🧪 STRATEGY LAB
EXPERIMENTAL MODEL v2 (BLENDED ELO + EDGE CAP) · A/B vs THE LIVE MODEL
← live scorecard

🧪 STRATEGY LAB · MODEL v2

Experimental model, tracked separately from the frozen live one — sharp-bettor discipline: market-anchored (25% model / 75% market), conservative Kelly, CLV-benchmarked.
ⓘ the 5 v2 rules
(1) Ratings blend live Elo 50% + FIFA-ranking 50% (kills single-source outliers like Ghana). (2) Market-anchored — final prob = 25% model + 75% de-vigged market, because the closing line is the most efficient price around (you must earn the right to disagree). (3) Edge sanity cap 25%. (4) Kelly on a conservative edge (−3pt haircut), EV net of 2% costs. (5) Benchmarked on Closing Line Value, not hypothetical EV. Next: isotonic calibration once enough matches are scored.

⚖️ v1 vs v2 vs MARKET

V1 BRIER
0.477
lower=better
V2 BRIER
0.419
lower=better
MARKET BRIER
0.410
✅ best
Head-to-head Brier over 47 scored match(es). Lower is better — this is the honest test of whether v2's changes actually help.
v2 EDGE PICKS
10
0% hit
v2 EDGE ROI
-100%
capped at 25%
FAVOURITE ROI
+17%
baseline
CLOSING LINE VALUE · 11 PICKS
-1.1
avg CLV (¢ of price) · beat the close 2/11

📉 CLV BY PICK

United States v Australiatook 0.17 → closed 0.18+1.0
Scotland v Moroccotook 0.17 → closed 0.15-2.0
Belgium v Irantook 0.12 → closed 0.10-1.0
Jordan v Algeriatook 0.15 → closed 0.15+0.0
Bosnia-Herzegovina v Qatartook 0.15 → closed 0.12-3.0
Curaçao v Ivory Coasttook 0.10 → closed 0.12+2.1
Paraguay v Australiatook 0.28 → closed 0.23-5.0
Senegal v Iraqtook 0.07 → closed 0.07+0.0
Egypt v Irantook 0.24 → closed 0.24+0.0
Congo DR v Uzbekistantook 0.21 → closed 0.17-4.0
Algeria v Austriatook 0.26 → closed 0.26+0.0
ⓘ how this works
CLV = closing price − price you took. Positive = the market moved toward your pick after you bet — the surest sign of real edge, and it shows up long before P&L. If you can't beat the close, no staking scheme rescues you.

📈 v2 EDGE PICKS (LIVE) · soonest first · tap a row for the why

No v2 edges ≥3% right now — after anchoring to the market, the model rarely disagrees enough to bet. That's the point.
MODEL v2 · TRACKED SEPARATELY · PREDICTIONS LOGGED PRE-KICKOFF · NOT BETTING ADVICE