Skip to content

Rule-Based Reasoning

Layer 2: Thinking
  • Enforces explicit rules [R#] that govern reasoning, keeping outputs aligned with policies, regulations, or domain constraints
  • Creates traceability by requiring references to rules when making claims
  • Finance, security, compliance, medical, or legal work with hard guardrails
  • Anytime you need deterministic decision logic or auditability
  • When coordinating multiple personas who must obey the same rule set
[RULES]
[R1] ...
[R2] ...
[R3] ...
[ANALYSIS]
- Reference rules explicitly (e.g., "Per [R2] we must…").
- Note conflicts between rules and how they are resolved.
[DECISION / OUTPUT]
- Summarize conclusions while citing relevant rules.
- Flag any rule that could not be satisfied and propose mitigations.
[R1] Max single-position = 10%.
[R2] Must maintain ≥30% fixed income.
[R3] New positions require downside stress test.
Analysis: Proposed NVDA increase violates [R1]; alternative is ETF exposure.
Decision: Keep NVDA at 9%, add SOXX per [R2]; run Clara stress test for [R3].
[R1] Prod secrets only in HashiCorp Vault.
[R2] MFA required for admin roles.
[R3] All changes logged to SIEM within 5m.
Analysis: New Jenkins pipeline currently stores secrets in repo; violates [R1].
Decision: Block deployment until Vault integration complete; document timeline for [R3] logging tests.
Pair WithWhen
Chain of ThoughtEach stage cites specific rules
Meta RulesStylistic or tone requirements exist
Clara, Kestra, Ada personasCompliance and rigor dominate