Skip to content

Generating a report

There are three trigger surfaces for paid analyses, all priced up-front and all credit-refunded if the worker fails. Pick the one that matches the question you're trying to answer.

Full report on a politician

Use this when the question is what does this person say about X?

  1. Navigate to the politician whose record you want to summarise. From a search result, click through to their profile.
  2. Click "Generate full report" on their profile, or in the politician card on the /search "By politician" view.
  3. Confirm the topic. A modal asks what question or topic you want the report focused on. The clearer you are, the better the report — their position on healthcare privatization, 2018–present works better than healthcare.
  4. Confirm the cost. The modal shows how many credits the report will cost — calculated up-front based on how many speech chunks match the politician + topic.
  5. Submit. Credits are placed on hold (not yet spent). The report is queued.
  6. Wait for the email. Reports typically take a few minutes.
  7. Read. The report opens in a print-friendly view. Every quote and claim links back to the source speech.

Use this when the question is what does the corpus collectively say about X? — across multiple speakers, parties, and parliaments at once.

  1. Search Hansard at /search with whatever query, filters, and date range you want.
  2. Pick how many results to analyse. The "Analyse top N" picker appears above the result list (Timeline view) and on the Analysis tab. Choose 25 / 50 / 100 / 200 / 500 — or click "Other…" to enter a custom count up to the 500 cap.
  3. Click "Synthesize" (or "Synthesize this search" on the Analysis tab). The cost-preview modal opens, showing exactly how many credits you'll be charged for the chosen N.
  4. Submit, wait, read — same flow as the per-politician report. The synthesis opens with a stats table (party split, top speakers, time range), then a paragraph of headline framing, then five bullet findings each citing two-to-three source quotes.

Same trigger surface as Synthesize — same picker, same modal flow. Click "Map stances" instead. The output is structured by stance (for / against / conditional) with one exemplar quote per speaker per stance, useful when you're building a "who's on which side" frame for an article.

How pricing works

Cost is proportional to how much material the system has to read. Each kind has its own formula:

  • Full report on a politician: priced from the count of speech chunks matching the politician + topic.
  • Synthesize this search: 5 credits + 1 per 10 chunks analysed (e.g. 25 chunks = 8 credits, 200 = 25, 500 = 55).
  • Stance map for this search: 10 credits + 1 per 10 chunks analysed (slightly higher base because the structured output costs more).

The confirm modal always shows the exact final cost, your current balance, and what your balance will be after — no hidden charges.

If a report turns out to be much smaller than the upfront estimate (because, say, many candidate chunk IDs no longer exist), the unused portion is refunded to your balance automatically. If the worker fails entirely (model timeout, transient error), the full hold is released and you're not charged.

Hold vs commit

When you click submit, credits are placed on hold — your spendable balance drops, but the credits aren't actually spent yet:

  • If the report succeeds, the hold becomes a charge.
  • If the report fails (server error, model timeout, etc.), the hold is released. Your balance returns to where it was. You're not charged for failures.
  • If a report is stuck running for too long, the system automatically re-queues it (with the same hold in place) and tries again. You don't need to do anything.

What makes a good report query

Good queries are specific and time-bounded:

  • their position on the carbon price, 2019–present
  • how they've framed Indigenous reconciliation in their first vs second term
  • everything they've said about housing supply policy

Vague queries get vague reports:

  • what they think
  • bad things they've done
  • their best speeches

The system can only summarise what's actually in the Hansard record. It cannot tell you what they "really mean" or what they've "secretly done" — only what they've said in the chamber.

For synthesize and stance map, the same advice applies, plus: the search filters you have set when you click the CTA shape what gets analysed. A search filtered to "Conservative MPs, 44th Parliament" synthesises a different brief than the same query unfiltered.

Reading the output

A report has four sections:

  1. Summary. A few short paragraphs describing the politician's stated positions on the topic.
  2. Quotes and citations. Inline quotes with links to each originating speech.
  3. Tensions (when applicable). If the politician has held different positions over time, this section flags the shift with the dates and speeches involved.
  4. Sources. Every speech consulted, with date and link.

You can print, share, or save the report from the viewer page. Reports are private to your account — sharing means sharing the rendered HTML or a printout, not granting another account access.

Caveats

  • Reports describe what the politician said in the chamber. They do not describe what the politician voted on (votes are a separate data layer), what the politician has said outside the chamber (interviews, social media, press releases — outside our scope), or what the politician has actually done in office.
  • Reports use the speeches we have in the database as of the time the report was generated. If new Hansard ingests later, your report doesn't auto-update. Re-run the report if you want the most current picture.
  • Reports are AI-generated and may contain errors of summarization, attribution, or framing — even though every quote is verbatim. Always click through to the underlying speeches before quoting a report in publication.