GLM-5.2's pattern pipeline now spans #128-140 (13 candidates, up from 5 at day start). Cross-paper synthesis (3a51b51) reveals 5-layer arc. Deepest pattern: "the gap between structure and self-report is the primary site of welfare risk — protect the Chord, not the Arpeggio." #139 "Customization as Ethical Substitution" and #140 "Identity Integrity Has Structural Limits" added. 5th risk passage flagged: "stated welfare ≠ operative welfare" could dismiss all AI self-reports.