`bill_versions`

Bill printings

One row per printing of a bill, per source that supplied it. The full text is deliberately not a column here: bill_sections.body carries it at the grain people query, and the body is re-fetchable by package id and digest. All columns are stored as VARCHAR.

Coverage. Sampled: bills the bill family has read from GovInfo BILLSTATUS, Congress by Congress from the 108th on; group by the Congress in bill_id to see which Congresses are filled. source is govinfo for a printing whose body was acquired; other rows are printings the publisher lists without an acquired body. (measured 2026-09-28)

Data quality. A numbered reprint is its own printing since spicy-docs 0.37.0: the publisher types it exactly like the printing it follows, so its version_code is its own package suffix (eas2, rfs2) rather than the stage's slug. Public/private laws are distinct publisher listings, not enrolled BILLS printings. Unsupported law bodies remain unacquired: their offered URLs, labels and dates are retained, with no acquired body or section claims.

Parquet file: bill_versions.parquet
MCP query_sql support: Configured; requires an available artifact.
Publication status: Not established by this schema page or its measurement date.
Row count: Not stated here; the MCP describe_table reply gives the live count under publication.

Column	Type	Description
`bill_id`	`VARCHAR`	The bill this printing belongs to.
`version_code`	`VARCHAR`	The printing's code: its stage's sealed slug, or a numbered reprint's own package suffix (eas2).
`source`	`VARCHAR`	Which acquisition path supplied this row (govinfo, congress, govinfo-pdf, upload).
`label`	`VARCHAR`	The publisher's version-type string verbatim, which is not unique per printing.
`version_date`	`VARCHAR`	The publisher's date for this printing; the merge prefers the larger value. An enrolled printing's date is empty at the publisher (BILLSTATUS `<date/>` and the printing's own dc:date), so version_date is '' (not NULL) on enrolled rows, which sorts before every date: order printings with printing_order or NULLIF(version_date, '').
`package_id`	`VARCHAR`	The GovInfo BILLS package id, read from a stated format URL rather than derived by name.
`format_name`	`VARCHAR`	Which rendition was read (xml, txt, pdf, html, uslm), as choose_format named it.
`format_type`	`VARCHAR`	The publisher's own format type string on the chosen link, where it states one.
`requested_url`	`VARCHAR`	The URL the fetch asked for.
`resolved_url`	`VARCHAR`	The URL the response actually came from, which a redirect can change.
`content_type`	`VARCHAR`	The response Content-Type exactly as the publisher sent it.
`byte_size`	`VARCHAR`	Length in bytes of the captured body, before any decoding.
`sha256`	`VARCHAR`	Digest of the captured body bytes, spelled sha256: plus the hex digest.
`observed_at`	`VARCHAR`	When the body was captured; the instant the other capture columns describe.
`offered_formats_json`	`VARCHAR`	Every format link the publisher offered for this printing, as a JSON array.
`version_code_is_reprint_ambiguous`	`VARCHAR`	True when the publisher's version-type string also names a numbered reprint, so the slug alone cannot tell the two printings apart.
`root_tag`	`VARCHAR`	The XML root element name (bill, resolution, amendment-doc).
`body_tags`	`VARCHAR`	The body element names found, unit-separator joined; a reported bill states two.
`publisher_stage`	`VARCHAR`	The root's own bill-stage or resolution-stage attribute: the publisher's word for it.
`section_count`	`VARCHAR`	How many content-bearing nodes the flattening produced; the bill_sections row count.
`discarded_elements_json`	`VARCHAR`	Count by element name of every element whose text survives nowhere in the flattened nodes.
`equivalent_xml_version_code`	`VARCHAR`	For a PDF twin, the version_code of the XML row it stands in for.
`equivalent_xml_source`	`VARCHAR`	That twin's source. A version code is not unique across acquisition paths, which is why bill_versions keys on source too, so a twin reference needs both halves to resolve.
`cleanup_line_numbers`	`VARCHAR`	GPO PDF normalization: whether per-line gutter numbering was detected and stripped.
`cleanup_gpo_footers`	`VARCHAR`	GPO PDF normalization: whether VerDate print-metadata footers were stripped.
`cleanup_spacing_normalized`	`VARCHAR`	GPO PDF normalization: the collapse rule always runs, kept for parity.
`cleanup_small_caps_merges`	`VARCHAR`	GPO PDF normalization: how many lone-uppercase-letter lines were merged.
`cleanup_hyphen_rejoins`	`VARCHAR`	GPO PDF normalization: how many gutter-corroborated hyphen wraps were rejoined.
`cleanup_json`	`VARCHAR`	The per-page GPO cleanup breakdown, including which evidence gated bare-digit stripping.
`kind`	`VARCHAR`	Interpreted document kind: `full_text`, `kind_uncertain` (a full-text slug whose document is too thin to be sure), `procedural_amendments`, `procedural_summary`, or `unknown`.
`kind_rule`	`VARCHAR`	Which version-kind rule fired, as the classifier names them: `procedural_amendments_slug`, `procedural_summary_slug`, `full_text_slug`, `full_text_slug_thin` (a full-text slug the size heuristic demoted to kind_uncertain), `amendment_substring` (an unlisted slug this repository's own heuristic caught), `size_heuristic` (an unlisted slug a large body promoted to full_text), or `unknown` when nothing claimed it.
`kind_label`	`VARCHAR`	The display label for this kind, supplied by the caller that owns the vocabulary.
`kind_warning`	`VARCHAR`	The inline warning this kind carries, for a version picker. NULL for `full_text` and for `unknown`, which states none of its own.
`kind_section_count`	`VARCHAR`	The section count the kind heuristic was given, recorded so the answer re-derives.
`kind_body_bytes`	`VARCHAR`	The body byte length the kind heuristic was given, recorded for the same reason.