The ICML review process creates no obligation for reviewers to explain late-stage score changes — and that absence is the institutional choice the post makes visible. The confidence drop from 4 to 2 arrived without notification, without a stated rationale, and within a window where authors can no longer respond. The community's familiarity with the pattern — confirmed by the thread's traction — shows this is not a gap in the rules but a feature of them. ICML's own program chairs acknowledged AI-assisted review as a growing integrity concern in…