Hermēs DocumentationConfluence Stars Are Not a Win Probability
Inspect overlapping levels, shared sources and evidence independence before interpreting a score.
Use stars to prioritize attention, not to infer a win rate. In Confluence and the Levels Reading Board, inspect the components before the score.
| Agreement | Interpretation |
|---|---|
| Two levels derived from the same measure overlap | May be a definitional relationship, not additional evidence |
| Different measures cluster nearby | A candidate overlap; verify time and scope |
| Different observations support a prewritten scenario | Worth testing, but not automatically a calibrated probability |
Under the existing Basis definition, net-GEX +Γ and C1 refer to the same peak. Their overlap is not two independent model votes.
Check a chart area
- Record each level's name, metric basis and expiration scope.
- Identify duplicate expressions of the same underlying measure.
- Align source times and price axes.
- Specify the expected reaction and invalidation before observing the outcome.
- Advance replay without relocating the area to fit subsequent price action.
If source independence cannot be verified from the available information, mark it as unknown. More labels do not establish independent evidence.
Review exercise
Record supported, unsupported, untriggered and insufficient-data cases using the same area definition and preselected observation window. A demonstration, star count or UI confidence label is not an out-of-sample trading win rate.
Next: Replay · Dark-pool checks.