Choose the passage structure
The research edition follows the standard 700-verse structure. Chapter 13 contains 34 numbered verses; the alternate opening question by Arjuna is preserved as an unnumbered recensional prelude.
Editorial method
A corpus should show its seams.
The purpose of this release is not to disguise uncertainty. It is to make uncertainty inspectable, reviewable, and correctable.
Keep representations distinct
Devanagari, IAST, and English are stored as separate representations with their own source roles. A change to one does not silently rewrite another.
Cross-check machines before people
Build checks enforce 18 chapters, 700 unique references, continuous numbering, NFC Unicode, Devanagari presence, reproducible exports, and matching SHA-256 hashes.
Review every passage twice
Two independent human decisions are required for Sanskrit, transliteration, English alignment, and provenance. No record can inherit approval from a chapter-level check.
Version every correction
A released file is immutable. Corrections create a new corpus version, regenerate every export, and appear in the changelog.
Important limitation
The English alignment is provisional.
Telang’s 1882 English translation is public domain, but the available digital transcription is chapter-level rather than verse-numbered. This release preserves the machine-assisted verse alignment and its confidence signal instead of presenting it as settled scholarship.
Low-confidence alignments and all other records still require two human review passes before the stable release.
Report a correction
A useful correction identifies the canonical reference, representation, proposed reading, source, and exact locator. Corpus curation remains repository-reviewed; there is no public editing or administration interface in V1.