Quantprove returns three 0-100 scores from one CSV. Edge Score grades a backtest across four buckets with six named tiers (86+ Exceptional down to below 21 No Edge). Stability Score measures how closely live trades track that backtest, with 60+ flagged stable. Health Score monitors rolling decay across the live trade record.
Quantprove produces three scores from your CSV trade log, one per mode. Edge Score comes from Backtest and grades a single backtest log 0-100. Stability Score comes from Validation and compares a backtest log against a live log 0-100. Health Score comes from Monitor and tracks a live strategy’s edge across the trade record as new trades arrive.
Each score answers a different question. Edge Score asks whether the historical record shows a statistical edge. Stability Score asks whether live execution reproduces that edge. Health Score asks whether the edge is holding or fading. You read them in that order: Edge first, then Stability, then Health.
All three start from the same per-trade data and never annualize R-multiples. The smart parser auto-detects whether your column holds R-multiples or dollar P&L, so the scores read the unit you uploaded.
Three scores, one CSV: Edge Score (Backtest) grades the past, Stability Score (Validation) checks the present, Health Score (Monitor) watches the trend.
Edge Score grades one backtest log 0-100 across four buckets that sum to 100 raw points: Edge Magnitude (30), Edge Consistency (25), Downside Risk (25), and Tradability (20). Inside those buckets sit metrics like EV per trade (15 points, caps at 0.50), Sortino (14 points, caps at 1.80), CVaR 95, and Loss Streak Distribution. The raw total then scales by a confidence multiplier.
Edge Score uses six named tiers so you can place a strategy at a glance: 86+ Exceptional, 71-85 Strong Edge, 56-70 Moderate Edge, 41-55 Developing, 21-40 Weak Edge, and below 21 No Edge. A Strong Edge or higher suggests the backtest is worth carrying into Validation. A Developing or Weak score suggests the edge is thin or the sample is still small.
Read the bucket breakdown, not the headline number alone. A 60 driven by strong magnitude but weak downside control reads differently from a 60 with even contributions. The Edge Score glossary entry holds the canonical definition and per-metric weights.
The confidence multiplier discounts scores built on thin samples. For Edge Score the formula is max(0.2, min(1.0, sqrt(n)/sqrt(500))). The multiplier reaches 1.0 at 500 trades and shrinks as the sample falls. The max(0.2, ...) term is a floor: an Edge Score never drops below 20% of its raw value. A 20-trade backtest sits near that 0.2 floor, so a raw 90 reports closer to 18.
This reflects how small samples overstate edge. The deflated Sharpe work of Bailey and López de Prado (2014) shows performance metrics inflate under limited data and selection. The multiplier applies that caution numerically: fewer trades, larger discount. No exceptions for pretty curves.
Stability Score uses the same square-root shape but has no 0.2 floor and keys off the live sample only. Read the next section for that difference.
Edge Score floor: max(0.2, min(1.0, sqrt(n)/sqrt(500))). A 20-trade backtest reports near 20% of its raw points.
Stability Score compares a backtest log against a live log and returns 0-100. It weighs four buckets: PNL Distribution (30, with an Anderson-Darling distribution test worth 18 and Win-Loss split worth 12), Drawdown Profile (30), Edge Quality (20, via Sortino Retention and EV Retention), and Annual Return (20). Higher means live trades track the backtest more closely.
Stability Score carries no tier labels. It is a number, and Quantprove flags a strategy stable at 60 or above. Read it qualitatively: a high score indicates live results reproduce the backtest’s distribution and drawdown shape; a low score indicates the live record has drifted from what the backtest showed. The Exceptional and Strong labels belong to Edge Score only.
Its confidence multiplier is min(1.0, sqrt(n_live)/sqrt(500)) with no floor, and it depends only on the live sample size. With few live trades the multiplier pulls the score toward zero, which is intended: a five-trade live log carries almost no confirming power. The Stability Score glossary entry is the canonical reference.
Health Score recomputes a rolling Edge Score across a continuous live record and plots it against trade number on the X axis, never dates. Monitor needs a minimum of 100 trades. Window sizing is adaptive: under 200 total trades it uses a window of 20 and a step of 8; at 200 or more it uses a window of 50 and a step of 15.
Health Score reports five verdict bands: 80+ Strong, 60-79 Promising, 40-59 Weak, 20-39 Poor, and below 20 No Edge. Decay detection compares the mean of the first three windows against the mean of the last three. A later mean below the earlier mean indicates the rolling edge has softened across the record.
Monitor is informs-only. It describes what the windows show and never prescribes an action. You read the trend and the band; Quantprove states the observation and stops there.
Health Score bands: 80+ Strong, 60-79 Promising, 40-59 Weak, 20-39 Poor, below 20 No Edge. Decay = first 3 windows vs last 3 windows.
The scores form a sequence: Edge to Stability to Health. A backtest earns an Edge Score. If that Edge Score reaches Strong or above, you run Validation to earn a Stability Score that checks whether live trades reproduce the backtest. Once a strategy is live with at least 100 trades, Monitor tracks its Health Score across the record.
Each step narrows the question. Edge Score asks whether an edge existed historically. Stability Score asks whether the edge survived contact with real execution. Health Score asks whether the edge is still present as the record grows. A strong Edge Score with a weak Stability Score is the overfitting signature: the backtest looked clean, the live record did not follow, and the Edge Quality bucket plus the distribution test surface the gap.
Start with the getting-started guide for a clean upload, then read the Quantprove metric glossary for every metric the three scores draw on.
Four errors distort the scores most. Watch for each before you trust a number.