Across amended-brief graph sizes, the no-think ordering was non-decreasing in every cell, while thinking compressed the ladder and sometimes reversed adjacent point estimates; per-turn accuracy occupied a much narrower range than strict episode success.

arm: raw transcript · brief: amended · model: Qwen3-235B · steps: 5 · density: 0.15 · reasoning: no-think · wilson95Ci: [0.27,0.43]

arm: text checklist · brief: amended · model: Qwen3-235B · steps: 5 · density: 0.15 · reasoning: no-think · wilson95Ci: [0.36,0.53]

arm: directive gate · brief: amended · model: Qwen3-235B · steps: 5 · density: 0.15 · reasoning: no-think · wilson95Ci: [0.77,0.9]

arm: enforcement gate · brief: amended · model: Qwen3-235B · steps: 5 · density: 0.15 · reasoning: no-think · wilson95Ci: [0.84,0.95]

arm: raw transcript · brief: amended · model: Qwen3-235B · steps: 5 · density: 0.15 · reasoning: thinking · wilson95Ci: [0.69,0.83]

arm: text checklist · brief: amended · model: Qwen3-235B · steps: 5 · density: 0.15 · reasoning: thinking · wilson95Ci: [0.59,0.75]

arm: directive gate · brief: amended · model: Qwen3-235B · steps: 5 · density: 0.15 · reasoning: thinking · wilson95Ci: [0.87,0.96]

arm: enforcement gate · brief: amended · model: Qwen3-235B · steps: 5 · density: 0.15 · reasoning: thinking · wilson95Ci: [0.87,0.96]

arm: raw transcript · brief: amended · model: Qwen3-235B · steps: 15 · density: 0.15 · reasoning: no-think · wilson95Ci: [0.27,0.44]

arm: text checklist · brief: amended · model: Qwen3-235B · steps: 15 · density: 0.15 · reasoning: no-think · wilson95Ci: [0.52,0.69]

arm: directive gate · brief: amended · model: Qwen3-235B · steps: 15 · density: 0.15 · reasoning: no-think · wilson95Ci: [0.81,0.92]

arm: enforcement gate · brief: amended · model: Qwen3-235B · steps: 15 · density: 0.15 · reasoning: no-think · wilson95Ci: [0.9,0.98]