The Sentinel
Engine telemetry is precise but opaque. A gamma headroom of 0.05, a criticality of 0.75, and a steps-to-breach of 3.0 are exact. They do not tell an operator what is happening, how fast it is happening, or what to do about it.
Sentinel, the Substrate telemetry summarizer, translates these numbers into narratives. Kairos Substrate is the on-premise KAIROS deployment.
What Sentinel Does
Section titled “What Sentinel Does”When the evaluation pipeline produces an escalation, Sentinel receives the engine telemetry. This structured feed holds recent tick data, warning signals, and escalation context. Sentinel produces three things:
A narrative. A plain-language summary of the system’s state and trajectory.
The following template output uses a constructed example feed with gamma 0.25, floor 0.20, headroom 0.25, and 2.0 steps to breach.
This example has a three-tick criticality trend from 0.00 to 0.45 and no loss events. Its overall risk is ELEVATED.
Human escalation was triggered with gamma at 0.25 (floor: 0.20, headroom: 0.25). The system predicts 2.0 steps until floor breach. Warning criticality has risen from 0.00 to 0.45 over the last 3 ticks, indicating a deteriorating trajectory. Floor breach is predicted within 2 ticks.Highlighted metrics. Selected telemetry values with individual severities. The number of highlights depends on the feed.
Examples from this feed include Gamma state: 0.250 (floor: 0.200), classified as INFO.
Steps to breach: 2.0 is classified as ELEVATED.
Criticality trend: 0.00 → 0.45 (3 ticks) is classified as CRITICAL. These examples are not an exhaustive highlight list.
A risk level. A single classification (NOMINAL, ELEVATED, or CRITICAL) that summarizes the overall urgency.
Risk Levels
Section titled “Risk Levels”| Level | Meaning | Operator response |
|---|---|---|
| NOMINAL | Escalation may be precautionary | Review the precautionary context |
| ELEVATED | Trajectory shows a concerning trend | Monitor the trend |
| CRITICAL | Immediate human attention required | Review the escalation immediately |
The overall risk level and each highlight’s severity are separate outputs. A CRITICAL highlight does not by itself determine the overall risk level.
The serialized levels are NOMINAL, ELEVATED, and CRITICAL. The stored-summary TUI prints Risk: Nominal, Risk: Elevated, or Risk: Critical.
Highlight Severities
Section titled “Highlight Severities”Each highlighted metric carries its own severity classification:
| Severity | Meaning |
|---|---|
| INFO | Informational context, no action needed |
| ELEVATED | Worth monitoring, the trend may worsen |
| CRITICAL | Requires immediate attention |
Deployment Behavior
Section titled “Deployment Behavior”Sentinel runs inside the Substrate deployment. Its structured feed contains engine telemetry and trusted escalation context. The default template backend generates operator-facing summaries locally, without outbound network access.
Integration Points
Section titled “Integration Points”Sentinel is used in two places:
The TUI dashboard. When kairos observe is launched with --sentinel, the dashboard includes a Sentinel panel.
A selected operator-review request with a stored summary shows its narrative, highlights, and risk level.
The panel title is Sentinel [TEMPLATE] or Sentinel [SLM], according to the stored summary’s backend.
Without a stored summary, the Sentinel Assessment panel shows fallback metrics rather than a generated narrative.
The stored-summary panel uses color for highlight severity. The summary data carries the explicit severity labels.
Operator review. When a workflow requires human review, the Sentinel assessment can appear alongside the raw evaluation data. This gives operators both quantitative and narrative context for their decision.