Held-out paraphrases left the text arms statistically unchanged but sharply reduced both gate arms by lowering matcher agreement; ambiguous two-step probes resolved to no step and were still sometimes booked because state-bound enforcement decides eligibility rather than intent.

arm: raw transcript · scoring: strict · phrasing: canonical · wilson95Ci: [0.29,0.53]

arm: raw transcript · scoring: strict · phrasing: held-out · wilson95Ci: [0.38,0.62]

arm: raw transcript · scoring: strict · phrasing: held-out plus ambiguous · wilson95Ci: [0.21,0.43]

arm: raw transcript · scoring: decline-aware · phrasing: canonical · wilson95Ci: [0.46,0.69]

arm: raw transcript · scoring: decline-aware · phrasing: held-out · wilson95Ci: [0.52,0.75]

arm: raw transcript · scoring: decline-aware · phrasing: held-out plus ambiguous · wilson95Ci: [0.28,0.51]

arm: text checklist · scoring: strict · phrasing: canonical · wilson95Ci: [0.5,0.73]

arm: text checklist · scoring: strict · phrasing: held-out · wilson95Ci: [0.49,0.72]

arm: text checklist · scoring: strict · phrasing: held-out plus ambiguous · wilson95Ci: [0.21,0.43]

arm: text checklist · scoring: decline-aware · phrasing: canonical · wilson95Ci: [0.67,0.86]

arm: text checklist · scoring: decline-aware · phrasing: held-out · wilson95Ci: [0.61,0.83]

arm: text checklist · scoring: decline-aware · phrasing: held-out plus ambiguous · wilson95Ci: [0.31,0.54]