Template Backend

The template backend is Sentinel’s default narrative generator. It uses deterministic rules to produce structured assessments from telemetry data — no ML models, no external dependencies, sub-millisecond latency.

How It Works

The template backend processes the SentinelTelemetryFeed through a rule-based assessment pipeline:

1. Gamma analysis — computes headroom, trend direction, and proximity to floor
2. Warning trend analysis — examines the historical window for escalating patterns
3. Steps-to-impact assessment — evaluates urgency from the escalation context
4. Risk level classification — combines signals into NOMINAL, ELEVATED, or CRITICAL
5. Narrative assembly — generates a human-readable summary from structured templates
6. Highlight extraction — selects the most important metrics for the operator

Risk Classification Rules

The template backend classifies risk based on these signals:

Signal	NOMINAL	ELEVATED	CRITICAL
Gamma headroom	> 25%	5–25%	< 5%
Warning trend	Stable or decreasing	Increasing	Rapidly increasing
Steps to breach	> 10	3–10	< 3
Loss event in window	No	No	Yes

Multiple signals are combined — a CRITICAL from any single signal elevates the overall assessment to at least ELEVATED.

Narrative Generation

The template backend constructs narratives from pre-defined sentence templates, parameterized with actual telemetry values:

• Opening — overall stability assessment based on risk level
• Gamma state — current gamma, floor, and headroom percentage
• Warning trend — describes the trajectory over the historical window
• Urgency — steps-to-breach and escalation type context
• Recommendation — action guidance based on the escalation type

Example output at CRITICAL:

System stability is deteriorating rapidly. Gamma has dropped to 0.250 against a floor of 0.200, leaving only 5.0% headroom. Warning criticality has escalated from 0.0 to 0.75 over the last 3 ticks, with impact projected in 3.0 steps. Immediate operator review recommended.

Example output at NOMINAL:

System stability is within normal parameters. Gamma is at 0.650 with comfortable headroom above the 0.200 floor. This escalation appears precautionary — no immediate action required.

Characteristics

Property	Value
Determinism	Fully deterministic — same input always produces same output
Latency	Sub-millisecond (typically < 1ms)
Dependencies	None beyond kairos-engine-core types
Feature flag	Always available (no feature gate)
Backend identifier	template in SentinelSummary.backend

When to Use Template

The template backend is the right choice when:

• Determinism is required — you need reproducible assessments for audit trails
• Latency is critical — sub-millisecond response time
• No model infrastructure — you don’t want to manage model bundles
• Deployment simplicity — zero additional dependencies

For richer, more contextual narratives, see the SLM backend.