What this site is, and how far to trust it

A site built from other sites

Most of what you read here was first published by someone else: Hansard and recorded votes from the House of Commons, by way of OpenParliament.ca; bill records from LEGISinfo; court decisions from the courts, by way of the Access to Algorithmic Justice project at York University; spending and results from the Treasury Board’s GC InfoBase; contracts and travel from Open Government Canada; reports from the Parliamentary Budget Officer; news articles and headlines from Canadian news organizations; polls as compiled on Wikipedia; biographical details from Wikidata and Wikipedia; and video from CPAC and other YouTube channels.

The site links each figure, quote and record to where it came from wherever it can. Those sources remain the authority. Where this site and a source disagree, the source is the one to rely on, and we would like to hear about it.

Information may not be accurate

Sources make mistakes, change their records and go out of date, and collecting from them adds its own errors: a name matched to the wrong person, a bill matched to the wrong session, a page that changed after it was read. Coverage is uneven. Some courts, years and records are not yet included, and some pages say plainly what is missing. Nothing here should be treated as complete, current to the minute, or free of error.

Where the site uses language models

Some pages assert things that no source states outright, because a language model read the record and drew a conclusion. These include:

• the written account of each sitting, and its verdict;
• ratings of answers in Question Period;
• members’ and columnists’ policy positions;
• readings of bills (how clearly drafted, how much they changed, whether they worked);
• gaffe readings;
• checks of claims made in television clips;
• profiles of influence groups;
• answers in Ask the record.

The site limits what models may do: quotes are checked word for word against the original, citations are kept only when the search behind them returned that source, and each reading shows the passages it rests on. Even so, a model can misread tone, miss context or reach a conclusion a careful person would not. Treat these readings as an informed interpretation to be checked, not as fact.

How the AI-rated figures are made

Question Period answers. A language model reads each exchange in Hansard and rates whether the answer addressed the question: answered, partly answered, deflected, or answered something else. The share of straight answers on the front page is the count of exchanges rated “answered” divided by every rated exchange on file.

Sitting scores. Each sitting is scored out of 100: 35 points for decisions (recorded votes and bills that moved a stage, counted from the record, not from the model), 35 for how fully Question Period questions were answered as the model rated them, 15 for the share of the day spent on business rather than procedure, and 15 for civility as the model rated it. A sitting with no rated Question Period is scored on the rest. Whether a sitting “decided nothing” is counted from the stored votes and bill stages alone.

The site’s own scores and measures

Sitting scores, department effectiveness scores, the lenses and the relationship maps are measurements the site designed. Each page states its method, and the word lists and weights are published so they can be argued with. A different method would give a different answer. A high or low score is a reading against that method, not a verdict on a person’s character, competence or motives.

Estimates

Some figures are estimates, labelled as such. The legal cost of laws in court, for instance, is built from published fees and budgets, because no government reports that spending law by law. Estimates are shown with ranges and their inputs. They indicate scale, not exact amounts.

Opinion and commentary

Television clips on the site are commentary by their own hosts and guests; their inclusion is not endorsement. Sections marked as opinion, such as a suggested course correction, are the site’s own view and say so.

Independence and use

Parliament AI is an independent civic project. It is not affiliated with or endorsed by the House of Commons, the Senate, the Government of Canada, any court, any political party or any news organization. Names, quotations and logos belong to their owners and appear for reporting and commentary. Nothing on the site is legal, financial or professional advice.

Corrections

If something here is wrong, please say so. Send a correction with a link to the source that shows it, and it will be checked and fixed. Because the site rebuilds itself from its sources, many errors are corrected at the source and flow through on their own.