Are the signals earning trust?

Performance

Daily results β†’
Cohort
Period
Market
All
⚾ Batter
Direction
Assists carries the strongest edge (+23.2 pts over base).

What's working

Every market's measured edge against how much of it we actually serve, ranked by lift. The tables further down read served predictions only, so a market that is good but never clears the floor is invisible there β€” and looks the same as one that does not work.

MarketAll predictionsServed (70+)Share served
Assists+23.2 pp n=275+32.6 pp n=16560%
β€œStarved” means the market beats its base rate by 5+ points across everything it predicted, yet under a tenth of those ever clear the floor β€” the scale cannot reach, so the edge is never offered. Judged on the engine running today: a superseded scorer’s serving share is not a fact about the current one. The columns stay pooled across versions.
Last 90 days Β· Other qualifying91–5462.8%n=1458 void Β· 0 pending Β· avg score 84 Β· 22 slates
All qualifying109–5666.1%n=165
Featured18–290.0%n=20
Other qualifying91–5462.8%n=145
Deeper diagnostics

Market pulse

Recent slate hit rates compared with each market’s own average. The right edge shows the selected period and the latest three slates versus the prior three.

MarketPeriodAug29Aug28Aug27Aug26Aug25Aug24Aug23Aug22
Assists63%n=145↓ -7 pp25%n=462%n=80%n=2100%n=260%n=520%n=570%n=1056%n=9
Above own averageNear own averageBelow own averageFaded cells have fewer than 5 decisions

Does a higher score perform better?

Not enough graded props yet to judge whether higher scores perform better. The 80–84, 85–89, 90–94, 95–98 bands remain small.

Current scoring engine only β€” 97 predictions from superseded versions excluded. Pooling engines once made the top band look anti-predictive when it was not; per-version records are in the model table below.

Score bandRecordHit rateSampleVoidsvs base
75–79Small sample0–0Not gradedn=01
80–84Small sample2–250.0%n=40+0.1 pp base 50%
85–89Small sample2–166.7%n=30+16.8 pp base 50%
90–94Small sample2–0100.0%n=20+50.1 pp base 50%
95–98Small sample5–362.5%n=80+12.6 pp base 50%
99–10020–1852.6%n=380+23.9 pp base 29%

Over vs Under

Over91–5462.8%n=145
Under0–0Not gradedn=0
MarketOverUnder
Assists62.8% n=145Not graded n=0

Edge finder

SegmentHit ratevs baseRecord30-day trend
Assists62.8% n=145+27.2 pp base 36%91–5463% 30d

Consistency

Last 7
52.8%
19–17 n=36
Last 30
62.8%
91–54 n=145
Prev 30
Not graded
0–0
Season
62.8%
91–54 n=145
All time
62.8%
91–54 n=145

By month

MonthRecordHit ratevs base
2026-0891–5462.8% n=145+27.2 pp base 36%

Model versions

VersionActiveRecordHit ratevs base
WNBA Assists
wnba-pra-v42026-08-21β†’2026-08-2931–2456.4% n=55+20.8 pp base 36%
4 earlier versions2026-08-03β†’2026-08-2060–3066.7% n=90+31.1 pp base 36%
All qualifying includes every published prediction scoring 70+. Featured is the main-page eight. Observed hit rates are not probabilities; prediction counts are correlated within the displayed slate count.