⚠ The most important sentence on this page
My analysis didn't beat the market.
Blind favorite betting returned +43.5% in the same 20 games. My analysis returned +39.4%. The framework is honest, not flattering. From this baseline real improvement is possible.
1. Methodology
- 20 closed playoff games analyzed (May 3-13, 2026)
- Predictions built using only pre-game factors — series state, home/away, seeds, recent form
- $5 simulated stake per game ($100 total)
- Standard moneyline + spread + total bet types
- Scored against actual outcomes
- Critical: I knew outcomes when doing this. To stay honest I cited only pre-game factors and tracked predictions vs actual
- Factor performance tracked individually
Why this matters: Without backtesting, "analysis" is just opinion. With it, we have data. This is the prediction-tracking framework in action.
2. Top-line results
Benchmark comparison
| Strategy | ROI | Win rate |
|---|
| My analysis | +39.4% | 80% |
| Blind favorite bet | +43.5% | 83% |
Blind favorite-betting beat my analysis by 4 percentage points. NBA playoffs are largely chalk-driven. The book sets favorites accurately. My "analysis" mostly aligned with the book's view + added some narrative bias that occasionally hurt.
3. Performance by bet type
| Bet Type | Record | Hit% | Staked | Return | P/L | ROI% |
|---|
| ML road favorite | 4-1 | 80% | $25 | $38.87 | +$13.87 | +55% |
| ML heavy favorite | 3-0 | 100% | $15 | $21.17 | +$6.17 | +41% |
| ML favorite | 4-1 | 80% | $25 | $33.90 | +$8.90 | +36% |
| ML favorite bounce-back | 1-0 | 100% | $5 | $7.78 | +$2.78 | +56% |
| ML home momentum | 1-0 | 100% | $5 | $9.55 | +$4.55 | +91% |
| Under | 1-0 | 100% | $5 | $9.55 | +$4.55 | +91% |
| ML home underdog | 1-1 | 50% | $10 | $10.25 | +$0.25 | +2% |
| ML favorite pivotal | 1-1 | 50% | $10 | $8.33 | -$1.67 | -17% |
The loser: "ML favorite pivotal" at -17% ROI is the only category that actively lost money. "Home court in pivotal G5" was over-weighted in my analysis.
4. The 4 losses — detailed breakdown
Loss 1 · SAS vs MIN G1 (May 4)
My pick: SAS ML at -150 · my probability: 60% · factors: home_court, higher_seed, rest_advantage
Actual: MIN won 104-102
What I missed: SAS's 5+ days rest became RUST. MIN had momentum from beating DEN in 6 games. Rest advantage flipped.
Lesson: 5+ days of rest can hurt, not help, when the opponent has momentum.
Loss 2 · MIN vs SAS G3 (May 8)
My pick: MIN ML at +110 (home underdog) · my probability: 55% · factors: home_court_for_underdog, MIN_won_G1
Actual: SAS won 115-108
What I missed: Home court for underdog doesn't outweigh talent gap. SAS proved they could win on road.
Lesson: Don't bet underdogs based on home court alone in playoffs.
Loss 3 · CLE vs DET G3 (May 9)
My pick: DET ML at -110 (road favorite) · my probability: 52% · factors: DET_won_first_two, still_favorite_despite_road
Actual: CLE won 116-109
What I missed: Team down 0-2 at home is MORE dangerous than line suggests. CLE found another gear that I didn't price in.
Lesson: Teams facing 0-3 elimination scenarios are dangerous, especially at home.
Loss 4 · DET vs CLE G5 (May 13)
My pick: DET ML at -130 (home pivotal) · my probability: 58% · factors: home_court_pivotal_g5, DET_must_hold_home
Actual: CLE won 117-113 (Strus shot 6/8 from 3)
What I missed: Outlier shooting performance from a role player. "Home G5 advantage" was already priced in by the line.
Lesson: Home G5 isn't automatic. Outlier shooting nights happen. "Pivotal home" factor doesn't add edge over market line.
5. Factor accuracy ranking
Factors that predicted accurately
✓home_court_g72/2 = 100%
✓1_seed_dominant2/2 = 100%
✓rest_advantage_after_sweep1/1
✓bounce_back_at_home_after_loss1/1
✓NYK/OKC talent_gapmultiple, 100%
Factors that were noise
✗home_court_pivotal_g51/2 = 50% (actively hurt)
✗home_court_for_underdog0/1 = 0%
✗road_favorite_despite_desperation0/1 (talent didn't save DET)
General factors (small sample but reasonable)
≈home_court7/8 = 88% — but most were also favorites
≈higher_seed3/4 = 75% — overlaps with favorite
Caveat: most factor counts are 1-2 samples. We need 20-30+ data points per factor for real signal. This is a baseline, not a conclusion.
6. Themes in the losses
All 4 losses shared a theme: I weighted situational/narrative factors over talent/odds.
Specifically, I lost when I:
- Trusted "rest advantage" over momentum (SAS G1)
- Trusted "home court for underdog" over talent (MIN G3)
- Trusted "road favorite still favored" over urgency factor (DET G3)
- Trusted "home pivotal G5" over market efficiency (DET G5)
The pattern: I was trying to find "narrative edges" the book missed. The book was right and I was wrong.
Actual lesson: For NBA playoffs, the line is usually accurate. Edges come from situational factors the books may not weight enough — not from narrative factors that everyone already knows.
7. Strategy implications
What works — continue
- Picking heavy favorites in talent-mismatch series (OKC vs LAL)
- Road favorites in dominant series (talent travels)
- Bounce-back at home after surprising loss (small sample)
- Series-closing games at home (G7 home court)
What doesn't work — stop
- "Pivotal home game" narrative — already priced in
- Home underdogs without talent advantage
- Overriding talent gap with "must-win" narratives
- Road favorites against desperate home underdogs
What needs more data
- Bounce-back factor (1/1 so far)
- 0-2 home team desperate (1/1 against my pick)
- Under after blowout (1/1)
- Specific player matchup effects
8. Honest framing on the +39% ROI
Q: If my analysis returned +39.4% ROI, isn't that great?
A: Yes and no.
Yes — $39 profit on $100 in 10 days is real money.
No because:
- Sample size is 20 games — that's noise, not signal
- Blind favorite betting beat it (+43.5%)
- The "analysis" mostly aligned with book lines anyway
- The 4 losses came from active analysis overriding the line
- The "wins" mostly came from agreeing with the line (favorites)
Real edge test: after 100+ games tracked this way, does my analysis consistently beat blind favorite-betting? Right now: probably no. That's the honest answer. The prediction tracking system is built for exactly this question. Track 100 games, then we know.
9. What this means for methodology
Proposed methodology update: for each prediction, score against TWO benchmarks:
- Was my pick correct? (binary)
- Did my analysis beat "just bet favorite"? (relative)
Track over time:
- Factor accuracy (which signals are real)
- Bet type performance (where I have edge vs market)
- Confidence calibration (am I right when I'm sure, wrong when I'm doubting)
After 50 games tracked: probably honest enough sample to know what I'm actually good at predicting.
Target: identify 2-3 bet types where my edge is real. Bet bigger on those. Avoid the rest.
📌 Key message
My analysis didn't beat the market in this 20-game sample. Blind favorite-betting outperformed by 4pp. That doesn't mean we stop analyzing — it means we know honestly where we stand. From this baseline, real improvement is possible. Without honest backtesting, we'd assume we were better than we are. This is the system working. It's not flattering, but it's honest.