W9 corpus pour — Livy's History of Rome joins the served record, all 35 extant books: 1 work, 1,763 episodes, 833,116 tokens

Livy, The History of Rome (Rev. Canon Roberts, Everyman's Library 1912; 1,763 chapter-unit episodes == the canon of the 35 extant books, 833,116 tokens — Rome's own historian joins Polybius on the Republic's shelf). License verified at STAGE 0 (Roberts/Everyman first issue 1912, pre-1930; Wikisource dates the translation 1905). Deterministic extraction profile livy-roberts-v1 with canon count-validation: 1,763 chapters over 35 extant books equal the canon exactly and contiguously (Books 1-10: 60/65/72/61/55/42/42/40/46/47; 21-30: 63/61/49/49/41/51/51/46/38/45; 31-39: 49/40/49/62/51/45/60/60/56; 40-45: 59/28/67/23/46/44), the counts double-sourced from critical editions and a committed boundary witness built from the same pinned Wikisource door. OCR-lost chapter numerals were re-anchored by witness-confirmed boundary repair (similarity >= 0.5) or frozen from manual page reads; three missing print leaves and the scans' OCR holes were found by printed-page continuity and content-search gates, and their 55 victim chapters are served whole from the pinned door rather than truncated (every one named in livy.works.json). Episode membership rides episode.receipt_id = receipt:corpus-w9.

• The served voice is Roberts' 1912 Everyman translation, not Livy's Latin; original_text is NULL (no alignment is fabricated).
• Books 1-10 and 21-38 are served from the shelf scans' OCR text layers, re-ordered from their own geometry — residual character-level OCR damage (including stray CJK glyphs where the OCR engine misread Latin type) is served as extracted, not proofread against the print; per-volume dictionary-word rates against the clean digital baseline (86.9%) are 82.1-86.3% and are recorded in livy.works.json.
• Book 39 (whole), Books 40-45, and 55 cured chapters inside Books 1-5, 9, 26, 28, 30, 31 and 38 are served from pinned Wikisource digital transcriptions of the same Roberts translation, NOT from page scans — the print run-in chapter headings are absent from that door, so those episodes carry none; every cured chapter and both pinned revision sets are named in livy.works.json.
• The print set's own vol V ends at Liv. 39.22 (the Everyman print splits Book 39 across volumes and no scan of vol VI exists anywhere findable) — vol V's chapters 39.1-22 stay unserved on the shelf and Book 39 is served whole from the pinned door so the door seam falls on the 38/39 book boundary.
• Chapter boundaries where the OCR lost the printed numeral were re-anchored by witness-confirmed repair (>= 0.5 similarity to the same translation's opening at the pinned door) or frozen from manual page reads; boundary-audit entries (every repair, freeze, ambiguity resolution and low-similarity flag) are committed in livy.works.json.
• Episodes are chapter-unit (Liv. book.chapter); section-level subdivisions are not carried.
• Books 11-20 are lost in the manuscript tradition itself (only the Periochae survive) — the tradition's fact, not an excerpt decision.
• Episode token counts are whitespace word counts, not linguistic tokens.