Model CLV: aggregate audit

Closing-line-value audit of the platform's picks: the ones we recommended, and every market we priced. CLV = closing price − the price we took, raw implied on both sides. The model is not in that formula, which is what makes it a test rather than a restatement of our own opinion. Positive mean CLV means our picks beat the close, the standard proxy for sharp skill.

+0.21pp
Ahead of the close on average; the range still includes zero.
Mean closing-line value across 79 recommended picks with a closing line; the price moved our way on 41.8% of them. 95% range -0.5 to +0.9pp, 0.79pp short of our +1.00pp target.

Both prices are the best on the board at their moment, when we posted and at the last capture before kickoff: this measures whether the market came toward us after we posted, not any one book's close.

Since 5 October 2026 our model starts from the market's price, so calls are rarer. Earlier calls stay in this record.

-0.68pp
Not beating the close.
Every market we priced, recommended or not: 4,694 markets; the price did not move our way on 62.0% of them. 95% range -0.8 to -0.6pp, 1.68pp short of our +1.00pp target.

This is the model's raw number, before the selection gate, and the one the plan below is about.

Bar for selling picks +0.00pp Target +1.00pp
Window (by game date):
30 days 90 days All time Since 12 Sep fix
View:
Per sport By sport + bet type
Methodology: The close is the last price captured before kickoff, and a row is recomputed whenever a fresher close is captured. Buckets with N<30 are hidden. The Since 12 Sep fix window starts at the closing-line capture fix of 12 September 2026, the cohort the selling test is judged on. CLV compares the best price available when we posted the pick with the best price available at the last capture before kickoff. Both ends are best-of-board, and the close is captured across at least as many books as the board we post from, Pinnacle where it is quoted, so the comparison does not tilt in our favour.

Every market we priced, by sport

Every market we priced, by sport: model CLV per bucket.
Sport N Mean CLV Median CLV % beating close
MLB 2,207 -0.55pp -0.57pp 39.1%
WTA Tennis 540 -0.41pp -0.22pp 42.4%
ATP Tennis 458 -0.95pp -0.76pp 38.9%
NCAAF 414 -1.78pp -0.92pp 18.6%
UFC 316 -0.96pp +0.00pp 48.4%
La Liga 150 -1.04pp -1.00pp 38.0%
EPL 106 +0.66pp -1.39pp 40.6%
Serie A 99 -1.57pp -2.29pp 33.3%
Ligue 1 95 +0.45pp -0.22pp 48.4%
NHL 81 -0.74pp -0.42pp 37.0%
NFL 79 -1.11pp -0.92pp 39.2%
Bundesliga 70 +1.97pp +0.19pp 51.4%
Champions League 32 -1.69pp -2.69pp 28.1%

Cumulative-mean CLV trajectory

Cumulative mean CLV, every market we priced · N=4,694 · 9 Jul 2026 to 7 Oct 2026 · ends at -0.68pp

What we are doing about it

A number like the one above is worth little without the response to it, so here is ours, and the figure that will show whether it worked.

  1. Narrow to where we are least behind. CLV is not evenly distributed. Today the largest sample is MLB, 2,207 markets at -0.55pp, and the widest gap is NCAAF, -1.78pp on 414. Coverage follows the evidence rather than the calendar.
  2. Pull the model toward the market where it disagrees most. In the band where our probability departs furthest from the closing price, the market has been the better estimator. That is a calibration problem with a known direction, not a mystery.
  3. Fix the instrument before trusting it. Some of what this page measured was measurement error, found and corrected in September 2026. Numbers here predate and postdate those fixes; the trajectory above, by game date, is the honest view.

The test: mean CLV on the picks we recommend, non-negative at 95% confidence. Today, on this window: +0.21pp, range -0.5 to +0.9pp, on 79 picks. Does not clear it. Until it does we are not selling picks, and this page will keep saying so.

See /trust for live calibration buckets across sports and /methodology for the signal pipeline behind each pick.