets.
🏒 NHL / ⚽ Soccer / Other
$0 — don't bet yet · build the framework first
Don't bet what we don't know yet.
Build sport-specific factor list + backtest 20+ games BEFORE risking money. Avoid the "but this game looks good" trap.
4. Confidence calibration — the MLB lesson
From the MLB backtest:
| Confidence | Record | ROI |
|---|
| STRONG | 4-4 | -19.4% |
| MEDIUM | 3-2 | +8.0% |
When I think I'm most certain, I'm probably wrong.
Possible reasons:
- Confirmation bias — when "sure", I ignore counter-evidence
- Heavy favorites have terrible payouts (risk $8 to win $4)
- The market knows what I know — "obvious" picks have no edge
New rule — inverse stake sizing (or flat)
- STRONG conviction: bet SMALL ($3)
- MEDIUM conviction: bet NORMAL ($5)
- MAYBE conviction: bet NORMAL or skip
Or, honestly: $5 flat across all bets until calibration is proven over 30+ games.
5. How we actually improve
The real strategy is the tracking system
- Make predictions with explicit reasoning
- Wait for game outcome
- Score honestly — was the pick right? Were the factors right?
- Track which factors actually predict over many games
- After 100 games: update strategy based on what actually worked
This is the difference between gambling and forecasting. Most casual bettors do step 1. Professionals do all 5. We aim for all 5.
📌 Sam's weekly bet budget: $25-50/week for analysis testing. Treat as learning expense, not income source. After 100+ games tracked, reassess.
Pending strategy v3 triggers
Update strategy v3 when:
- 100+ games tracked in the prediction log
- Specific bet type shows ≥55% hit rate over 30+ samples
- Specific factor shows ≥65% accuracy over 20+ scored outcomes
- Starting-pitcher data pipeline is integrated for MLB