Skip to content

The Lab

Nine questions we are putting to the data, in public, before we know the answers.

Each one is registered before we run it, states what we expect to find, and is then handed to a process whose job is to prove it wrong. Where the data disagrees with the prediction is where the bugs and the stories live, so we publish that too.

207

watch house months, out of 63 watch houses across 17 months, when the average number of people held each day was higher than the number of beds. The table only exists because somebody asked for it on notice.

DocumentedQueensland Police Service, answering in the Queensland Parliamentasserted 2026-05-13Question on Notice 504 of 2026read 2026-07-31

What this does not claim

That we have read the documents. The public record layer holds what the state said, under obligation, on a date, and every entry in it is marked as unread, because relevance came from titles and subject lines rather than from opening the PDF. This figure is one of the few we have opened. And the same question asked for medical incidents and staffing in those watch houses: the Queensland Police Service refused those sections, so that data does not exist. A question's subject line says what was asked, not what was answered.

Ledger updated 2026-08-21

EXP-003LIVE · READ NIGHTLY

The watch-house ledger

How many children are in Queensland police watch houses tonight, how many are First Nations, and how long has the longest-held child been there?

3 of 6 gates passedNext: Ben signs the July digest, then automate
Built4
Testing4
Needed4
EXP-001CHECKED · AWAITING RIGHT OF REPLY

Who gets the youth justice money in Queensland

Which organisations win contracts, how much, and was there a competitive process? If not, was a reason recorded?

4 of 6 gates passedNext: Right-of-reply letter to the department, then publish
Built3
Testing1
Needed2
EXP-002STORY LIVE

Two invoices: detention vs community

What does one day of youth detention cost, what does one day of community supervision cost, and what does each buy?

6 of 6 gates passedNext: refreshes on each annual ROGS release
Built2
Testing1
Needed1
EXP-007NEXT · NEEDS THE MONITOR

Last updated: which official datasets have gone quiet

Which government data feeds have silently stopped publishing?

1 of 6 gates passedNext: Build the freshness monitor over every source
Built1
Testing0
Needed2
EXP-004PROPOSED

Who funds the alternatives

Where does the money for youth justice organisations actually come from, and how many are one contract away from closing?

1 of 6 gates passedNext: Define the fragility measure from ACNC filings
Built1
Testing0
Needed2
EXP-006PROPOSED · CO-OWNED

One honest place picture per justice reinvestment community

What do crime trends, school disadvantage, income support and detention exposure really look like in each community, and what does the community itself report?

1 of 6 gates passedNext: Pilot with one or two willing communities first
Built1
Testing0
Needed2
EXP-005BLOCKED · DATA CLEAN-UP DONE, QUEUED

Announced. Then what?

When a government announces youth justice funding, can the money be traced to a contract and a payment?

1 of 6 gates passedNext: Queued behind the first two publications
Built1
Testing0
Needed2
EXP-008ONE LAYER LIVE · 3 HELD

Every youth justice organisation in Queensland

Who are ALL the youth justice organisations and programs in Queensland, what money reaches each of them, and will they claim their own record when it is handed to them?

5 of 6 gates passedNext: Auspice worksheet confirmation (9 of 52 rows need a human), then the grant layer
Built6
Testing2
Needed3
EXP-009RUN · PREDICTION HELD · 99.6% COVERAGE

Who governs the organisations, and what does that connect

Across Australia, who sits on the boards of Aboriginal community-controlled organisations working in justice, what else do those people govern, and does the sector connect through shared leadership or is it genuinely locally held?

3 of 6 gates passedNext: A human read of the multi-board clusters before anything is published. Some are one Country and one family of organisations; some are the same four people across the same trusts.
Built6
Testing2
Needed3

How the lab works: six gates, in order, every time.

01

Register the question

Before touching the data we write down the question, why it matters, and which sources we will use.

02

Predict the answer

We state the expected result in advance. Where the data disagrees with the prediction, that is where the bugs and the stories live.

03

Run it with plain code

Every published number comes from a script anyone can re-run. AI helps us read and draft; it never produces a figure.

04

Try to break it

An independent check re-derives every figure from scratch and hunts for double counting, bad parsing and wrong assumptions.

05

Right of reply

Anyone named critically sees the finding before the public does, and their response is published alongside ours.

06

Publish, with receipts

Every figure carries its source, its date and its query. When we get something wrong, the correction publishes at the same volume.

9 experiments · 2 published · 1 live