Boris Chen revolutionized fantasy football rankings by using a Gaussian Mixture Model to cluster experts into tiers instead of ordinal lists. Within a tier, picks are statistically indistinguishable.
Same principle applies here: a 62% and 65% confidence bet aren't really different. Our model can't reliably tell them apart. Focus on which TIER a pick falls in, not its exact percentage.
📊 Tier Definitions
🟢 Tier 1
60%+ confidence
Take it. Clear edge ~5%+ above break-even.
🟡 Tier 2
55-59% confidence
Selective. Only if odds compensate.
🟠 Tier 3
52-54% confidence
Usually pass. Barely above break-even.
🔴 Tier 4
<52% confidence
Never bet.
Important: Tier alone isn't enough. A Tier 1 (high confidence) pick can still be NEGATIVE EV if the odds are too heavy. See Lesson 1 (EV) for the second filter — both tier AND EV must agree.
🎯 This Week's Picks — Tier Visualization
Each rank-1 pick from this week's analyses plotted by confidence. Vertical lines mark tier boundaries.
Observations: 4 picks cluster in Tier 1 (60%+) and 4 in Tier 2 (55-59%). The mid-Tier 2 picks (56-57%) are statistically equivalent — don't bet one over another based on the 1% difference. Cross-reference with EV (Lesson 1) — some Tier 1 picks have negative EV due to vig (Braves Wed @ -160, COL @ -200).
📖 Key Insights
Don't fight precision the model doesn't have
Saying "this is a 62.3% confidence pick" implies precision we can't deliver. The model estimates 60-65%. Treat them as one bucket.
Tier + EV is the real filter
Tier tells you "is this a confident pick?" EV tells you "do the odds compensate?" Both must say YES. Either alone is misleading.
No tier 5/6/7
Anything below 50% is just an uncorrected loss. We don't grade those — we don't bet them.