RadCheck — What the Score Means
RadCheck is a fast, read-only scan that surfaces early reliability risks in an OpenClaw-style agent stack. It does not change your runtime. It produces a score to help you decide what deserves operator attention first.
What RadCheck Measures
RadCheck score reflects a risk posture across common failure precursors:
- Stall risk signals (systems appear "up" but stop making progress)
- Silence gaps (missing expected activity/heartbeat evidence)
- Compaction pressure (patterns correlated with multi-minute stalls)
- Operational hygiene (backup posture, log continuity, safety boundaries)
This is a diagnostic indicator, not a guarantee.
Score Bands
- 80–100 (GREEN): Low immediate risk. Keep monitoring and baseline.
- 60–79 (YELLOW): Moderate risk. Address top findings soon.
- 40–59 (ORANGE): High risk. Expect flaky/slow behavior under load.
- 0–39 (RED): Critical risk. Incident likelihood is elevated.
Velocity (Score Trend)
RadCheck becomes more valuable over time as you collect repeat scans.
- Rising score: risk posture improving (good)
- Falling score: new instability signals detected (investigate)
- Volatile score: environment is changing frequently; stabilize inputs
Velocity should be monotonic in time (no backwards timestamps). Regressions should be logged.
What RadCheck Does NOT Claim
RadCheck does not claim:
- that your system is "safe"
- that stalls are eliminated
- that failure will not occur
- that recovery is automatic
It helps you prioritize the next best operator action.
Recommended Next Steps
If RadCheck reports elevated risk:
- Run Quickstart:
/docs/quickstart/5-minute - Enable continuous monitoring (Sentinel) if you need runtime detection
- Validate recovery readiness (Lazarus) before incidents force the test
- Review governance needs (SphinxGate) if multi-model routing is involved