`section_classifications`

Section classifications (model)

One row per label a model assigned to one section of one printing. vocabulary_hash is a digest over the sealed label list, so a vocabulary change is visible in the data rather than silently reinterpreting old rows. All columns are stored as VARCHAR.

Coverage. Empty by owner decision 34: written only by a language model during bill-family runs, which no scheduled run enables, so no row is produced. (measured 2026-10-03)

Parquet file: section_classifications.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 whose section was classified.
`version_code`	`VARCHAR`	The printing whose section was classified.
`source`	`VARCHAR`	Which acquisition path supplied that printing; without it an XML row and its PDF twin collide.
`seq`	`VARCHAR`	The classified section's bill_sections seq, its key within the printing.
`match_path`	`VARCHAR`	The section's normalized cross-version key, unit-separator joined.
`body_index`	`VARCHAR`	Which body element the section came from.
`label`	`VARCHAR`	One of the five sealed classification labels.
`confidence`	`VARCHAR`	The model's own confidence, checked to be between 0 and 1 before it was stored.
`model`	`VARCHAR`	The model id the call was made against.
`prompt_version`	`VARCHAR`	The prompt version the label was produced under.
`prompt_hash`	`VARCHAR`	Digest of the exact prompt sent, spelled sha256: plus the hex digest, so a prompt edit is visible per row.
`batch_index`	`VARCHAR`	Which batch of the run this section belonged to.
`requested_at`	`VARCHAR`	When the call was made.
`completed_at`	`VARCHAR`	When the answer came back; the merge prefers the larger value.
`vocabulary_hash`	`VARCHAR`	Digest over the sealed label table, so a vocabulary change is visible in the data.