Genuinely out-of-sample: predictions recorded before matches were played, graded against real outcomes. Generated 2026-07-11 01:07 UTC. Click any matchup row for a feature-level breakdown of why the model favors that side.
Live from Polymarket US
Pulled directly from Polymarket US's public API when you click refresh —
genuinely live, not a stored snapshot. Model probabilities are the model's current read on
each match (refreshed every record_predictions run, matched by event id) —
available even for matches too far out to be in the forward-test betting window yet; matches
with no estimate at all show market price only. Decision support only — nothing here is
placed automatically.
Recorded
2
Scored
2
Pending
0
Model log loss
1.8521
▲ trails market by 0.6782 log loss
Market log loss
1.1739
Model Brier
0.6813
Total staked
$108.29
Total P&L
-$75.55
▼ -69.8% ROI on 2 bets
Snapshot — bets to consider (0)
Not yet resolved, soonest first. Snapshotted once each match fell into the betting window (see `paper_trading.snapshot_window_min_hours`/`snapshot_window_max_hours`) — a proxy for the price at actual bet-placement time, captured the last time record_predictions ran, not a live feed (see "Live from Polymarket US" above for that). Decision support only, nothing here is placed automatically.
None yet.
Model vs. market log loss over time
ModelMarket
By league (min. 3 scored matches)
No league has enough scored matches yet (need 3+).
Simulated P&L (fractional Kelly, decision support only — no real money)
Scored — graded against real outcomes (2)
Start (ET)
League
Matchup
Model P(A)
Market P(A)
Edge
Stake
Full Kelly %
Result
P&L
2026-07-10 04:00
MSI
LYON vs G2 Esports
0.265
0.248
+0.018
$8.29 on LYON
1.7%
LYON
+24.45
Feature
Raw diff (A − B)
Pull toward LYON
/ G2 Esports
Log-odds contribution
Recent win rate
+0.10
+0.093
Blue/red side win rate
n/a (imputed)
+0.000
Head-to-head record
n/a (imputed)
+0.000
Rest advantage
+0.17
+0.000
Player Elo rating
-63.52
-0.635
Momentum
+0.23
+0.069
Region strength
-18.10
-0.065
Feature sum (log-odds) → implied P(LYON) per game
-0.538 → 36.9%
Best-of-5 series adjustment (Polymarket US prices the series, not one game) → P(LYON wins series)
26.5%
Side (blue/red) isn't known before a match, so the model's real ~53% historical
blue-side edge is averaged over both teams here rather than credited to whichever one happens to be
listed first — otherwise it would silently read as a home-side advantage for an arbitrary label.
LYON avg 1624
Inspired
1730
Dhokla
1656
Saint
1611
Berserker
1566
Isles
1555
G2 Esports avg 1687
Hans Sama
1736
Caps
1733
BrokenBlade
1687
SkewMond
1667
Labrov
1613
2026-07-09 04:00
MSI
Bilibili Gaming vs Hanwha Life Esports
0.093
0.386
-0.293
$100.00 on Hanwha Life Esports
75.5%
Bilibili Gaming
-100.00
Feature
Raw diff (A − B)
Pull toward Bilibili Gaming
/ Hanwha Life Esports
Log-odds contribution
Recent win rate
-0.10
-0.093
Blue/red side win rate
n/a (imputed)
+0.000
Head-to-head record
+0.50
+0.135
Rest advantage
-0.85
-0.000
Player Elo rating
-54.89
-0.549
Momentum
+0.10
+0.030
Region strength
-189.71
-0.682
Feature sum (log-odds) → implied P(Bilibili Gaming) per game
-1.159 → 23.9%
Best-of-5 series adjustment (Polymarket US prices the series, not one game) → P(Bilibili Gaming wins series)
9.3%
Side (blue/red) isn't known before a match, so the model's real ~53% historical
blue-side edge is averaged over both teams here rather than credited to whichever one happens to be
listed first — otherwise it would silently read as a home-side advantage for an arbitrary label.