Skip to content

Disclaimer

This page tells you, in plain terms, what the data and tools on this site are and aren't. Two main things to know about: AI-generated reports and scraped social-media content from paid monitoring.

The data we publish

Most of what you see on this site is drawn from official sources: Hansard transcripts published by federal and provincial legislatures, the legislatures' own roster pages, official bill texts, official vote records. We re-index this content to make it searchable — we don't claim authorship of it, and we link back to the upstream record where possible.

Accuracy guarantee: as faithful as we can be to the upstream source. When the legislature corrects Hansard, we correct ours. When the official roster changes, we update ours. We don't independently fact-check legislators' statements — if an MP says something in the House that isn't true, our copy of Hansard says that thing. That's how Hansard works.

If you spot a transcription error, a roster mistake, or a vote mis-attribution, submit a correction — we'll fix it. Accepted corrections from signed-in users earn credits.

How politician social-media handles are sourced

Each politician's profile shows the social-media accounts we've linked to them. Those handles come from four kinds of sources, and we tag each with a confidence score so you can tell them apart:

  1. Open North's roster feed — the original handles we ingest with the politician record itself. Most baseline coverage comes from here.
  2. Wikidata SPARQL + OpenParliament.ca + Mastodon directory walks — deterministic enrichers that match politicians to public structured sources by name and jurisdiction.
  3. Operator manual entry — for high-profile politicians where sources 1 and 2 are silent, we add handles by hand with the source tag manual_operator. These are still verified live with the same liveness checker as everything else.
  4. Daily LLM web-research pass — a small autonomous Claude Code session runs each morning, web-searches for handles of the 25 most-active politicians with the worst gaps, and inserts what it finds with an evidence-weighted confidence score (0.3 weak → 0.9 strong). Every row carries the source URL it was found at, so if you're curious where we got a specific handle, that's the audit trail. Source tag: claude-code-agent.

We don't independently verify that a given handle is operated by the politician personally rather than by a staffer — the same caveat that applies to Hansard's recording of who's speaking. If a handle on a politician's profile is wrong, submit a correction and we'll review.

AI-generated reports

The premium report feature uses a large language model to summarise matching speeches into a multi-section synthesis. Specifically:

  • A report is a summary of speeches that match a query, not an independent investigation. The model is shown the relevant speech text from our corpus and asked to organize and summarise it.
  • The model can make mistakes. It can mis-attribute a quote between two MPs who spoke on the same topic, drop important context, or over-confidently summarise something a politician qualified heavily in the original. The cited source speeches are always linked — read them for the authoritative version.
  • Citations point to our copy of Hansard, not the original legislature URL (when we have an upstream URL on file, we surface it too). The speech text underlying any quoted material in a report is in the indexed dataset — you can verify any claim by clicking through.
  • Reports are private to the account that generated them. Other users can't see them, and they don't appear in public search.
  • We don't fine-tune on user content. Your queries and the speech context shown to the model are not retained by the model provider past the response cycle; they're not used for training. See the privacy notice for our model-provider arrangement.

If you publish a report (paste it into a thing, quote it in an article, share it with a colleague), the citation discipline is on you: cross-reference against the linked source speeches. We provide the tool; we don't endorse a specific report as a finished work.

If you've subscribed to monitor a politician's social-media accounts, your dashboard shows posts captured on the cadence you chose (weekly / monthly / quarterly). Some things to know:

  • All captured content comes from public accounts. We don't access private accounts, DMs, locked tweets, friends-only Facebook posts, or anything not accessible to anyone with a web browser. If a politician has a public X account, that's the data we capture.
  • It's a snapshot, not a stream. A scrape captures what was visible at the time it ran. If a post was deleted between scrapes, we may have the deleted copy — see takedown for the policy on this.
  • Engagement counts are point-in-time. The "238 likes" we recorded is what the post had when we scraped it; it may have grown since.
  • Coverage is partial. Politicians who don't have public accounts on a platform won't show data. Some accounts use display-name changes, deletions, or restrictions that affect what we can see.
  • Translation and tone are preserved. We don't summarise or rewrite. The text in our database is the text the politician posted, with HTML stripped and engagement metadata attached.
  • Visibility is public on the politician's profile page. Posts captured by paying subscribers appear in the Recent posts tab of the politician's profile, visible to anyone (including anonymous visitors). Subscriber identity is anonymous by default: attribution reads "Scraped via paid monitoring." Subscribers can opt in to public attribution when they configure their monitor, in which case the post line reads "Funded by @theirhandle." The opt-in is per-subscription and defaults off.

Fair use and editorial framing

We publish this data to support journalism, academic research, civic literacy, and the public's right to scrutinise the people who govern them. The framing across the site is broadly pro-transparency and pro-democratic-accountability — we don't pretend to be neutral on the proposition that voters should be able to easily see what their representatives have said and done.

We aren't an investigative outfit. We don't editorialise on specific politicians. We do publish blog posts and analysis from time to time; those are signed and clearly marked.

Use of any data exported from the site:

  • Permitted — research, journalism, civic-tech apps, academic work, classroom use, personal curiosity, building something useful.
  • Permitted with attribution — incorporation into derivative works, visualisations, publications. Cite as "Canadian Political Data (canadianpoliticaldata.org)" or link.
  • Not permitted — using the scraped content in ways that violate the upstream platforms' terms (e.g. for spam, harassment, or bulk-republication as if you'd scraped it yourself).
  • Not permitted — using the data to identify, target, or harass private individuals incidentally mentioned in legislative business.

Future changes (read this if you care)

A few things are in active development that may affect the framing above:

  • Aggregate views on profile pages. We may, in a future release, surface aggregate statistics (post counts over time, posting cadence, topic clustering) on the public politician profile page alongside the individual captured posts. The same subscriber-anonymity default would apply.
  • More platforms (TikTok, Threads, Facebook page metadata) may come online. The same scope and discipline as above will apply.

Who to contact

admin@thebunkerops.ca for anything about this page. For specific content concerns, the takedown and DSAR pages have the detailed workflow.

This page itself is versioned in this site's documentation; you can see the history of when it changed by looking at the file in the documentation repository.