What are my odds?

Every review you enter is collapsed into one number on the venue's own rating scale — the effective mean. Your raw mean is the starting point; confidence-weighting, disagreement between reviewers, reviewer count and the rubric dimensions each nudge it by a fraction of a rating point. The probability is then read off that venue-year's curve. The breakdown above shows every nudge.

• Extra features barely matter. On ICLR, everything beyond the mean rating is worth about +0.004 AUC. They are here because they are slightly better than nothing, not because they are the point.
• The interval is not a guarantee. It covers sampling uncertainty in the historical curve (and, for recovered years, the transferred selection-sharpness parameter). It does not cover this year differing from last year, area chair discretion, or your paper being unusual.
• Withdrawals are excluded. The denominator is papers that reached a decision, not all submissions — because you have scores and are not withdrawing. That is why the base rate shown is higher than the acceptance rate the venue publishes.
• Rating scales are not comparable between years. ICLR 2026 uses {0,2,4,6,8,10} where 2025 used {1,3,5,6,8,10}; NeurIPS collapsed to 1–6 in 2025. Each year has its own model. Pick the right one.
• Nothing here knows what your paper is about. A calculator that reads scores cannot tell a borderline paper an AC will champion from one they won't.

Built from papercopilot/paperlists, iclr-insights, and Figure 1 of the NeurIPS 2025 program chairs' review-process post.