Automated Neuro-Symbolic Epigraphic Extraction and Secular Earth Rotation (Delta-T) Constraints from Medieval Indian Eclipse Inscriptions (522–1584 CE)

Stone and copper-plate donation inscriptions (dana-shasanas) across the Indian subcontinent record over a millennium of dated solar and lunar eclipses alongside village-level topocentric coordinates and five-fold traditional Panchanga calendrical parameters (Saka year, 60-year Jovian Samvatsara, lunar month, paksha/tithi, and weekday vara). However, large-scale geophysical exploitation of digitized epigraphic corpora (Epigraphia Carnatica, Epigraphia Indica, and South Indian Inscriptions) has been bottlenecked by severe 19th-century OCR diacritic corruption and scribal gate (expired) versus vartamana (current) era-year ambiguities. We introduce a 3-Agent Neuro-Symbolic Epigraphic Pipeline combining high-recall fuzzy lexical triage (Levenshtein <= 1), LLM schema extraction with persistent caching and deterministic rule fallback, 60-year Jovian Samvatsara calendrical disambiguation, and topocentric NASA JPL DE440s/DE441 ephemeris verification. In a blind calibration audit against 50 ground-truth inscriptions from Vahia et al. (2013), our pipeline achieves 98.0% detection recall, 96.0% (48/50) exact Julian date recovery without manual OCR repair, and a 0.0% false positive rate. Scaling to an open-corpus sweep across 19 digitized archival volumes (1,195 candidate eclipse chunks), our pipeline extracts and astronomically verifies 29 additional dated medieval eclipse inscriptions (100.0% verification yield, 676 CE–1584 CE). Sweeping Earth's clock-drift parameter Delta-T = TT - UT1 across the combined 77 verified topocentric visibility corridors from 522 CE to 1584 CE yields a secular quadratic curvature of c2 = 29.24 +/- 0.63 s/cy^2, providing an independent low-latitude geophysical constraint complementing the 32.0 s/cy^2 canonical spline of Stephenson, Morrison & Hohenkerk (2016).