alea-institute/alea-gov-xml

ALEA government XML

Tasks over real government XML elements: evaluate an XPath expression, locate the element holding a phrase, read citations and the citation hierarchy, read headings, render XML to text, list defined terms, plus raw XML documents.

75,476 documents, 49,757,939 tokens (alea-v2b-sw32k, the ALEA R4 tokenizer, counting each document's start and end tokens).

Part of the ALEA R4 training mix

This dataset contains exactly the documents of this family that ALEA's R4 model trained on in its stable pretraining mixture (second data stage, one pass each). Every row is byte-identical to the trained text; text_sha256 is the SHA-256 of text. Chat-format rows that the generator reserved for supervised fine-tuning are not included.

Held-out evaluation items generated by the same code are not part of this release.

Composition

task	documents	tokens
xpath	17,301	10,070,036
doc	14,527	10,819,249
citation	13,282	8,467,901
locate	12,226	8,504,903
heading	11,250	6,778,622
render	6,573	4,733,703
terms	317	383,525

phase tag	documents	tokens
midtraining	60,949	38,938,690
pretraining	14,527	10,819,249

How it was made

Written by deterministic code; no language model was used. Units are real elements serialized by lxml: eCFR sections with their ancestor divisions, Federal Register documents (contact blocks removed by tree operation), and bill sections. Federal Register units with an e-mail address, phone number or street address left in the text were dropped entirely (detection only).

Sources

• eCFR bulk XML, Federal Register daily XML 2000–2023, and bills of the 113th–118th Congresses (U.S. government works, public domain); no stratum above 2.2% of tokens

Verification

• Gold comes from XPath evaluated by lxml on the unit, with exact checks on whitespace-normalized text; rendered text must have the same word multiset as the unit.
• Decontamination: training rows that shared a word 13-gram with a held-out evaluation item of this or any sibling family were removed, and rows overlapping ALEA's evaluation sets were removed or logged according to ALEA's decontamination policy.
• Sampled records were reviewed blind by a separate reviewer who recomputed or re-derived the answers; defect patterns found in review were fixed in the generator, or the affected rows dropped, before the final build.

Columns

column	content
id	row id from the generator (not always unique: a few families reuse an id across variants, and a handful of exact duplicate documents were trained twice and appear twice)
text	the document exactly as trained; chat documents use ALEA's chat template, whose control tokens appear as text (`<
task	task kind
phase	the generator's placement tag: pretraining for plain documents, midtraining for instruction-framed tasks; every row here was trained in R4's stable pretraining mixture
generator	what wrote the row (deterministic = code only)
group_id	source passage / record group; all variants of one group share it
prompt_chars	length in characters of the request part of text (0 for plain documents; null where not recorded)
n_tokens	tokens under alea-v2b-sw32k, including the document start and end tokens
text_sha256	SHA-256 of text
meta	JSON: the generator's structured fields (source, presentation, answer form, gold values, checks)

Known limitations

• The documents are generated and more repetitive than natural text.

License

CC-BY-4.0 for this compilation and the generated text. Public-domain and public-record source text keeps its status; Wikidata is CC0; OpenGloss and SHELF are CC-BY-4.0.

Published by the ALEA Institute.