How Vision Grade works.
Vision Grade (VG) is a 0–100 score for the strength of a strategy's validation evidence. It is not a probability of profit: VG 82 does not mean an 82% chance of making money.
Vision Grade assesses evidence quality and strategy characteristics. It does not predict future returns, guarantee profitability, or constitute investment advice. Weights, thresholds and caps are AlgoNexta policy, not academic constants, and VG is not an academically standardized or regulator-approved score.
Ten evidence domains (weights sum to 100)
A · Integrity & anti-bias
12 ptsLook-ahead, repainting and data-leak controls
B · Out-of-sample validity
15 ptsGenuine out-of-sample evidence, not a re-sliced backtest
C · Multiple-testing control
12 ptsTrial registry, effective trial count, DSR / SPA-style correction
D · Overfitting control
12 ptsPBO / CSCV and parameter-neighborhood checks
E · Statistical evidence
12 ptsPSR / DSR and confidence intervals, not point estimates
F · Economic realism
10 ptsCosts, slippage and implementation evidence
G · Robustness
10 ptsStress behavior, regimes and sensitivity
H · Drawdown & tail risk
7 ptsDrawdown depth, recovery and tail behavior
I · Sample adequacy
5 ptsEnough independent observations to judge at all
J · Forward / live evidence
5 ptsLocked-version forward or live track record
Mandatory gates
Any failed gate means NOT GRADED — VALIDATION FAILED. A gate that is not yet verified is never treated as passed (NOT GRADED — VALIDATION INCOMPLETE).
- • G1 · No look-ahead bias
- • G2 · No undisclosed repainting
- • G3 · Realistic costs included
- • G4 · Genuine out-of-sample evidence
- • G5 · Reproducible evaluation
Score caps
VG = Σ(weight × domain score) ÷ Σ weight × 100, each domain scored 0, 0.5 or 1. Bands: 90+ Exceptional · 80–89 Strong · 70–79 Substantial · 60–69 Moderate · 50–59 Limited · below 50 Weak.
- • No forward or live evidence → maximum 89
- • Effective trial count unknown → maximum 69
- • Broad search without formal multiple-testing correction → maximum 79
- • Inadequate sample → "Insufficient Evidence", not a grade
Required inputs
- • Locked strategy version
- • Evidence manifest and data window
- • Cost and execution assumptions
- • Data snapshot id and evaluation engine version
- • Validation results and known gaps
Regrade triggers
- • Source or parameter change
- • New validation or forward evidence
- • Evidence integrity issue
- • Published methodology change
Integrity rules
- 1. The same normalized inputs, evidence package, and methodology version must produce the same result.
- 2. A published assessment cannot silently change; new inputs create a new snapshot.
- 3. Missing evidence reduces confidence and must remain visible to the reader.
- 4. Commercial relationships cannot change the computed score.
See the methodology in context
Trace a sample grade back to its evidence.