XY:Ref — a reference to a place
Every value has a receipt.
Planning due diligence is still a manual job. A dozen mapping systems, a dozen tabs, and then the numbers get copied by hand into a feasibility template — where every paste is a chance to get one wrong, and no one downstream can tell which cell came from where.
XY:Ref exists to remove the copying, not the judgement. It reads the NSW public registers live the moment you ask, and fills the template you already use — with the query that produced each number sitting behind it. The facts arrive filled. The judgement stays yours.
A due-diligence sheet gets checked by someone else. A number that cannot be traced back to where it came from is a liability, not an answer — so the test we hold every part of this product to is whether an assessor, handed the sheet and nothing else, can re-run the value themselves and get the same result.
The team behind
The registry is data, not code. That is not an accident.
XY:Ref is led by Danny Liang, its founder and technical lead. His doctorate — Western Sydney University, on model-driven web application development — was about building an application from a model of it rather than by hand, and the reference implementation from that thesis was commercially adopted by the university and by an industry manufacturer.
That is the idea the connector registry is: every source is an entry in a file — endpoint, fields, clauses, licence — so a new register is data rather than a release. It is what keeping pace with a moving set of government endpoints takes.
The engineering decisions here are his. The thinking behind them is not his alone: it comes out of years spent inside other teams — data, front-end, infrastructure — on problems that were not this one, and out of watching how the people who do this work for a living actually use what gets built for them.
He is an Adjunct Fellow at Western Sydney University — co-supervising PhD candidates, guest lecturing, mentoring final-year capstone students — and sits as an external industry representative on university and professional accreditation panels for computing, including the Australian Computer Society panel that accredited the university's computing degrees.
He co-founded an award-winning Sydney agency and website platform and led research and development on its content management system, and has since rebuilt that platform as an AI-native content management system with a built-in agent for layout design and content generation. Alongside it he has delivered web platforms and business systems for enterprise, council and university clients — among them a web-based approval workflow for a national telecommunications company. He has won at the Australian Web Awards and now sits on its judging panel.
For the past three years he has been engineering lead on a separate spatial data platform, embedded with the client team: address resolution, geospatial data engineering, mapping and vector tile layers, machine learning and propensity modelling. He coordinated a data team alongside external front-end and infrastructure teams, and integrated live data from NSW Land Registry Services — the operator of the state's land titles registry — and from commercial data providers.
That history is also why the core engine here is deterministic by design. The same address returns the same answer, and no language model sits in the value path, so no retrieved value is a guess. Artificial intelligence does the work it suits, at the edges — reading documents and instruments, binding the labels in your template to fields — where its output is graded and can be checked against the value underneath it. That is an engineering judgement about where each tool is trustworthy, not a limit on what he can build.
Machines have never been better at describing a place. They can pin an address to six decimal places, trace a lot boundary from the air, measure the fall of the land, and track and visualise every address on the market at once — the sparkle of dots behind the homepage's headline is exactly that, a snapshot of Sydney's listings. What they cannot do — however precise they become — is weigh what a place means: the significance an address carries in one person's memory and in no one else's. A career spent building for the web taught Danny to respect that line, and this product is drawn on it. XY:Ref takes the half a machine can honestly own — the facts of a place, each traced to the register that stated it — and leaves the other half where it has always lived. A reference to a place. The meaning of it stays yours.
- Doctorate
- PhD, model-driven web application developmentWestern Sydney University
- Appointments
- Australian Computer Society accreditation panel
- Award
- Australian Web Awards
- Domain
- Three years as engineering lead on a spatial data platform
- Sectors
- Enterprise, councils, universities, telecoms
The engineering decisions behind the product — why there is no database, why a failed source is named rather than dropped, why the setback is an input and not a constant — are written down on How it works.
What we commit to
Five claims you can test.
Every value carries a receipt
The source, the exact query that produced it, the minute it was read, its licence, and the clause it answers — beside the value, not in a metadata block you have to join back to it. Paste the query and you get the same answer.
The engine is deterministic
A parser, a scoring function, and arithmetic against named clauses. Same address in, same answer out — no model anywhere in that path. Paid features that do use AI are labelled as such, draw only on receipts the engine has already produced, and never set a number.
A gap is a fact too
A register we cannot reach is reported by name; a field we cannot answer says so and says why. An absent value and a failed lookup are different facts, and neither one renders as a blank cell.
There is no copy of the register
Your browser reads the government endpoints directly — no database in the middle, no nightly harvest. That is why the answer is as current as the register, and why there is no pipeline to go quietly wrong between the source and your screen.
Your template, filled — in your browser
Not another report to transpose by hand. Your Excel or Word file is unzipped, edited and re-verified locally: nothing is uploaded, formulas are left intact, prose is never overwritten, and the file is re-opened and checked before you are offered the download.
How each of these is actually done — address resolution, the concurrent reads, and the four confidence labels every value wears — is on How it works, alongside the published list of what this cannot answer.
What we will not do
The shortcuts are the whole product.
Each of these would let us fill more cells, cover more of the state, or answer faster. We don't take them.
- Present a value a model produced as if a register had answered it.
- Invent a number to fill a cell — a missing height stays missing, and no footprint is drawn from it.
- Claim a jurisdiction before its sources are in the registry and tested.
- Write over the formulas that make your workbook a model rather than a form.
- Charge for an address that failed to resolve.
What's next
Open source first — it makes the rest testable.
Open source registry
Every source is a JSON entry — endpoint, fields, clauses, licence, routing. Adding a jurisdiction is adding data, not writing code. Published so anyone can audit what we query and what we claim it answers.
Commercial connectors
Title and deposited plan, ordered and filled in place rather than arriving as a PDF to retype. On- and off-market status. Sydney Water feasibility. Significant trees. Construction cost rates — the missing half of a feasibility.
Interpretation, behind its own boundary
A model may read receipts the engine has already produced and draw a clause-level conclusion, citing the deterministic value underneath every sentence. It may never produce a number. It ships labelled, or it does not ship.
Where the panel runs
A Chrome extension for environments where a bookmarklet is blocked by policy, and native iOS and Android apps — the wallet, on the device the metaphor came from.
Beyond NSW
Other states, later and in order. A new jurisdiction is registry data rather than a rewrite — the newest connector already reads a national dataset that works unchanged in any state — and Victoria and Queensland are the natural next two. Each arrives here only once its sources are in the registry and tested against live addresses.
Coverage
New South Wales, today.
Built on NSW Government and Geoscape Australia services under their own open licences, which every value names. The connector registry is a file of endpoints and clauses rather than code, so a second state is data — but we will not put a state on this page until its sources are in the registry and tested against live addresses.
Questions, corrections, or a register you think we have read wrong: [email protected]. A reported wrong value is the most useful mail we get — it comes with a replayable query attached, so it can be checked in a minute.