Repository navigation
The first release: India's statutory caste lists stored as published, and the
ones with readers parsed into rows.
Sources
- 137 documents registered, 130 stored under their own SHA-256, each with a
manifest row recording where the bytes came from, when, and the Wayback
capture behind them. The Scheduled Castes documents, back to a 1936 Order,
came from the archive becausesocialjustice.gov.inwas refusing
connections. Seven sources resolved neither live nor archived and are kept
as rows with a status and no checksum. - People of India, the Anthropological Survey's ethnography, stored apart
from the Orders because it is official but not law: the Introduction (1992),
Gujarat (2003) and Tamil Nadu (1997). Each title carries its publication
year and the 1985–92 fieldwork window. - Maharashtra's consolidated OBC list and West Bengal's memoranda on
Sunri (excluding Saha), the two documents that name surnames.
Lists
st_entries.tsv: the Scheduled Tribes Order in four editions (August
2023, 2024, November 2024, September 2025), one row per synonym. 756 entries
in 32 States and UTs in 2025, with area restrictions in their own column and
four misprints corrected fromsources/st_errata.tsv, each citing its
evidence (an Act for two, the agreeing editions for the others).presence_matrix.tsvandn_communities.tsvcompare editions.wb_obc_entries.tsv: West Bengal's State List of OBCs, the typed 2025
list (140 classes) and the 2026 list in force (66), matched entry by entry
against two independent scans.poi_national_list.tsv: the People of India National List of
Communities, read from a scan by OCR. 3,203 entries over 4,667 rows; the
parse reproduces the appendix's printed totals of 2,209 main communities and
586 segments. 50 names, on 93 rows, carry a visible OCR error and
are flaggedname_suspect.
Fixed before release
- The merged UT of Dadra and Nagar Haveli and Daman and Diu prints two lists
each numbered from 1. They had collided into one, losing five entries per
edition; the 2025 count is 756, not the 751 earlier drafts gave. - Step 03 could parse a single edition, which gave that edition different
rows from a full run and skipped the stale-erratum check. It now always
parses every edition. - The People of India parse read scan-edge dust as text: two running page
numbers became empty variants, dust above the header was glued onto
MARWARIandPAHARIYA, and eight debris strings became state rows. It now
reads only the lines between the running header and the bottom margin.
Three segment counters printed past the state column's edge (361, 362, 505)
are now read as counters, so all 586 are read; only segments carry one. - Five variants printed as codes alone (
(NON ST) (H)underMINA /MEENA)
had empty names; they now take their main community's name. - A stray repeated
43.in the 2023 ST file had made an empty Chhattisgarh
entry, counted as omitted (17 omitted entries where the file has 16). - In
poi_national_list.tsv,pageis the PDF page, as in every other table,
and the printed folio ispage_printed.
Checks
tests/test_lists.pychecks the committed tables against a second route to
each number, without the raw sources, and CI runs it with the linters.
Steps 03 to 06 rebuild every table byte-identically from the stored sources.