Platform
Decide How Much Reasoning A Request Deserves.
- Context Length↑
12 docs · long context raises depth need
- Task Type↑
Cross-document comparison, not classify
- Semantic Ambiguity→
Clause language moderately ambiguous
- Required Reasoning↑
Renewal risk needs multi-hop judgment
- Tool Need↓
No live writes; extraction only
- Failure Cost↑
Customer-facing financial interpretation
- Latency Budget→
Interactive ok under 4s
- Output Constraint↑
Structured clause list required
Routing decision
Deep Reasoning → Structured Extraction → Quality Check
- Fast Path excluded — cross-doc + high failure cost
- Tool Agent not required — no live system writes
- Escalate if quality gate fails factuality or completeness