ztzoff.tech

Project

PledgeScope

Every member of Congress, what they promised next to what the record shows, with the source behind every number. The model reads, code verifies, people decide.

539 member profiles · 1,025 campaign sites preserved · a language model that is never allowed to decide

GeminiFirebaseCloud RunTypeScriptOpenTelemetry

What it does

PledgeScope puts what each member of Congress promised next to what the public record shows. One page per member: roll-call votes, sponsored and cosponsored bills, campaign money from FEC filings, dated press releases, and exact quotes from the member's own campaign site as it was archived before election day. Each quote carries a status against the record and the records behind it. No scores, rankings or ratings.

Readers start with a ZIP code, which is looked up in the browser from a static table and never sent anywhere, and end at a source. Every number links to Congress.gov, the House Clerk, senate.gov or the FEC. If it cannot be sourced, it is left out.

The product booklet

Ten pages on the method: the member profile, the pipeline, the five promise statuses, the guardrails, and the agents that keep the record current.

How a promise is checked

  • Code, deterministic and tested. Fetches the campaign page as the Internet Archive saved it before election day and stores the bytes with a hash. Checks every returned quote against that text; anything not found verbatim is thrown out. Publishes only through a database that refuses without a recorded approval.
  • Model, one bounded call. Gemini reads the page text and returns up to twelve exact quotes, each with a neutral summary and a topic. Later it gets a closed list of the member's bills and votes dated after the promise and says which relate. It never chooses what to read and has no write path.
  • Person, signed in and on the record. An editor accepts or rejects each suggested record, picks a status, writes a rationale and attaches evidence. Every transition is logged under the editor's identity.

Five statuses, none of them about the person

Kept, Working on it, No action found, Contradicted, Reversed position. Trying and failing counts as keeping your word: a bill that dies in committee is still a bill the member put their name to. "No action found" is not a verdict; it says what was searched. There is no "liar" or "hypocrite" status, because those describe a person. These five describe a promise against a record.

Guardrails that hold even if the prompt is wrong

  • A quote that is not in the preserved page, character for character after whitespace, is rejected and logged.
  • Every source page and API response is stored as retrieved, with a SHA-256 hash and timestamp. Database triggers refuse edits and deletes.
  • A row cannot reach the published state without a recorded human approval and at least one https evidence link. Application code cannot bypass it.
  • Assessments are append-only. A new one adds a version; the previous one stays.
  • Public data carries no model name, editor identity, internal note or stored page copy.
  • For any promise an editor can export an evidence pack: the quote, the preserved page with its stored and recomputed hash, a fresh quote check, the review history and every assessment.

Automation

More than twenty scheduled workers keep the record current: votes hourly, Senate roll calls daily, campaign finance and press releases on their own cadences. None of them can publish. Everything they produce lands in a review queue or in the immutable evidence store.

Why this matters for your project

Most agent projects fail on the boundary between what the model proposes and what the system commits. PledgeScope is a worked example of drawing that line in the database, not in the prompt: the model has no write path, the verification is code, and the approval is a logged human act. If your agent touches records that someone will be held to, this is the shape we would start from.

Have a bounded AI system to build?

30 minutes. Problem, owner, budget, date, and success criteria.

Book a call
Book a discovery call