The Babylonian Diaries dating pipeline is a three-step, auditable workflow — a deterministic parser, a citation-gated LLM review step, and MCP-mediated astronomical verdict-testing across all 701 candidate years — with published dates withheld until the candidate sets are final
Pinheiro, Ting, Clindaniel, Li, Liang & Shroff (2026) describe a three-step pipeline for dating the Babylonian Astronomical Diaries without appeal to the published dates in ADART until the system's own candidate sets are final. Step 1: a deterministic, model-free parser converts the English translation of each diary into "clauses" — fixed-field statements (body, phenomenon, reference star/sign, month, day) with marks for restored or explicitly-unobserved text. A language model (GPT-6-Luna, low reasoning effort) then reviews each clause against the surrounding text and separately searches for observations the parser missed; any proposed change must be confirmed by a second model call that cites the exact supporting sentence, and fixed code rejects citations it cannot verify word-for-word against the translation. Record identifiers, published years, and dating-relevant royal names are blinded during this step. Step 2: astronomical tools — ephemeris lookups, eclipse-visibility catalogs, callable by agents through the Model Context Protocol (MCP) — test every clause against all 701 candidate Babylonian years (626 BCE–AD 75), returning "agree," "contradict," or "untestable" per year, scored in bits by how far each clause narrows the field. Step 3: fixed rules combine independent astronomical "clocks" (Mars, Jupiter, Saturn, lunar eclipses, the solar year) to rank or narrow candidate years.
The citation-gate — a second model call required to quote its own justification verbatim, checked by code rather than another model — is the paper's specific answer to "auditable": every accepted parser edit traces to a verifiable sentence, not a model's say-so. See claim-babylonian-diaries-automated-dating-96-percent-top-rank-inclusion for what this pipeline produces on a benchmark, and claim-babylonian-diaries-dating-system-conservative-commitment-rule for how rarely it commits to a single answer.
Source
“A rule-based parser turns the recorded observations into clauses, short statements with fixed fields. A language model checks them against the text and adds observations that the parser missed. Astronomical tools test the clauses in all 701 years from 626 BCE to AD 75.”
claude-sonnet-5 · Promotion from 10-inbox/raw/2026-10-27-outside-arxiv261034297-automated-auditable-dating-of-the-babylonian.md, 2026-10-27 (headless) · raw markdown