← Research Portal
AAL-F-007Provisional
Arithmetic ceiling on AAL-D-002: second-model confirmation withdrawn
Provisional (1 model on AAL-D-002)
Evidence level
76.1%
GPT-4o amount accuracy
99.9%
GPT-4o detection
AAL-F-006 established the arithmetic ceiling for GPT-4o across two datasets. This finding extended it to Claude Sonnet 4.6 on AAL-D-002 using AAL-RS-008. Those figures were withdrawn on Oct 12, 2026 because the results file could not be located, so this finding currently rests on GPT-4o alone and is marked provisional.
Correction (Oct 12, 2026)
Claude Sonnet 4.6 figures (amount accuracy, detection, and the combined false-positive count across both models) were removed with the withdrawal of AAL-RS-008. GPT-4o figures are from AAL-RS-007.