This dataset is released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 licence (CC BY-NC-SA 4.0): research, academic, personal and other non-commercial use is permitted with attribution; adaptations must be shared under the same licence; commercial use is not permitted. The original public source data remains under the Government Open Data License - India (GODL-India). Commercial use: contact surya@judicialmind.ai or hello@judicialmind.ai.

India-District-Briefs

Licence: CC BY-NC-SA 4.0 — research and non-commercial use only; derivatives must be shared under the same licence. Commercial use: contact surya@judicialmind.ai or hello@judicialmind.ai

Access is approved automatically when the email you enter matches your Hugging Face account email and belongs to the organisation you name; other requests are reviewed manually.

Short English and Hindi briefs for 1,588 district-language pairs (794 districts x 2 languages), written by a language model only from computed facts (HMIS health, mandi prices, Kisan Call Centre themes, Jal Jeevan Mission tap water, eShram, PM-KISAN, MGNREGA, NRLM self-help groups, rainfall and NDAP indicators). Every number in a brief is tagged with the id of the fact it comes from and checked against the fact table (number-lock); share of numbers matching a fact: 1.0000. The 552,603 facts ship in facts.parquet with source ids and GODL attribution. Snapshot May 2026.

Quickstart

The repository is gated: accept the terms on the dataset page and log in with huggingface-cli login first.

# pip install pandas pyarrow huggingface_hub
import pandas as pd
from huggingface_hub import snapshot_download

root = snapshot_download("judicialmind/India-District-Briefs", repo_type="dataset")

briefs = pd.read_json(f"{root}/briefs.jsonl", lines=True)
print(briefs.groupby("lang").size().to_dict())
b = briefs[(briefs.district == "Pune") & (briefs.lang == "en")].iloc[0]
print(b.text.split("\n\n## ")[6])          # the eShram section; [fact ids] point into facts.parquet

facts = pd.read_parquet(f"{root}/facts.parquet")
f = facts[(facts.district_lgd == b.district_lgd) & (facts.domain == "eshram")]
print(f[["fact_id", "value", "unit", "period", "rank_in_state", "n_in_state"]].round(2).to_string(index=False))

Output: the brief counts are {'en': 794, 'hi': 794}; the Pune eShram section reads

Workers (eShram)

As of August 2024, 10,76,174 unorganised workers were registered on the eShram portal in Pune district, ranking fifth among 36 districts [eshram.workers_registered]. Of these, 58.74% were women, ranking 11th among 24 districts [eshram.female_share]. The top occupation among registrants was agriculture, accounting for 32.52% of registrations, ranking 30th among 34 districts [eshram.top_occupation_share].

Pune eShram facts behind that section

Fact id	Value	Unit	Period	Rank in state	Districts in state
eshram.workers_registered	1,076,174	persons	2024-08	5	36
eshram.female_share	58.74	%	2024-08	11	24
eshram.top_occupation_share	32.52	%	2024-08	30	34

The same brief is in en/<district_lgd>.md (and the Hindi one in hi/<district_lgd>.md). With datasets: load_dataset("judicialmind/India-District-Briefs", split="train") loads the briefs; read facts.parquet with pandas as above.

Check the number-lock (standard library + pyarrow; works on your own briefs in the same JSON-lines format):

python numlock.py briefs.jsonl --facts facts.parquet
# {"briefs": 1588, "numbers": 96923, "numeric_match_rate": 1.0}

Files

file	content
`briefs.jsonl`	1,588 rows: `district_lgd, district, state, lang` (`en` / `hi`), `text` (Markdown brief), `facts` (the facts the brief was written from), `numlock` (number-lock result at generation)
`en/<district_lgd>.md`, `hi/<district_lgd>.md`	the same briefs as Markdown files, 794 per language
`facts.parquet`	552,603 district facts used for the briefs
`numlock.py`	number-lock checker

Brief sections, in order: Health (HMIS), Mandi prices, Farmer queries (Kisan Call Centre), Census 2011, Schools (UDISE+), Tap water (JJM), Workers (eShram), PM-KISAN, Enterprises (UDYAM), Other indicators (NDAP); a section with no facts for the district says "No data in this snapshot." (Hindi: "इस स्नैपशॉट में डेटा उपलब्ध नहीं है।"). Each figure is followed by its fact id in square brackets.

`facts.parquet` schema

column	meaning
`district_lgd`, `state_lgd`	Local Government Directory district / state code
`domain`	`ndap`, `health`, `mgnrega`, `jjm`, `nrlm`, `rain`, `eshram`, `pmkisan`, `kcc`, `mandi`
`fact_id`	fact id as cited in the briefs (`<domain>.<name>.<period>`)
`label`, `value`, `unit`, `period`	what the fact is, its value, unit and period
`rank_in_state`, `n_in_state`	rank of the district within its state (1 = highest) and the number of districts ranked
`source_uuid`	data.gov.in resource id, `ndap-<id>` or `jjm:<table>`
`attribution`	publisher, title, resource URL and licence of the source

Per-brief facts entries carry fact_id, value, rank_in_state, n_in_state, period, unit; numlock carries n_numbers, n_ok, rate, llm, rewrites, dropped (numbers found, numbers matched, their share, whether the text came from the language model, sentences rewritten and dropped during the check).

Results

check	briefs	numbers	numbers matching a fact
number-lock (`numlock.py briefs.jsonl --facts facts.parquet`)	1,588	96,923	1.0000

A number matches when, after reading Indian formats (1,23,456 / 1.2 lakh / 3.4 crore / Devanagari digits / लाख / करोड़ / %), it equals a fact's value, rank or ranked-district count at the precision the text shows; years and period labels of the facts are allowed.

Coverage

794 districts in 36 states and union territories, English and Hindi. Facts by domain: NDAP 470,881; HMIS health 60,244; MGNREGA 5,095; JJM 4,888; NRLM 3,749; rainfall 2,832; eShram 1,955; PM-KISAN 1,535; Kisan Call Centre 829; mandi prices 595. Snapshot May 2026; each fact's period gives the year or month it refers to.

Source data

Facts are computed from data.gov.in and NDAP resources; each fact names its source in source_uuid and attribution (a data.gov.in resource id resolves at https://api.data.gov.in/resource/<id>; ndap-<id> at https://ndap.niti.gov.in/dataset/<id>). Main resources: Ministry of Health and Family Welfare, HMIS sub-district monthly reports (0c2d45a3-d0b5-4053-bf9a-e00395319472); Agmarknet daily mandi prices (35985678-0d79-46b4-9ed6-6f13308a1d24); Kisan Call Centre query logs (cef25fe2-9231-4128-8aec-2c948fedd43f); Ministry of Labour and Employment, eShram (1d4d1c5a-feea-423b-be4f-c3636fdd1d82); Ministry of Agriculture and Farmers Welfare, PM-KISAN (388208c6-d82a-4190-90df-91aa2c326fec); Ministry of Rural Development, MGNREGA district data at a glance (ee03643a-ee4c-48c2-ac30-9f2ff26ab722) and village-wise SHG counts (d4206736-a28b-4552-8900-7e0c23c707ac); Department of Drinking Water and Sanitation (Ministry of Jal Shakti), Jal Jeevan Mission block tables; NITI Aayog NDAP (57 datasets, including district-wise rainfall, dataset 7319).

Original data: published by the Ministry of Health and Family Welfare (HMIS), Ministry of Agriculture and Farmers Welfare (Agmarknet, Kisan Call Centre, PM-KISAN), Ministry of Labour and Employment (eShram), Ministry of Rural Development (MGNREGA, DAY-NRLM), Ministry of Jal Shakti (Jal Jeevan Mission) and NITI Aayog (NDAP) on the Open Government Data Platform India (data.gov.in) and, where marked, NITI Aayog's National Data and Analytics Platform (NDAP), under the Government Open Data License - India (GODL-India). The original public data remains under GODL-India; the CC BY-NC-SA 4.0 licence of this repository covers JudicialMind's compilation, derived fields, annotations, benchmarks and models.

Citation

@misc{judicialmind2026briefs,
  title={India-District-Briefs: Number-locked English and Hindi district briefs from Indian open government data},
  author={JudicialMind},
  year={2026},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/datasets/judicialmind/India-District-Briefs}}
}

License

CC BY-NC-SA 4.0 (see LICENSE): research, academic, personal and other non-commercial use, with attribution to JudicialMind and to the original data publishers; adaptations must be shared under the same licence. Commercial use requires a separate licence.

Commercial use: contact surya@judicialmind.ai or hello@judicialmind.ai.