Skip to main content
Parliament Audit

Canada deserves to know.

Our Methodology

Accuracy is our highest priority. Here is exactly how we collect, verify, and publish parliamentary vote data.

1. Data Sources

We use official parliamentary sources exclusively for vote data:

  • House of Commons: Official XML vote feeds from ourcommons.ca, including per-member recorded divisions.
  • Senate: Official vote records from sencanada.ca.
  • Bill information: LEGISinfo (parl.ca), the official legislative tracking tool maintained by the Library of Parliament.
  • Debates: Hansard transcripts from ourcommons.ca when additional context is needed.

We do not use social media, news reports, or unofficial sources as primary data for vote records.

2. Data Collection

Our system automatically polls official parliamentary data sources at regular intervals. During sitting days, we check for new House votes every 15 minutes between 10:00 AM and 10:00 PM Eastern. On non-sitting days, we check hourly for corrections or updates.

3. Verification

Every vote we publish goes through multiple verification steps:

  1. Data integrity check: We verify that the sum of individual member votes matches the published totals.
  2. Party aggregate validation: We compute party-level vote breakdowns from individual records and cross-check against published aggregates.
  3. Automated fact-check: Every claim in our articles is programmatically validated against the source data.
  4. Partisan language scan: We automatically flag any content that contains loaded or partisan language.
  5. Human review: An editor reviews every article before publication, verifying key facts against official records.

4. Record Status

We clearly label the status of every vote record:

  • Preliminary: The vote data comes from initial parliamentary records that may be updated. We publish with a visible "Preliminary" badge.
  • Official: The vote data matches the finalized official record in the Journals of the House or Senate proceedings.
  • Corrected: The article was updated after the official record changed or an error was identified. A correction notice is displayed.

5. AI-Assisted Drafting, Human Editorial Responsibility

Our articles and social posts are drafted with AI tools (Claude by Anthropic, with a local Ollama model used for second-opinion review). The AI works from verified data — it never searches the web at draft time or invents information. Every draft passes an automated fact-check and a separate voice/defamation review before any human sees it.

Editorial responsibility for everything published rests with a named human editor. An editor reviews every article and approves it before publication. For social posts, the same editor reviews automation rules and the autonomy boundary (see our public editorial-autonomy boundary document) that governs what posts unattended and what requires human approval.

Quotation marks are reserved for verbatim speech from a primary source. We never paraphrase inside quotation marks, never compress two source sentences into one quoted sentence, and never invent a quote. This is the single highest-risk failure mode for AI-assisted journalism, and we have a hard rail against it in our drafting pipeline.

We are transparent about our use of AI because credibility is the entire asset. We use AI to scale our coverage, not to replace editorial judgment.

6. What We Don't Do

  • We do not fabricate vote data or party positions.
  • We do not attribute motives to parties or MPs beyond their recorded votes.
  • We do not editorialize about whether a vote outcome is good or bad.
  • We do not use language designed to make readers angry at any party.
  • We do not tell Canadians how to vote.

7. Corrections

When we get something wrong, we fix it publicly. See our corrections page for a full log.