Signals
Four things worth detecting.
Most operational data is inert. These are the four shapes that reliably carry a decision inside them — and what each one is actually evidence of.
Signal classes
Select a class to inspect.
Period 12
78index
vs baseline
+56%
| Period | Value (index) | Baseline (index) |
|---|---|---|
| 1 | 50 | 50 |
| 2 | 51 | 50 |
| 3 | 49 | 50 |
| 4 | 52 | 50 |
| 5 | 54 | 50 |
| 6 | 53 | 50 |
| 7 | 57 | 50 |
| 8 | 60 | 50 |
| 9 | 63 | 50 |
| 10 | 68 | 50 |
| 11 | 72 | 50 |
| 12 | 78 | 50 |
Drift
The baseline moved and nothing announced it
A quantity separates slowly from its own historical trend. No single reading is anomalous, which is exactly why threshold alerting never catches it.
Evidence of: a condition that changed upstream of the metric.
Two halves of the same series, early and late. At the left extreme they sit on one curve; as the period advances the later half separates, which is what drift is — a pattern that was stable and no longer is.
The series below are illustrative of shape only. They are not measurements of any operator and are not claims about outcomes.
All signal classes
- Drift — The baseline moved and nothing announced it
- A quantity separates slowly from its own historical trend. No single reading is anomalous, which is exactly why threshold alerting never catches it. Evidence of: a condition that changed upstream of the metric.
- Separation — One group stopped behaving like the others
- A subset diverges from a population that was previously homogeneous. In aggregate the total looks unchanged, so the aggregate is the thing hiding it. Evidence of: a segment boundary that is now real and was not before.
- Concentration — The distribution has a hot region
- Occurrences cluster far more tightly than a uniform process would produce. The mean is unremarkable; the concentration is the finding. Evidence of: a localised cause, not a general condition.
- Absence — Something that should be there is not
- A signal that has always been present goes quiet. Absence is the hardest class to detect because nothing arrives to trigger a rule. Evidence of: a process that stopped, or a pipe that broke silently.
Contact
If you already have the data, we can show you what it says.
We take on a small number of operators at a time. Engagements start with a read-only audit of what your existing systems already know.