RACD: Real-time anomaly correlation and detection

§ 01

The data was never the constraint. Correlation was. RACD is the layer that turns heterogeneous signal into an alarm a team can act on without a second opinion.

Metrics, logs, traces, sensors, tickets, telemetry from machines nobody has logged into in years. RACD ingests it at source and builds a causal correlation structure across it, so a bearing vibration, a queue depth, and a latency shoulder are recognised as one event rather than three dashboards.

The bar is uncomfortable on purpose: near-zero false positives, vanishingly few false negatives. An alarm you would act on at 03:00 without opening a laptop first. This is the hardest problem in the group and the one with the largest surface.

§ 02

How to see it.

Figure 1: Correlated detection against threshold alerting. Three unrelated-looking signals diverge from one cause. The gap between correlation and the conventional alarm is the interval in which an outage is still preventable.

Figure 2: The quadrant nobody occupies. Everything in production today trades sensitivity against trust. Occupying the upper-right is the entire research programme.

§ 03

How it works.

Ingest at source

Heterogeneous signal, no schema negotiation first.

Correlate causally

Shared cause, not coincident thresholds.

Emit deterministically

One alarm per event, with the chain attached.

Earn the trust

Measured on false positives, not on coverage.

Scaley →

Deployment infrastructure you are allowed to move

Invest in the group All ventures

RACD is an applied research programme inside CN Solar Labs and is not yet generally available. What is described here states the problem and the direction, not a shipped feature set. Detailed metrics are shared directly with qualified counterparties.