← All guides / Signal Lab

Signal Lab

What it is

Signal Lab is the historical-backtest view of edge categories. Every edge that hits the platform falls into a signal bucket: 'soft book lag', 'model disagreement', 'sharp consensus diverging from public', etc. Signal Lab shows the historical hit rate, ROI, and CLV for each bucket, sliced by sport and market.

When to use it

How to read it

FieldMeaning
SignalCategory label. E.g. soft_book_lag, model_consensus_disagree, sharp_steam.
Sport / MarketSlicing dimension. Same signal can perform differently across sports.
Sample SizeHow many historical bets in this bucket.
Hit Rate% of bets that won.
ROINet profit divided by total wagered, expressed as a percentage.
Avg CLVAverage closing line value across this bucket. Strong predictor of future performance.
SharpeRisk-adjusted return. High mean / low variance is what you want.

Worked example

Example
Signal:       soft_book_lag
Sport:        NBA
Market:       player_props
n:            12,400
Hit rate:     53.8%
ROI:          +4.2%
Avg CLV:      +1.7%
Verdict:      Real edge, modest variance, scale up

The 'soft_book_lag' signal on NBA player props has shown a 4.2% ROI over 12,000+ bets with positive CLV. That is durable. Compare to a signal with 200 bets, 7% ROI, negative CLV: that is luck masquerading as skill.

Common mistakes

  1. Trusting small sample sizes. Anything under ~500 graded bets is suspect. Large samples are the only way to separate signal from luck.
  2. Ignoring CLV. Hit rate without CLV is just variance. CLV confirms the edge is real.
  3. Cherry-picking signals. If you only run signals after they have a good month, you are chasing variance.
  4. Mixing signals into one bet. Two signals on the same bet are not independent. Treat the bet as one signal.
  5. Forgetting market regime. A signal that worked in 2023 NBA may not work in 2026. Recheck regularly.

Related