Atlas
Find where legislation blocks a reform.
A reform touches provisions across many Acts. Atlas reads the Acts in scope against your reform questions and finds provisions that may block, duplicate, overlap or conflict with the change. Each finding quotes the law word for word and shows its reasoning.
In pilot with a federal department, scanning Commonwealth legislation. Findings are flags for expert judgement, not legal advice.
How it works
It starts from your reform
We set up each review with your policy team. The team supplies the reform objective, the questions to ask of the law and the Acts in scope. It signs off the set-up before any reading starts.
Atlas loads each Act from the Federal Register of Legislation at a fixed version, so every finding is read against the same text. We can load your own policy documents too.
Each Act gets its own reader
Atlas first codes every provision of the Acts once, using the same scheme for all of them. Each question then selects the provisions it needs, so you know how much there is to read before reading starts.
An AI reader for each Act then works through those provisions. It has tools to search and read every other Act in scope, so it can compare them as it reads.
Every quote is found in the law
Each finding has an issue type from your review, such as conflict, overlap or duplication. It quotes the provisions it relies on and sets out its reasoning. Where something could weaken the finding, it says so.
Before your experts see a finding, Atlas finds every quote word for word in the stored Act. A finding whose quote cannot be found is dropped.
Findings are read against each other
Each question is read on its own, so Atlas then reads the findings of every question against each other.
Where findings together show something none of them shows alone, Atlas records a combined finding that rests on each of them. Findings that make the same point are marked as related.
Your experts make the call
Your experts score the findings on two scales. Supported asks whether the quoted law says what the finding claims, and Material asks whether it is worth acting on. The pair of scores gives the verdict.
Scoring starts with calibration, where everyone scores the same few findings and the scores are compared. After that, each expert scores without seeing anyone else’s scores. An expert can reject a finding on a fuller reading of the law, and the reason is recorded.
Evidence your team can keep
Each finding has an evidence pack with the finding, its issue type, the law quoted word for word and the reasoning. A pack downloads as one file that opens without a network connection.
Your team can also download the findings list as a CSV file, or the whole evidence bundle. The bundle lists every file with its checksum, so anyone can check that nothing has changed.
Ask a question answers your team’s questions about the Acts and the findings. It shows a sentence only when every quote it rests on is found in the law. An answer can be saved as a Word document.
Hosting and security
- Hosting
- Hosted in the AWS Sydney region.
- Sign-in
- Only people added to the review can sign in. Sign-in requires multi-factor authentication.
- Roles
- Experts score the findings. Others on the review can read and download them without scoring.
- Record
- Every score and rejection is recorded with the person who gave it.
How a pilot runs
- 01
Set up
We map your process and your law into the product. Your security review runs at the same time.
- 02
Pilot
Your staff use it on your own documents, alongside the current process.
- 03
Decide
You get measured results, such as time saved and issues found.
Try Atlas on your own reform
A pilot runs on your own reform questions and the Acts in scope, alongside your current process.
Start a pilot