Candidate evidence
Each sample produces several imperfect clues. Median absolute deviation keeps the baseline useful when the history contains occasional spikes; slope and volatility detect motion rather than only magnitude.
Signal intelligence // telemetry
Shape before threshold.
The detector compares the active metric trajectory with historical normal motifs. Robust level, slope, volatility, derivative-DTW distance, and continuity state are combined so a sustained drift can become visible before a static red line is crossed.
detector stateloading
shape score-
evidence time-
contacting specular-edge
Trajectory evidence
Select a metric. The line is the persisted edge history; the amber area marks warning or critical observations.
No evidence loaded.
Current explanation
- Value
- -
- Robust z
- -
- Slope z
- -
- Volatility ratio
- -
- Derivative DTW
- -
- Decoder confidence
- -
- First divergence
- -
- Warm-up remaining
- -
Constrained alignment
The active window is compared with earlier normal windows using a Sakoe-Chiba-bounded DTW path. Derivatives carry more weight than absolute level, which makes trajectory geometry explicit without erasing amplitude evidence.
Continuity decoder
State transitions penalise implausible jumps. A single sample can raise the score, but warning and critical states require temporal support. Recovery also has to earn its way back to normal.
Metric evidence
| Metric | State | Score | Value | DTW | Slope z | Confidence |
|---|---|---|---|---|---|---|
| Loading. | ||||||