Excess ReturnFor every date and horizon: stock forward return − equal-weight active-universe forward return. Positive excess means the signal beat its contemporaneous market universe.
Advanced Rankv4.4 uses Advanced Score for cross-sectional ranking after it passed separate development, validation and untouched holdout checks. Elite Setup calibration remains separate for signal labels.
ConfidenceSample confidence rises from Low to Medium, High and Very High as evaluated observations increase. Small groups are intentionally discounted in the validation score.
No Look-aheadHistorical scores use data only through the replay date. Entry and every forward horizon require an exact positive close on the common market trading date; future OHLC is used only for evaluation.
Strong Buy TrustThe frozen Advanced Rank top-8% Strong Buy cohort is validated separately across 20D, 40D, 60D, 100D and all history. BUY NOW remains locked unless model reliability and current trust are acceptable.
Trading Dates0automatically reconstructed
Base Observations0exact signal-date close available
9D Evaluable0exact 9th market-date close available
9D Strong Buy Excess—average return above universe
Evidence Quality—diagnostic score, not forecast accuracy
Advanced Rank Quality—top percentiles vs bottom 20%
Model Reliability—rank, holdout and signal stability combined
Holdout Rank—final 20% of trading dates
Calibration Health—automatic rolling stability status
Decision Gate—scanner permission for new entries
SB Time Consistency—positive 9D edge across rolling blocks
Rolling Calibration Health
Compares calibrated Final Signals across 20D, 40D, 60D, 100D and all history. v4.4 separates actionable Strong Buy evidence from the independently validated Advanced Rank and excludes non-trading observations.
| Window | Dates | Samples | Strong Buy 9D | Buy 9D | Watch 9D | Avoid 9D | 20D SB−Avoid Spread | Ranking Quality | Signal Ordering | Health |
|---|
| Rolling calibration health will be calculated automatically from downloaded OHLC data. |
Strong Buy Trust—current evidence gate
SB Samples0all-history calibrated samples
SB 9D Edge—vs active universe
SB 20D Edge—vs active universe
SB 9D 95% Range—Newey–West date-clustered diagnostic
Strong Buy Trust Validation
Separates Strong Buy reliability from overall ranking quality. Trust improves when Strong Buy has positive excess, beats Avoid, has adequate samples, and remains stable across medium windows.
| Window | SB Samples | 6D SB | 9D SB | 15D SB | 20D SB | 9D SB−Avoid | 20D SB−Avoid | Trust |
|---|
| Strong Buy trust will be calculated automatically. |
Strong Buy Comparative Validation
Shows whether calibrated Strong Buy adds value versus Buy, Watch, Avoid and every non-Strong-Buy observation. This remains the primary actionable-signal diagnostic in v4.4.
| Comparison | 3D Spread | 6D Spread | 9D Spread | 15D Spread | 20D Spread |
|---|
| Strong Buy comparative evidence will be calculated automatically. |
Strong Buy Time Consistency
Splits history into non-overlapping 20-trading-date blocks. A robust edge should appear across multiple periods rather than depend on only one unusually profitable regime.
| Period | Dates | SB Samples | 9D SB | 9D SB−Non-SB | 20D SB−Non-SB | Result |
|---|
| Time consistency will be calculated automatically. |
Chronological Holdout Validation
Evaluates the frozen Advanced Rank on the first 60%, next 20% and untouched final 20% of dates. No period is shuffled, and Strong Buy is reported separately from generic rank skill.
| Period | Dates | Samples | Rank Quality | Top 10% 9D | Bottom 20% 9D | 9D Spread | SB Samples | SB 9D | Rank Test |
|---|
| Chronological holdout validation will be calculated automatically. |
Reliability reasons will appear after reconstruction.
Legacy / Raw Signal Validation
Diagnostic only. This table validates the older raw signal vocabulary and is not used for BUY NOW. A zero Strong Buy count here does not mean the calibrated Final Signal has no Strong Buy observations.
| Signal | Samples | Confidence | 3D Raw | 3D Excess | 6D Raw | 6D Excess | 9D Raw | 9D Excess | 9D Excess Win% | 15D Excess | 20D Excess | MFE | MAE | Validation |
|---|
| Elite performance will be reconstructed automatically from downloaded OHLC data. |
Final Rank-Signal Validation
Reconstructs the frozen Advanced Rank bands: Strong Buy top 8%, Buy 8–20%, Watch 20–80% and Avoid bottom 20%. Tradeability and entry readiness remain separate decision gates.
| Final Signal | Samples | Confidence | 3D Excess | 6D Excess | 9D Excess | 9D Excess Win% | 15D Excess | 20D Excess | MFE | MAE | Validation |
|---|
| Calibrated signal validation will appear after automatic reconstruction. |
Advanced Rank Percentile-Band Validation
Non-overlapping per-date Advanced Rank bands test whether better-ranked stocks deliver stronger forward excess. These bands do not redefine Strong Buy labels.
| Percentile Band | Samples | Confidence | 3D Excess | 6D Excess | 9D Excess | 9D Excess Win% | 15D Excess | 20D Excess | MFE | MAE | Validation |
|---|
| Percentile-band validation will appear after automatic reconstruction. |
Advanced Ranking Diagnostic
Tests broad cross-sectional Advanced Rank quality. Rank skill and Strong Buy trust are separate requirements inside the v4.4 reliability gate.
| Rank Group | Samples | Confidence | 3D Excess | 6D Excess | 9D Excess | 9D Excess Win% | 15D Excess | 20D Excess | Validation |
|---|
| Percentile ranking validation will appear after automatic reconstruction. |
Universe Context
The full reconstructed universe should have approximately zero excess return versus its own equal-weight benchmark; this table is context, not a ranking-skill score.
| Horizon | Samples | Raw Avg | Benchmark Avg | Excess Avg | Excess Win% |
|---|
| Holding-period analysis will appear after automatic reconstruction. |