Ipsoot.Data

What We’ve Built / Legislation

Tracking what Congress does in the counties you organize

A civic-engagement program follows 350 counties. It needed one answer: what are the members of Congress who represent those counties actually doing?

By countyfiltered to the counties you organize
Every billyour House members sponsored or cosponsored
Changesevery amendment proposed to those bills
Statuswhere each bill stands, and how it got there

What the program gets

One table the program team can filter by county: every bill the county’s House members sponsored or cosponsored in the current Congress, every amendment proposed to those bills, and each bill’s current status. A separate history shows how each bill’s status changed over time, because “where does this stand now” and “how did it get here” are different questions.

Why it was hard

Answering the question means walking from counties, to the congressional districts covering them, to the House members holding those seats, to every bill they sponsored or cosponsored, to every amendment proposed to those bills. The program had inherited a pipeline that did this, written by someone who had moved on, and it couldn’t finish a run. Congress.gov allows 1,000 requests an hour, the original code assumed five times that, and a full run needs several thousand. Every attempt hit the ceiling and died partway. Worse, each run wiped the tables and started over, so a failed run left less data than before it started.

What we fixed

We rebuilt the collection so that finished work stays finished: it records progress as it goes, resumes exactly where it left off across hourly windows, and running it twice changes nothing. It also stopped asking the same question repeatedly, with one lookup per district instead of per county: roughly 226 calls where it had made 434. Then we fixed the quieter problems, the kind that don’t crash anything:

  • The member list was pulling in senators alongside House members.
  • Amendment sponsors were always blank. The source returns them from a different endpoint than the one being used, so a feature that appeared to work had never worked at all.
  • A filter was case-sensitive, silently dropping records over a capitalization difference.
  • The API key was being written into logged URLs.

Keeping it current

Freshness checks are tuned to the legislative calendar, so a normal congressional recess doesn’t trigger a false alarm, but a genuinely broken feed does. The pipeline reruns safely, resumes after failure, and is covered by 51 automated checks.