anusuchi

Statutory caste lists for India, stored as published and parsed into rows.

Anusuchi (अनुसूची) is the Schedule of the Constitution — the one that gives Anusuchit Jati and Anusuchit Janjati their names.

A land record, an electoral roll or a survey records a jati: ପାଣ, कुर्मी, Bagata. Almost every question worth asking needs a category — Scheduled Caste, Scheduled Tribe, Other Backward Class, or none of them. The mapping between the two is law, published in gazetted Orders, and it is not in any of those datasets. This repo holds the law and the mapping, so that analysis does not have to invent either.

Three layers, kept apart

layer	what it is	changes when	who judges
sources/	gazette PDFs and pages exactly as fetched	the government republishes	nobody — bytes
lists/	one row per gazetted entry	sources/ changes	nobody — mechanical parse
crosswalks/	one row per dataset label → entry	a dataset is added, or a call is revised	a person, on the record

lists/ is shared by every state and every downstream repo. crosswalks/ is per dataset and is the only place judgment lives. Mixing them is how a legal list quietly acquires someone's opinion.

What this does and does not license

The chain has two links, and they do not have the same standing.

jati → category is official. The gazette settles it. That is what this repo holds.

surname → jati is irreducibly statistical. No Order names a surname, and no amount of gazette work changes that. In Odisha's land records, of 2,052 surnames carried by 100 or more people, 67 (3.3%) belong to a single jati, covering 0.3% of tenants; the median surname spans 12 jatis and the widest spans 421. Behera has 153,016 bearers across 421 jatis, its commonest at 14%.

Nothing here may be described as an official mapping from a name to a caste.

Two legal facts that change the arithmetic

Scheduled Caste status is conditioned on religion. Paragraph 3 of the Constitution (Scheduled Castes) Order, 1950:

no person who professes a religion different from the Hindu, the Sikh or the Buddhist religion shall be deemed to be a member of a Scheduled Caste

Scheduled Tribe status is not. The ST Order carries no equivalent clause.

So in Odisha's own labels, ପାଣ ଖ୍ରୀଷ୍ଟିୟାନ — Pana Christian, 40,729 tenants — is legally not SC, while କନ୍ଧ ଖ୍ରୀଷ୍ଟିୟାନ — Kandha Christian, 30,121 — is still ST. A crosswalk that reads only the caste half of the label gets one of those wrong.

Many entries are scheduled only in named places. 34 of 772 ST entries carry an area restriction: Gujarat's Kunbi only in the Dangs district, Andhra's Dhulia, Paiko, Putiya only in Vishakhapatnam and Vijayanagaram. Category is therefore a function of (label, district), not of label alone, which is why area_restriction is its own column.

Pipeline

Numbered, re-runnable, each step writing only its own layer.

step	writes
scripts/10_fetch_sources.py	sources/raw/<sha256>.<ext>, sources/manifest.tsv
scripts/20_parse_st_list.py	lists/st_entries.tsv, lists/st_parse_report.tsv

Step 10 fetches every URL in sources/registry.tsv, records the newest Wayback capture and submits one where none exists, stores bytes under their own SHA-256, and writes a manifest row per artefact. A source that fails is a row with a status and no checksum, not a silence. Nothing is ever overwritten: a republished document is a new file and a new row beside the old one.

This is not ceremony. socialjustice.gov.in was refusing connections while this was written, and every Scheduled Castes document here — the consolidated Order and the amendment trail back to a 1936 Order predating the Constitution — was recovered from the Wayback Machine rather than the live site.

Step 20 reads the ST Order's two-column layout by word position, splitting on the gutter found per page. 772 entries across 32 States and UTs — every State and UT that has Scheduled Tribes yields at least one. It recovers the Order's own synonym lists (Bhottada, Dhotada, Bhotra, Bhatra, Bhattara, Bhotora, Bhatara is one tribe, Odisha entry 5), splits the 34 area restrictions into their own column, keeps section headings as context rather than emitting them as phantom tribes, and picks up the entries the Order prints without a number.

What the source gets wrong, and how that is handled

The Order is a scanned-and-retyped legal document and it has typos. Each is recorded rather than corrected in place: state carries the canonical name and state_as_printed carries what the page actually says.

• The Order misspells Uttarakhand as Uttrakhand, and spells Jaunsari as Jannsari under that heading while spelling it correctly under Uttar Pradesh. Matching only the correct spelling lost the whole State — its five tribes were in the document the entire time.
• A heading can wrap. Dadra and Nagar Haveli and Daman / and Diu is one UT split by the column width.
• Not every entry is numbered. Puducherry's sole tribe and Lakshadweep's descriptive entry are bare text.
• Some lines are sections, not tribes. Assam's list splits into I. In the autonomous Districts of... and II. In the State of Assam...; Arunachal opens All tribes in the State including:. These are kept in a section column as context for the entries beneath them.
• Puducherry's entry trails into editorial boilerplate (Note: In case of any discrepancies...), which is stripped.

Known gaps

• SC, OBC and PVTG lists are fetched but not yet parsed. Only the ST Order has a reader.
• 6 of 74 registered sources resolved neither live nor in the archive. They are manifest rows with a status and no checksum; four are amendment Acts and two are administrative circulars.
• tribal.nic.in does not present a complete certificate chain; verification is disabled for that host alone and the manifest records tls_verified=false on the affected rows.

Not yet built

crosswalks/ is empty. The first will map Odisha's Record-of-Rights jati labels onto these entries, using the status / method / evidence vocabulary proven in land/crosswalk/caste_crosswalk_additions.tsv: every row carries how it was decided and a sentence saying why, and nothing is silently dropped.