Every graded pick we've ever made. Timestamped before the
fight, graded after. Nothing removed, ever.
Following only the recommended bets
+$75
28-21 on
49 recommended bets, 5.9% return on money staked,
on a $1,000 bankroll at the locked fair price.
The scoreboard below counts EVERY prediction,
wins and losses alike; that is the receipts. This line is what the bet tags on the
picks page actually paid. Most picks carry no bet tag: no edge at the price means
no bet, and coin flips stay watched, not wagered. Recompute it yourself from the
raw data below.
117
picks graded
80
winners
68.4%
hit rate
65.1%
pre-launch validation (897 bouts)
+$1,079
flat $100 per pick (10.4% roi)
39/58
locked picks that beat the closing line
Flat-bet line assumes $100
on every graded pick at the fair no-vig market price recorded when the pick locked (for the
few picks locked before the books priced the bout, the closing line is used instead). Prices
at real books differ. Analytics, not betting instructions.
Do not take our word for the record above. Replay it. Pick a bankroll and a
strategy, and this chart walks through every graded pick in order, at the fair no-vig price
recorded when each pick locked. Losses included, obviously.
...
Fair no-vig prices, not real book prices, and past
picks do not predict future ones. Analytics, not betting instructions.
Audit us
The whole ledger is machine readable: every pick with its lock timestamp, our
probability, the market at lock, the closing line, and the result. Pull it, run your own
numbers, catch us if you can. No tout service offers this because no tout service could
survive it.
curl https://cagemath.com/api/record
How honest is the number itself?
Accuracy alone can be gamed by only picking favorites. The harder test is
calibration: when we say a percentage, does reality agree? Measured on 897 blind
predictions before launch:
When the model says
The pick actually wins
50-59%
56.9%
60-69%
68.8%
70-79%
76.6%
The same test, running live
The table above was the pre-launch exam. This one rebuilds itself from the
public ledger as it grows, so you can watch the calibration prove itself (or not) in real
time. Small samples show as counts; percentages have to be earned.
CAGE