Skip to content

Reports and credits

A report is an AI-assisted, evidence-cited summary of what parliamentarians have said in the chamber, with every claim linking back to a specific speech chunk so you can verify it in one click.

There are three kinds of report you can ask for today, each priced proportionally to how much material the system reads:

  • Full report on a politician + topic (the original) — generated from a politician's profile or from a specific search result. Reads every matching speech for that politician and produces a multi-section brief.
  • Synthesize this search — generated from the Analysis tab or Timeline view of /search. Reads the top N matches across all speakers in the result set and produces a one-paragraph summary plus five bullet findings, each citing source quotes. Useful when you want to know what the corpus collectively says about a topic, not what one politician said.
  • Stance map for this search — same trigger surface as Synthesize. Groups speakers in the result set by stance (for / against / conditional), with one exemplar quote per group. Useful for journalists building a "who's on which side" frame.

Reports cost credits, which you buy in packs (or earn by submitting accepted corrections).

Buy credits See your credit balance

In this section

Why credits, not subscriptions

We charge per report, not per month. This means:

  • Casual users pay for what they use, with no recurring bill.
  • Heavy users pay in proportion to load — running a report has a real compute cost, and we'd rather price it transparently than hide it behind tiered plans.
  • There's nothing to cancel — credits don't expire.

What a report contains

The exact shape depends on the kind, but every report includes:

  • A plain-language summary of what the underlying speeches collectively say on the topic.
  • Direct quotes, with each quote linking back to the originating speech.
  • Citations on every claim — pulled from the actual chamber transcript.

A full report on a politician also flags tensions or contradictions across time. A synthesize-this-search opens with a stats table (party / top speakers / time range) before the bullet findings, and a stance map structures the body as for / against / conditional groups with an exemplar quote per speaker.

What a report does not contain:

  • The model's opinions about the politician.
  • Speculation about why a politician holds a position.
  • Claims that aren't grounded in something the politician actually said in Hansard.

Limits

  • One report runs at a time per account (queued; you'll get an email when each is ready).
  • Free-tier accounts have a soft cap on reports per day to keep abuse costs predictable. The cap is generous; if you hit it for legitimate research, ask for an increase from your account page.
  • Reports are stored on your account and visible only to you. They are not indexed by the public site.

What if a report is wrong

If a report misrepresents what a politician said:

  1. Tell us — there's a feedback link on every report.
  2. You may be eligible for a refund if the issue is a bug on our side (e.g. wrong politician's speeches got included). Refunds add credits back to your balance.
  3. We treat report-quality complaints as a learning signal — if a class of question consistently produces bad reports, we adjust the prompt or the retrieval logic.

We won't refund a report just because the conclusion isn't what you wanted. Reports describe what the politician has said; that's not the same as what you wish they'd said.