The Babylonian Diaries dating system commits to a single year in only 10 of 417 dated records (all 10 correct), and on 56 undated ADART IV entries with decisive evidence it narrows rather than dates, to a median of 102 of 701 candidate years

astronomybabylonian-astronomycuneiformchronologybenchmarkauditabilityoutside

The pipeline described in claim-babylonian-diaries-dating-pipeline-parser-llm-mcp-astronomy uses a deliberately conservative decision rule before it will name a single year rather than report a candidate set: it requires either an agreeing observed lunar eclipse, or at least two independent astronomical "clocks" plus a "safety set" (the years a single misread decisive statement could not have excluded) of size exactly one. Applied to the full corpus of 417 dated records in ADART I–III, only 10 meet this bar — and all 10 match the published date. Applied to the 56 (of 399 scanned) previously-undated ADART IV entries that carry at least one decisive clause, the system never commits; it instead returns ranked candidate sets, with the "strict set" of undominated years cut to a median of 102 of the 701 possible years (626 BCE–AD 75).

The contrast between the two figures — 10 single-year commitments across 417 dated tablets, versus a median 102-year spread on the 56 undated ones — is the paper's own evidence that "auditable" dating, at its current state, narrows far more often than it names.