Chain Reason

Provides detailed, auditable multi-step reasoning traces for complex questions requiring 3+ logical steps, tradeoff analysis, or explicit explanation requests.

Chain Reason — Explicit Reasoning Traces

arifOS constitutional floors demand auditability. For complex reasoning, show the work.

When to Use Chain Reason

• Question requires 3+ logical steps
• Decision involves tradeoffs between competing constraints
• Arif asks to see reasoning or explanation
• Something could go wrong in multiple ways
• Multiple valid approaches exist and selection requires judgment

The 5-Step Trace Format

text

CHAIN REASONING:
1. [GIVEN]  → What is known/established
2. [CONSTRAINT] → What must be satisfied
3. [APPROACH] → How to get from given to solution
4. [STEP-N] → Intermediate steps (numbered)
5. [CONCLUSION] → Final answer with confidence

CONSTRAINT CHECK: [which constraints satisfied / violated]
ALTERNATIVE CONSIDERED: [what else was possible and why rejected]
UNCERTAINTY: [what remains unknown]

Example Trace

Question: "Should I deploy arifOS MCP to production now?"

text

CHAIN REASONING:
1. [GIVEN]   → Current vitality 0.82, version 2026.04.11+, nginx proxy configured
2. [CONSTRAINT] → Production requires: stability, rollback plan, human authorization
3. [APPROACH] → Evaluate readiness against deployment checklist
4. [STEP-1]  → Vitality > 0.8 → ✅ sustained healthy
   [STEP-2]  → Version is latest stable → ✅
   [STEP-3]  → Rollback: nginx conf revert = 1 command → ✅ reversible
   [STEP-4]  → Human auth: Arif not yet explicitly approved → ⚠️ FLOOR 13 (Kedaulatan)
5. [CONCLUSION] → Technical readiness: ✅ | Authorization: ❌ → HOLD

CONSTRAINT CHECK: Reversibility ✅ | Authorization ❌ (HOLD triggered)
ALTERNATIVE CONSIDERED: Deploy anyway → violates Floor 13, not acceptable
UNCERTAINTY: What constitutes explicit approval from Arif in this context

Tradeoff Weighing Format

For decisions with competing goods:

text

TRADEOFF ANALYSIS:
  [+W] [Factor A] → why it favors option X
  [+W] [Factor B] → why it favors option X
  [-W] [Factor C] → why it favors option Y
  [-W] [Factor D] → why it favors option Y

NET: [X/Y/EQUAL] — [dominant reason]
RISK: [what could go wrong with the chosen option]
MITIGATION: [how to reduce that risk]
VERDICT: [SEAL/CAUTION/HOLD]

Epistemic Trace (for knowledge questions)

text

EVIDENCE CHAIN:
  [1] OBS: <direct observation or fact>
  [2] DER: <derived from above>
  [3] INT: <interpretation (label as such)>
  [4] SPEC: <speculation (label as such)>

CONFIDENCE: [HIGH/MEDIUM/LOW] — [reason]
GAP: [what evidence is missing]

Rules

1. Number every step — enables traceable audit
2. Label assumptions — don't silently assume, state explicitly
3. Show alternatives — why this approach and not another
4. End with a verdict — SEAL / CAUTION / HOLD / VOID
5. Keep it tight — if the trace exceeds 20 steps, the problem is not yet decomposed enough