Public codebook · Definition 1.2.0

How the ledger turns records into claims

The site preserves the difference between source observations, reviewed resolutions, derived metrics, and unknowns. This guide explains what each public result establishes—and what it does not.

Start here

Four interpretation principles

Evidence before interpretation

Every defensible claim identifies its unit, source, observation window, denominator, uncertainty state, release, and route back to evidence. A scheduled event, a documented proceeding, an individual roster entry, and a named vote are different facts.

Missing is not absent

A member is described as present, absent, excused, remote, or partially present only when an identified source supports that state. A missing roster, unavailable artifact, unresolved identity, or unreviewed record remains unknown.

Activity is not productivity

Votes, committee records, sponsorships, actions, publications, and reviewed speech are separate observable evidence channels. None measures total work, effort, effectiveness, intent, constituent service, policy impact, or personal worth. The project publishes multiple lenses and no composite score.

Source disagreements are data

Conflicting calendars, journals, minutes, media, and roster assertions remain independently inspectable. Reconciliation may select an analytical interpretation, but never erases the assertions that produced it.

What one row represents

Units of analysis

Find the corresponding release tables
UnitIdentityAppropriate use
Chamber sittingStable sitting ID, chamber, session, dateSchedule, documented occurrence, quorum, journals, roll calls, media
Member–sittingMember service and sittingPrinted attendance and named vote evidence
Committee meetingCanonical committee proceedingNotice, agenda, minutes, motions, documents, media
Member–meetingMember service and meetingSource-scoped attendance and named participation
Member–dayMember service and calendar dateCross-source observable-activity context
MeasureSession, chamber, type, identifierSponsorship, action, committee, and vote relationships
Source observationSource object and acquired artifactProvenance, conflicts, processing, and missingness
Media assetPlatform and external identityDuration, proceeding linkage, transcript, and review evidence

Words have controlled meanings

Evidence and missingness states

Documented present
An identified official source explicitly lists the member present for its stated scope.
At least one recorded vote
A named yea or nay exists. This is positive participation evidence, not proof of presence throughout the sitting.
Documented absent or excused
The exact source supports that label. The project does not infer a reason.
Not individually recorded
The proceeding may be documented, but the collected source prints no usable status for this person.
Source unavailable
An expected source could not be obtained. This is different from a source that was obtained but had no relevant statement.
Processing incomplete
The artifact exists, but acquisition, parsing, transcription, linkage, or review is unfinished.
Conflicting official records
Two identified assertions disagree. Both remain visible; neither silently becomes truth.
Undetermined
The available evidence does not support a preferred factual interpretation.
Excluded
The observation is outside the defined opportunity set—for example, before service or outside an assignment interval.

The denominator comes first

Eligibility and observation windows

Member comparisons use versioned windows shared by every selected member. Eligibility accounts for service start and end dates, chamber, committee assignment intervals, and the event date. Proceedings outside service or assignment are excluded rather than counted as missed opportunities.

Every denominator-based metric is partitioned into included, known but not included, unknown, and excluded. These counts are mutually interpretable; unknown is never folded into the negative class.

Raw totals are contextual

Offices, leadership roles, committee assignments, source practices, and opportunity counts differ. A larger raw count is not automatically more, better, or more effective work.

Definition 1.2.0

Published member metrics

Inspect a complete example

Metric labels and rules are release-versioned. A later definition may correct or refine eligibility; comparisons must not silently mix versions.

Proceeding outcome, not personal attendance

Quorum classification

A sitting is quorum confirmed only when the collected record supports establishment of quorum. It is confirmed no quorum only when explicit proceeding language establishes failure. Otherwise it remains undetermined.

The classifier preserves source text, method, schedule state, journal status, roll-call counts, and evidence coverage. Recording duration can prioritize review, but duration alone never determines quorum. A no-quorum outcome does not identify who was present or absent.

Inspect monthly denominators and every classified sitting.

Source scope stays attached

Committee meetings, attendance, and votes

A notice can establish scheduling; minutes can establish documented occurrence; a roster can support a member status; a motion or named choice supports that discrete action. These records are related but not interchangeable.

Canonical committees retain source aliases and chamber identities. Joint bodies preserve the contributing source chamber. Participation rows retain the document or roster scope and do not claim continuous presence.

Browse committees, meetings, documents, motions, and recordings.

Search aid, not official quotation

Media, duration, and transcripts

Media records preserve platform identity, official links, date, duration, acquisition state, proceeding linkage, and transcript provenance. Duration may come from official metadata, media inspection, or transcript metadata and is useful for triage—not as independent evidence of attendance, quorum, or work.

Transcript text, punctuation, timestamps, and speaker labels may be machine-generated. Consequential wording must be checked against the recording. A model label such as SPEAKER_04 is not a confirmed person identity. Full transcript redistribution remains subject to the source-rights policy.

Search transcript passages with timestamped recording links.

Do not repair disagreement invisibly

Identity, totals, and source reconciliation

Stable public identifiers connect records across changing source systems. Original labels, source IDs, acquisition metadata, and aliases remain available. A canonical link means the project resolved an analytical identity; it does not assert a legal rename or erase the source record.

When a vote provides both aggregate totals and named choices, the site compares them. A mismatch is displayed as a discrepancy. Neither side is silently edited to force agreement.

Release boundaries

Coverage and known limitations

Inspect all 44 measured surfaces

The normalized release spans multiple years and source families, but comparable percentage coverage rows currently measure only 2025–2026 House surfaces. Missing Senate or earlier-biennium rows mean unmeasured, not zero.

  • PUBLIC RECORD BOUNDARY: No public observation on a date is not evidence that no legislative work occurred.
  • TRANSCRIPT AUTOMATED: Published transcript search text is automated and may contain recognition or speaker-attribution errors.
  • COVERAGE LEDGER SCOPE: The structured source-coverage ledger currently measures 2025-2026 House surfaces only. It must not be generalized to earlier sessions or the Senate.
  • SOURCE RIGHTS PENDING: Source access does not itself grant republication rights. Consult source_register before reuse; all source families still require legal review.

Public records do not comprehensively capture private constituent work, informal negotiation, caucus activity, preparation, travel, or every public event. Therefore the dataset supports claims about observable evidence—not total work.

From page to frozen evidence

Reproduce and cite a result

  1. Preserve the complete page or API URL, including filters and observation window.
  2. Record release 2026-07-28.3, schema 1.9.0, and data-through date 2026-07-27.
  3. Inspect the denominator, unknown, excluded, coverage, and definition fields.
  4. Follow the project evidence link to the official source and exact locator.
  5. Use the immutable release-pinned API or downloaded artifact and verify its SHA-256.
  6. Cite the project record, dataset release, and official source as three distinct layers.