`committee_reports`

Committee reports

One row per published part of a captured GovInfo committee report package, keyed (package_id, part_id). bill_id is the measure the report accompanies, read from the package's own MODS metadata. All columns are stored as VARCHAR.

Coverage. Sampled. Selected committee report packages from GovInfo, a mixed-Congress selection that grows as packages are added; not complete historical or constituent-granule coverage. committee_report_reads records each package read and its outcome. (measured 2026-09-28)

Data quality. Only the MODS PRIMARY bill fills bill_id. The thirteen CBO columns come from spicy-docs read_cbo_estimate on the same body already acquired. The cover recital gates publication of a letter: a CBO heading alone is not an estimate. report_states_estimate=false records an absent recital; NULL means the rule has not run. A positive recital with no resolved letter span remains distinct from an extracted letter. recital_bill_id comes from the printed recital and is retained beside the MODS bill_id. bill_id is the part's own PRIMARY bill: a multi-part package's root states none, so each part of CRPT-119hrpt494 carries the H.R. 3495 its constituent record names, where its row read before parts were keyed has NULL; its separate recital_bill_id is a text interpretation. Every part a record states is read, all or none: a package whose parts cannot all be read is refused in committee_report_reads and keeps its prior rows. Parts, by decision 29 (docs/research/fork-delivery-decisions-2026-09-22.md): a report published in one part is one row whose part_id is its package id and whose part_number is NULL; a report published as parts is one row per part, part_id the part's granule id (CRPT-119hrpt455-pt2) and part_number 1 to N, except that a Part 1 the publisher left unsuffixed (CRPT-119hrpt494) has part_id equal to the package id and part_number 1. A package whose one part is spelled -pt1 (CRPT-119hrpt811; spicy-docs docs/tables.md names the others it measured) is (package-pt1, 1), the same shape as Part 1 of two, so a reader counts a package's rows to tell a lone part from Part 1 of several. title, date_issued and last_modified are the package's and repeat on every part row. A row published before part_id existed is kept as its package's one part (part_id the package id, part_number NULL) until the package is read again, which replaces its rows as a set; the committee_report_reads rule version re-reads every report once for this, which is what re-keys the CRPT-119hrpt811 row published 2026-09-24 as (CRPT-119hrpt811-pt1, 1).

Parquet file: committee_reports.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
`package_id`	`VARCHAR`	The GovInfo package id: with part_id, this row's identity. Summary facts (title, date_issued, last_modified) are the package's and repeat on every part row.
`collection`	`VARCHAR`	The GovInfo collection the package belongs to (CRPT, CHRG).
`congress`	`VARCHAR`	The numbered Congress, parsed from the package id's own grammar.
`report_type`	`VARCHAR`	The report's document-type code (hrpt, srpt, erpt).
`report_number`	`VARCHAR`	The report's number within its Congress and type.
`chamber`	`VARCHAR`	The chamber, from the package id's document-type code rather than from its first letter.
`title`	`VARCHAR`	The package title as the keyed summary states it.
`date_issued`	`VARCHAR`	The date the package was issued.
`last_modified`	`VARCHAR`	When the publisher last modified the package; the merge prefers the larger value.
`bill_id`	`VARCHAR`	The bill this package concerns, where a linkage exists; nullable by design (C11).
`format`	`VARCHAR`	Which rendition was read (htm, xml, txt, pdf).
`media_type`	`VARCHAR`	The response media type, proved against the format before the body was accepted.
`requested_url`	`VARCHAR`	The URL the body fetch asked for.
`resolved_url`	`VARCHAR`	The URL the body actually came from.
`byte_size`	`VARCHAR`	Length in bytes of the captured body.
`sha256`	`VARCHAR`	Digest of the captured body bytes.
`observed_at`	`VARCHAR`	When the body was captured.
`page_count`	`VARCHAR`	How many pages the extraction read, where a page-based extractor ran.
`text_sha256`	`VARCHAR`	Digest of the extracted text, so a re-extraction that changed nothing is visible as such.
`estimate_rule`	`VARCHAR`	Which rule read this report's cost-estimate statement. Named even though there is one today, for the reason document_citations names its own: a second reader over another rendition would have to say which one produced the span.
`estimate_rule_version`	`VARCHAR`	That rule's version, a digest over all patterns, flags, rejects, heading thresholds and rule revision, so a re-read under a corrected rule is attributable.
`report_states_estimate`	`VARCHAR`	Whether the report's own cover carries the statutory recital `Including cost estimate of the Congressional Budget Office`, printed in brackets. This is the gate and a heading is never one: three retained reports print a CBO heading over a section saying the estimate was not received, and one prints the estimate under a heading no pattern set had. `false` is requested-empty with estimate_absence_reason attached, never absence: 3 of 13 reported bills whose index named an estimate had none in the report.
`recital_bill_id`	`VARCHAR`	The measure the cover states this report accompanies, keyed the way congress_bills.bill_id is, with the Congress taken from this package's own identity because the cover states none. The print's own answer to the report-to-bill join, which is what cbo_cost_estimates.bill_id joins against; bill_id beside it stays whatever index record the caller read.
`estimate_heading`	`VARCHAR`	The section heading the located span opens on, as the committee spelled it, collapsed to one line where GPO wrapped it across two. Committee-specific prose no index states.
`estimate_heading_rule`	`VARCHAR`	Which entry of the heading vocabulary matched. That vocabulary is a floor: it only ever locates a span the recital already declared, and one retained report uses a spelling the routes measurement's five patterns missed.
`letter_span_start`	`VARCHAR`	Character offset of the reprinted letter in the extracted text text_sha256 digests, counted from zero. The letter is not carried in the row, so the span is how it is re-read.
`letter_span_end`	`VARCHAR`	Offset just past the letter, so text[letter_span_start:letter_span_end] is it.
`letter_text_sha256`	`VARCHAR`	Digest of exactly those characters, so a consumer can prove the span it re-read is the span that was measured. A re-extraction that moved one character moves every offset after it, which is why text_sha256 says which text these offsets index into.
`letter_end_rule`	`VARCHAR`	How the end was found: `director_attribution` on the letter's own close, or `next_heading_in_series` where the report prints a summary table with no attribution at all. A declared letter whose end could not be found publishes a NULL span rather than a guessed boundary.
`letter_signatory`	`VARCHAR`	The Director as the letter names them; NULL where the attribution states no name, which the one table-shaped estimate does.
`estimate_absence_reason`	`VARCHAR`	The publisher's own paragraph saying why the estimate is not here, verbatim, where the cover declares none. Carried whole rather than as a span, unlike the letter: it is a paragraph, and a row that holds the words needs no offset to read them.
`estimate_absence_rule`	`VARCHAR`	Which reason pattern matched: `not_available` (the estimate had not arrived when the report was filed) or `not_received` (the committee asked and CBO had not answered). NULL where the report gives no reason, which is not the same as having none to give.
`part_id`	`VARCHAR`	The part of the report this row is: the publisher's own granule id for it, which is also the file stem its body was read at (`CRPT-119hrpt455-pt2`). The package id itself where the part's stem is the package's: a report published in one part, and a Part 1 the publisher left unsuffixed (`CRPT-119hrpt494`). Never NULL, because it is half the identity; a package's part rows are replaced as a set.
`part_number`	`VARCHAR`	The part's number as the publisher states it (its `partNumber`, or the `-pt{N}` its id carries, which must agree). NULL on a report published in one part at the package id, whose record numbers none; a report the publisher issued as a lone `-pt1` is (`{package}-pt1`, 1), so a reader counts rows per package to tell it from Part 1 of several.
`body_completeness`	`VARCHAR`	What the read text states about itself (`sources.govinfo.bodies.publisher_body_status`): `publisher_placeholder` where it is the publisher's own notice that the text is only in the PDF, so text_sha256 digests that notice, not the document; `pdf_extracted` for text extracted from a PDF, including one read in place of a placeholder; `not_flagged` otherwise. No value asserts the text is complete. NULL on a row not re-read since the column was added.
`text_derivation`	`VARCHAR`	The derivation that produced the text text_sha256 digests (`markup-reader`, `text-rendition-cleanup` or `pdf-extraction-gpo-normalized`), from the body sha256 digests; NULL on a row not re-read since the column was added.