From Regulation to Rules
"Do I need a permit for this?" The answer usually sits across an Act, its regulations and its guidance, and it depends on the place and the activity.
The person asking wants to know where they stand. The regulator wants to give a reliable answer, based on the clause. In both cases, the next question is "Why?"
What we built
Our approach starts with the Act. We encode trustworthy rules that can be tested so the answers are reliable. Every result carries four things: the outcome, the reason, the relevant clause, and whether an officer review is needed.
For the example in the video, we started with a question for the Environment Protection and Biodiversity Conservation Act 1999 (EPBC Act): "I want to clear regrowth, do I need a permit?"
A Google search would recommend trawling 1,200 pages of legislation: that's hard. What if we could change the question from a Google search to a conversation with the Act itself?
Encoding, testing and building the demo app took a few hours on our existing rules platform.
Source credits: Act text from the Federal Register of Legislation, CC BY 4.0. Satellite imagery © Google and its suppliers, as credited on the map.
What does this mean for regulators?
Our approach is built on four principles:
Rules you can read. Each clause sits beside the rules made from it, in plain if/then form. The Act's words come from the Federal Register of Legislation, word for word. Guidelines are encoded as guidelines, below the Act.
Proof before anyone relies on it. We write scenarios to thoroughly test the rules, including both sides of every boundary: 49 metres and 50, the day before a deadline and the day after. All of them will run again whenever a rule changes.
An answer you can explain later. The same facts give the same answer every time. Each answer names its clause: a review, a complaint or a question starts from the provision.
You make the decisions. Where the law asks for a decision-maker's view, the result reads "Needs a judgement the rules cannot make" and names the provision.
How it was built
- Take the law as published: The Act, the regulations and the standard, from the Federal Register of Legislation under its Creative Commons licence.
- Encode each provision as a rule beside the text it comes from.
- Write the test cases for assurance.
- Attack it. A separate review ran hundreds of tests, looking for one failure above all: a confident wrong answer. It found several. We fixed each one and rebuilt.
- Let public data answer what it can: We use open datasets, each named on screen beside the fact it supplied.
- Log every decision: No black boxes - a reviewer can always see how the system arrived at its conclusion.
Testing is the difference between trusting a result and hoping a search result might be correct when you ask. A defect in a rule can be found, fixed and shown to be fixed: applying standard software development to business rules.
If you work with regulation or compliance and would like to see your own legislation, regulation or rules become digital, contact us. We'd love to walk you through this example.
Legalese and stats
This is a demonstration built on public legislation by Regsoft, with no affiliation with the agencies that administer this law. It checks against the rules we encoded and is provided as general information, not legal advice. The scenario in the video is invented, and the land is unallocated State land near Millaroo, Queensland. The checks cover Commonwealth law; State vegetation and planning law applies separately. The Act text is compilation 70 of 24 August 2026. Each check names the parts of the law it leaves outside the encoding when a question lands there.
| Sections covered for referral | Act ss 12 to 28, 38 to 43B, 67, 68, 74B, 84A, 523 to 524A, 527E, 527F and 527H; Regulations 4.05 and 5.01 |
| Words from the Act | about 7,000 |
| Test scenarios | 168 |
| Checks on those scenarios | 890 passed, none failed |
| Adversarial review runs | 231 |
| Time to a decision | about 20 milliseconds on a laptop |
