Inspirational Vision Background
The Ideas-to-Life Philosophy

Deterministic Guardrail

Use a non-LLM, rule-based agent to enforce safety constraints with predictable, explainable outcomes.

Confidence: high Pattern ID: deterministic-guardrail exploring
Agentic AI Architecture #agentic-ai #safety #guardrails #heuristics #deterministic #risk-assessment February 15, 2026
Diagram for Deterministic Guardrail

Intent

Enforce safety constraints without relying on probabilistic LLM outputs.

Context

Applied to injury-prevention risk assessment where rules are well-defined (for example, mileage spikes). This pattern fits scenarios where safety is critical and the decision logic is better expressed as explicit thresholds and conditions.

Agentic profile • System shape: unknown • Orchestration mode: unknown

Agent-to-agent interaction • Present: unknown • Mechanism: unknown • Evidence: unknown

Tool protocols • MCP: unknown • Tool calling: unknown • Evidence: unknown

Optimisation target • Primary: unknown • Secondary: unknown • Notes: unknown

Simplicity vs autonomy • Position: unknown • Rationale: unknown

Forces • Safety criticality • Rule clarity • LLM unreliability

Solution • Implement a dedicated agent class that encodes safety heuristics as hard-coded logic. • Produce a risk assessment signal deterministically from input features and reports.

Implementation signals • No LLM client in agent • Rule-based logic (if/else)

Evidence • src/agents/guardrail_agent.py#InjuryFatigueGuardrailAgent — Uses explicit thresholds for mileage increase

Consequences

Benefits • Guaranteed safety check • Explainable logic • Fast execution

Costs • Limited by defined rules • Requires manual rule maintenance

Failure modes • False positives/negatives if heuristics are flawed

Reuse notes • Wrap critical safety checks in non-LLM agents. • Prefer when constraints can be expressed as clear rules and must be predictable.

Confidence

High — the guardrail agent is evidenced as an explicit, heuristic-based component with threshold logic.

Sources & References

  • Formalising the Pattern Publication System