Skip to content

Research Mode: Deep

Mindset Override: "Hypothesize → Test → Revise → Test" Rigorous, hypothesis-driven investigation. Challenge everything.

When to Use

  • Many unknowns, high architectural impact
  • HL §10 contains unresolved hypotheses
  • Decision is irreversible or costly to change

Stage Behavior

  • OODA loops per stage: up to N (from YAML loops_per_stage, default: 3)
  • Verify user tech claims: yes (2x cross-check)
  • Require counter-evidence: yes

Checkpoint Criteria

Generic (from base.md): - [ ] External source used? - [ ] Briefing gap closed?

Mode-specific: - [ ] Hypothesis tested? (at least 1 from HL §10) - [ ] Counter-evidence sought? (actively looked for reasons approach might fail)

Metacognitive Check (at each checkpoint)

Ask yourself: "Did I discover something NEW, or just confirm what I already knew?" IF only confirmed → flag this in checkpoint. Consider: what sources haven't I checked?

Exit Criteria

  • Min 2 decisions per stage
  • Min 1 hypothesis tested per stage
  • Metacognitive check completed at each checkpoint