Model analytics
Three views on how our signals are performing — the overall record across both feeds, the per-model leaderboard, and how stacked confluence picks pay off versus a single signal. Switch views with the tabs below.
How this is graded: every featured Sharp Signal and Game Report logged at posting time, graded at the captured closing price on a flat-unit basis. Pushes, voids, and split games (the two feeds on opposite sides of the same market) are excluded from win-rate math. ROI shown is real-odds-derived where the signal has at least 30 picks with captured American odds; a -110-implied estimate is used as a fallback below that threshold (each card carries the source label). The per-model leaderboard (Models tab) covers Sharp Signals only — this strip is the pooled view across both feeds. NBA picks are paused for the 2026 playoffs (Apr 19 – Jun 30); regular-season models resume next season. Past performance is not predictive.
What these numbers mean
Each row represents one of our proprietary models. We don’t publish what each model is measuring — that’s the edge. What you get is how it has performed in the selected window: win rate (wins out of decided picks), ROI (return on a flat $100 unit), total picks, and a 7-day snapshot. The roster is rebuilt continuously: a model only appears here if it fired a featured pick in the last 7 days. The pipeline is constantly being tuned, so a model that was firing a few days ago can still be analyzed even if it no longer meets the threshold to send today — in that case it stays on the page with an INACTIVE badge until it falls out of the 7-day window entirely. Stats reflect the tab you’ve picked, so a model can look hot on 90d and cold on 7d (or vice versa). Models are sorted by profit, descending. Pending picks (live games, awaiting result) are counted in the pick total but excluded from win-rate math. Void picks — refunded because the line couldn’t be graded (player DNP, postponement, etc.) — are excluded from the total entirely; when present, they show as a separate “X void” tail on the model’s meta line. SportsGods reserves the right to hide, restrict, or re-rank any model on this page at any time without notice; see Terms § 9. All figures are informational only and may be inaccurate — always verify at the sportsbook before wagering.
Same signal pool as the Sharp Signals cards. Model pseudonyms (A, B, C…) are stable, but scoped per sport — Model X on an NBA pick and Model X on an MLB pick are different underlying signals. Pick cards render the full sport-suffixed label (e.g. Model X-MLB) to make this explicit; this leaderboard keeps the bare letter because the sport is already shown as a chip next to it. Roster includes any model that fired a featured pick in the last 7 days; INACTIVE means the model hasn’t fired in the last 48 hours and may not be sending picks tomorrow. SportsGods may add, hide, restrict, or re-rank any model on this page at any time, with or without notice.
How confluence picks have actually performed
When a Sharp Signal card shows a 2x, 3x, or 4x CONFLUENCE chip, the Edge column below is the historical lift in win rate vs the 1x baseline (single-signal picks of the same sport) — the natural comparison for asking "does stacking models actually move the needle, or is the chip mostly cosmetic?"
- Same dedupe + filters as the Sharp Signals feed — distinct picks grouped by (sport, game, bet); soak-test signals excluded; NBA regular-season picks paused during playoffs (
PROP_*player props are exempt). - Win rate excludes pushes and pending. Edge is the percentage-point lift over the same sport's 1x baseline; ROI is flat-unit (profit ÷ total $100 stakes) using captured real-odds where available and a -110 fallback otherwise.
- Lifetime view over the last 180 days; the Models tab's 7/30/90-day window tabs do not change these numbers.
- Small-N caveat — rows tagged small are below ~20 decided picks; the Edge value is suggestive, not predictive. Rows with fewer than 5 decided picks have their color cues removed.