What the record contains
Exact saved passages and the individual vote on the official question. Recorded material is not automatically spoken floor debate. Inserted statements are labeled; third-party text is excluded.
Starting the interactive map
You can explore the records while the map loads.+ United States Explore a state. Follow the impact.
LOCAL PILOTFictional people & records. No findings about real politicians.
Follow the evidence
Evidence, not a verdict
Why the records relate
Explicit measure references first. Debate context and narrowly defined policy connections need supporting evidence. Similar sentiment alone never establishes the same proposition.
What can be concluded
Every assessment requires a facts pass and a later contrary-context pass. Conditions, procedural votes, changed text and uncertain chronology stay visible. No honesty score. No inference of motive.
What this dataset covers
A fixed fictional corpus covering direct references, agreement, tension, procedural actions, conditions, absence, uncertain same-day order, missing text and no-match outcomes. Not a sample of actual congressional behavior.
- Action window
- 2026-08-01 — 2026-08-31
- Statement lookback
- 30 days
- Sources / passages / actions
- 20 / 9 / 8
- Last source event
- 2026-08-14
- Published release
- 2026-09-15 · sd-be69cc2ae6dbe004
Exclusions and gaps
An example with missing operative text remains in the private review queue.
1 unmatched candidate; 2 candidates with publication blockers. “No match in this dataset” never means “did nothing.”
Local processing, deliberate publication
Input files → archived source versions → validated claims → candidates → review decisions → immutable public snapshots. Browsing never calls a model or government API. Unreviewed candidates and raw model output stay on disk, outside the public folder.
The demo uses transparent deterministic extraction. External AI extraction files can propose meaning, stance and sentiment, but cannot approve or publish. Sentiment is not voting intent.
The nine synthetic evaluation cases are regression tests, not a 95% real-world accuracy claim. Real source verification, held-out evaluation and physical-phone performance measurements are still required before a real evidence release.