How the record was read
This site reads a single cloud database built by the insight-bridge pipeline from the public record of the Joint Select Committee on Artificial Intelligence's inquiry: 170 submitters, 188 evidence documents, 2,949 passages, 2,225 submitter–topic memberships, 510 submitter-level stances, 105 sector perspectives and 1,401 answers to the committee's 14 terms of reference. Everything on every page is a live query against Cloudflare D1 (the run's own tables, copied whole) and Vectorize (the embeddings), through fully-typed Drizzle. The same data is exposed to agents over an MCP server at /mcp.
What was done, in order
Each step below is a stage this run actually executed, read from the run's own record of itself; a stage the run was configured to skip is not listed. The first stage, the collection, is the one thing the run cannot know about itself and is written from the collection's own notes.
- Collecting the evidence from the Parliament’s site
Every public submission record on the inquiry’s page was discovered on 18 September 2026 — 170 submission records and 188 written PDFs — and each PDF was downloaded with its submission number, the Parliament’s own descriptor (Submission, Attachment 1, …) and its DocumentStore link, so every document here links to its original. Text was recovered from all 2,006 pages with native extraction, and OCR where a page carried no text: 570 pages in 101 documents use accepted OCR, with 25 hash-pinned page reviews and duplicate page text removed. The inquiry held one public hearing (18 September, Canberra) with a program but no published transcript, so there is no hearing evidence in this record: every submitter here wrote to the committee. One source defect is confirmed and kept visible: Housing Industry Association’s submission 146 has pages that are black in the published original, so its available text is partial and its views are not to be inferred as complete. Four map pages in submission 170 must be read in the original PDF. The bundle is a snapshot of the evidence published at discovery, not of everything the inquiry eventually received.
- Reading in the evidence documents
Every evidence document was stored whole under its submitter, its text cut into passages small enough to compare — 2,949 of them across 188 evidence documents from 170 submitters — and each passage embedded with text-embedding-3-large, which is what makes a search by meaning, and the grouping below, possible at all. Each submitter carries what the inquiry records about it (its submission number, what it lodged) and each document its type, the Parliament’s descriptor and the link to the original file.
- Reading each evidence document end to end
A language model read every evidence document in full against this run's profile and recorded what it found as key points, each with the verbatim quotes that support it: 2,252 key points and 6,999 quotes. No quote is written by the model — each is selected from the evidence document's own sentences and checked back against its text before it is stored; 99.8% verify, and a quote that does not is never shown here. The same reading classified each submitter's sector from what it wrote, and put the committee's 14 terms of reference to every document: 1,401 answers were found, each with a position, key points and quotes. An unaddressed clause means no answer was located in the available text, never that a submitter has no view.
- Settling what counts as one submitter
The names the model inferred were put to a person beside the names the Parliament lists, before anything was counted, so that one submitter appearing under two spellings is counted once in every figure on this site. The listed name stands; the model's reading is kept beside it.
- Grouping the passages into topics
Passages were grouped bottom-up by what they are about, with HDBSCAN. A group had to hold at least 5 passages to be a topic at all, and a penalty of 1 on passages from the same submitter kept one submitter with many passages from forming a topic on its own. The settings were tuned at a review gate: 9 attempts were tried before these were committed (they are listed below). Each submitter is then placed against every topic at a graded strength — exemplar above 0.9, high-value above 0.8, member above 0.6 — which is why a submitter can belong to several topics at once. 26 topics resulted, over 186 documents' passages.
- Writing each topic up, then reading positions against it
Each topic's proposition and key points were synthesised from its own members FIRST. Only then was each exemplar and high-value submitter's position assessed against that finished proposition — 510 written stances, and the same done for each value of sector: 105 perspectives. The order matters: a position here records how a submitter stands against one topic's particular framing, and is never a verdict on the subject in general. The model never writes quote text at this stage: every citation is chosen from a catalogue of the submitter's own sentences.
- Removing the groups that carry nothing
A model flagged topics whose synthesis reported no substantive shared content, and a person confirmed or rejected each flag — and could confirm a topic the model had not flagged. 3 topics were confirmed and are excluded from every figure on this site. The flag alone never removes anything.
- Offering the topics upward into themes
The topics were grouped again, generation by generation, each generation clustering the one below it until it stopped merging, held to 1 generation. A topic with no theme above the attachment threshold was attached to its nearest theme rather than left alone. This site reads the tree as the run left it, however many generations that turned out to be — here, 7 families above the topics, each named by the model from its members.
- Placing the topics in a sky
Each topic's place in the galaxy was computed once and stored with the run, seeded so it is the same sky every time. That is what lets these pages draw the constellation as plain graphics, with no WebGPU and nothing solved in your browser.
- Loading this site
Outside the pipeline, before loading: the run database was copied whole to the cloud — every table as the pipeline wrote it, with the passage embeddings moved to a vector index — and two facts were derived from it for these pages, each document's and each passage's dominant topic. Nothing was flattened, re-verified or re-classified here; every figure on every page is a live query over the run's own tables.
The clustering attempts
The tuning gate walked these settings before the last row was committed. Minimum cluster size and minimum samples decide how dense a group must be to count as a topic; the submitter penalty keeps one submitter's passages from forming a topic on their own; noise is the passages left in no topic.
| Attempt | Min cluster size | Min samples | Submitter penalty | Topics | Noise passages |
|---|---|---|---|---|---|
| 1 | 3 | 3 | 0.25 | 15 | 2,824 |
| 2 | 3 | 3 | 1 | 13 | 2,831 |
| 3 | 3 | 2 | 1 | 22 | 2,768 |
| 4 | 3 | 1 | 1 | 69 | 2,604 |
| 5 | 3 | 2 | 1 | 22 | 2,768 |
| 6 | 2 | 2 | 1 | 67 | 2,738 |
| 7 | 3 | 2 | 1 | 22 | 2,768 |
| 8 | 3 | 1 | 1 | 69 | 2,604 |
| committed | 5 | 1 | 1 | 26 | 2,704 |
What each submitter was classified by
Every facet the run's profile declares. A facet marked compared is one the analysis wrote a perspective for, value by value — the lens you can read a topic through. The inquiry's own records (submission number, document type, text quality) ride beside it as facets of the record without being compared.
What a position means here
Read weakest claim first. Supports, Builds on agree with a topic's proposition; Mixed, Redirects, Opposes push back against it; Unclear reads as neither.
Position is relative to the cluster proposition, not the source's tone. Nuance alone is not Mixed, and constructive engagement alone is not Builds on.
How to read the output
These rules apply to every Insight Bridge corpus. They are properties of the method, not caveats about a particular run.
- Positions are relative to a proposition. A source’s position records how it stands against that cluster’s particular framing — not whether it agrees with some absolute claim. The same source can support one cluster and redirect a neighbouring one that covers similar ground differently.
- Propositions are synthesised from the cluster’s own members. Because the argument is built from the sources that were grouped together, and those sources are then assessed against it, a degree of agreement is built into the method. Comparisons between groups carry weight; a corpus-wide agreement rate does not.
- Counts describe the corpus, not the world. Every corpus here is curated. “N sources say X” measures what was collected and is never a measure of how common X is in the field.
- Clusters differ in how many distinct sources back them. A long document can fragment across many clusters, so weight a theme by the distinct sources beneath it rather than by how many clusters it contains.
- Every extraction and position is a model judgement. Key points, stances, propositions and syntheses are produced by a language model reading the source. They inherit its calibration and are not determinations of fact.
In this run
Removed after review. “Evidence Base: No substantive position provided” — The summary explicitly states that the source supplied no identifiable proposition, policy direction, argument or evidence. “AI Investment: Pool public capital nationally” — confirmed by the reviewer without a model flag. “National Adoption: Pair productivity with strategic control” — confirmed by the reviewer without a model flag. They are excluded from every figure on this site, including the 23 topics counted above.
Partly unreadable in the published original. (Housing Industry Association (HIA) — Submission — User manually confirmed the black pages in the original HIA submission and could not locate another published copy, including on the HIA website. This is a confirmed source defect, not a pipeline extraction failure.). What is here is that submitter's available text, not its complete evidence; do not infer absent views from it.
No hearing evidence. The inquiry's one public hearing had no published transcript when the evidence was collected, so every submitter here wrote to the committee and the record carries written evidence only.
The terms of reference are answered per document. A submitter with an attachment may answer a clause twice; the Terms page counts a submitter once. An unanswered clause means no answer was located in the available text, never that a submitter has no view.
What else was asked of every evidence document
Beyond the standard reading, this run declared its own questions and put each of them to every evidence document.
Which model did what
Named because it matters: the reading, the writing-up and the grouping were done by these, and a different model would have produced a different record.