section_diffs

Bill version comparisons

One row per compared pair of consecutive printings of one bill. engine_name, engine_version and engine_revision record which diff engine produced the row. All columns are stored as VARCHAR.

Coverage. Sampled: consecutive printings of a bill are compared only when both bodies were acquired and parsed, from the Congresses the bill family has read. (measured 2026-09-28)

Data quality. A consecutive pair that cannot be diffed — because one or both sides has no parsed tree — is a counted refusal named by table, never a row. Measured on the cold-start run of the 119th (receipt d1-measured-run-2026-09-19/): 3,185 refusals against 128 published comparisons, which is what a 600-printing cap over 22,064 printings implies, since a pair needs both of its sides fetched in the same run. A pair runs from a printing to its neighbour in spicy-docs' printing_order: publisher date, then a dateless enrolled printing by its stage (pair_rule consecutive_by_date_then_stage). A pair's neighbour can be a procedural printing (a four-section Senate engrossed amendment between the calendar printing and the enrolled bill), which the summary route refuses but this table compares, so the pair reads as nearly everything removed and then everything added. In section_diff_items, financial_amounts_changed is true on every added or removed row that carries a figure, by construction; only a modified row says a figure changed in place. Floor-amendment annotations ((increased by $X)) are left out of financial_*_amounts_json though they change the text; financial_has_amendment_annotations marks them. Enrolled printings carry new element ids at the publisher, so element_id does not join into enr. All columns are stored as VARCHAR.

• Parquet file: section_diffs.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 both printings belong to.
from_version_code	VARCHAR	The earlier printing's version code.
from_source	VARCHAR	Which acquisition path supplied the earlier printing.
to_version_code	VARCHAR	The later printing's version code.
to_source	VARCHAR	Which acquisition path supplied the later printing.
from_version_date	VARCHAR	The earlier printing's publisher date.
to_version_date	VARCHAR	The later printing's publisher date; the merge prefers the larger value.
pair_type	VARCHAR	Which comparison strategy this pair called for: xml-xml, pdf-pdf or pdf-xml.
pair_rule	VARCHAR	Why these two printings were compared: neighbours by date, then by stage, never every pair.
added_count	VARCHAR	Sections present only in the later printing.
removed_count	VARCHAR	Sections present only in the earlier printing.
modified_count	VARCHAR	Sections present in both whose body text differs.
unchanged_count	VARCHAR	Sections present in both whose body text is identical.
moved_count	VARCHAR	Sections the move round paired at a different path.
item_count	VARCHAR	How many section_diff_items rows this diff produced.
engine_name	VARCHAR	Which diff engine produced this row; the stand-in for a rule name.
engine_version	VARCHAR	The engine's declared version.
engine_revision	VARCHAR	The exact pinned commit the engine was installed from.
computed_at	VARCHAR	When the comparison ran, from the caller's injected clock.