How to read results
The tool looks for a zone where data behavior changes. The detected zone tells you where to look.
The before/after table shows what changes: mean, variance, slope, volatility. Contributions show which columns move most around the zone.
Xi/TCI metrics are displayed only in AI train_loss / val_loss mode. For sensors, series and tables, the reading remains experimental.
Short glossary
Detected zone : place where behavior seems to change.
Contribution : column that moves most around the zone.
Change score : indicative intensity of change.
Examples : some are public/classic, others are synthetic or educational.
Data
Load or paste a CSV. The engine automatically chooses the mode: canonical AI if train / validation losses exist, otherwise experimental multi-column reading.
Built-in examples: classic public series, synthetic examples and educational edge cases. Hover over a button to read what it is.
Selected state columns
State reading
Displayed metrics
Official metrics are displayed only in canonical AI mode.
Change score : indicative intensity of before/after change. The higher it is, the more the zone deserves examination. This score remains educational.
| Metric | Value / status | Reading |
|---|
Multivariate before / after
Mean = general level. Variance = dispersion. Slope = trend. Volatility = local agitation.
If the mean changes, the signal level shifts. If variance or volatility increases, the regime becomes less stable. If the slope changes, the trend reverses or accelerates.
| Measure | Before | After | Delta |
|---|
Column contributions
In experimental mode, contribution = relative size of before/after change per normalized column. It helps locate what moves, without certified diagnosis.
Simple null models
Local comparison with 40 shuffled versions of the state columns. Educational control, not definitive statistical proof.
Displayed limits
AI mode is faithful to the pack for train_loss / val_loss logs. Multi-column mode is an experimental reading inspired by INSACERMO-U: it is not yet the canonical Xi/TCI protocol from the reproducible pack. Exploration tool only: no certified medical, industrial or scientific diagnosis, no automatic decision. Data remains local in the browser. Do not use it to target, exploit, disrupt or profile systems or people without authorization.