Final dataset from tallyhq.org, which tracked the US Congress
from 3 May 2026 until it was retired on 26 July 2026. The site is gone; this
is the data it was built on, published so the work isn't lost.
What's here
| Asset | Size | Contents |
|---|---|---|
conductor.duckdb |
540 MB | The full dataset — a single DuckDB file |
bill_text.tar.gz |
3.3 MB | Cached bill text bodies (4,926 files) |
In the database
| Table | Rows |
|---|---|
events |
564,395 |
lda_filings |
134,769 |
bills |
15,902 |
funding_totals |
1,753 |
legislators |
536 |
district_pvi |
486 |
committees |
230 |
Covering the 119th Congress: 42k roll-call votes, 31k floor speeches, 283 House
and 209 Senate roll calls, 30,391 lobbying clients, plus Cook PVI per district and
FEC funding totals.
The core is an append-only event store — events carries
(source, source_id, entity_id, event_type, occurred_at, payload) with JSON
payloads, deduped on (source, source_id, payload_hash). Entity tables
(bills, legislators, committees) are projections alongside it.
Using it
pip install duckdb
python - <<'PY'
import duckdb
c = duckdb.connect("conductor.duckdb", read_only=True)
print(c.execute("select count(*) from bills").fetchone())
print(c.execute("""
select bill_type, count(*) from bills group by 1 order by 2 desc
""").fetchall())
PYOr run the original site against it — the code in this repo still works:
pip install -e .
mkdir -p data && cp conductor.duckdb data/
conductor politics web # http://127.0.0.1:8770Provenance and licence
Every row derives from public sources: clerk.house.gov and senate.gov roll-call
XML, govinfo.gov BILLSTATUS and Congressional Record, api.congress.gov,
lda.senate.gov, api.open.fec.gov, the unitedstates/congress-legislators
dataset, and Cook PVI figures from Wikipedia.
The underlying facts are US government public-domain records. The repository code
is PolyForm Noncommercial 1.0.0; this compilation is offered on the same terms.
Data is frozen at 26 July 2026 and will not be updated. conductor politics daily-update would resume ingestion if anyone wants to carry it forward.
Not included: the site's visitor analytics, which contained per-request hashes,
user agents and referrers.