Methodology

How we count, and how fresh it is.

KRAYN states a number: 90+ government datasets. A number like that is worth nothing unless you can check it, so this page gives you the counting rule, the arithmetic it produces, and what that arithmetic excludes. It does not yet publish the register dataset by dataset: the refresh table below is the curated Australian subset we hold ingest dates for, not the whole count. Disagree with the rule and you have enough here to argue with it.

The counting rule

One published dataset, per publisher, per check. A multi-layer service counts once — NSW ePlanning’s Principal Planning Layers is one dataset, not six, and NHVR’s seven restricted-structure layers are one, not seven. Genuinely separate registers from the same publisher count separately, because they are separate instruments with separate consequences: Victoria’s Heritage Register and its Heritage Inventory are two, and the NSW coastal SEPP’s clauses 2.7, 2.10 and 2.11 are counted apart because one makes your works designated development and the others impose consideration duties.

A dataset only counts if it is ingested and on a read path a report actually executes. Data sitting in our database with nothing reading it is not a capability, and counting it would turn a claim about what we check into an inventory of what we store.

The arithmetic

109government datasets wired into live consumption

− 16region-scoped to the US, UK or NZ — real, but they cannot fire on an Australian address

= 93readable at an Australian address

→ 90+what we say, rounded down

We round down, deliberately. On a page whose argument is that we tell you what we cannot be sure of, the only safe direction to be wrong in is under-claiming.

We audited that subtraction on 17 August 2026, one source at a time, and it held: the sixteen excluded sources are each genuinely US, UK or New Zealand, and nothing internal, non-government or foreign sits in the 93 behind the figure. Internal ingest jobs, crowd-sourced data and non-government sources are all excluded by name and always were — the same exclusions the refresh table below applies, which is the point: one rule, both surfaces.

Three numbers, three names

KRAYN publishes more than one count and they measure different things. They are easy to conflate, and conflating them would flatter us, so we keep them apart and labelled.

Per report

checks stated exactly

The checks a report runs at your boundary — flood, bushfire, heritage, zoning and the rest. Every report states exactly how many checks ran at that address, and which ones didn't. The exact checks run depend on the address, state and available coverage.

90+

government datasets

The published sources those checks read. One check often reads several; some datasets serve several checks.

238

automated checks

Checks on KRAYN itself, not on your site — they run against the platform to catch a figure, a claim or a rendered surface drifting out of line. 186 of them block a release: the deploy stops rather than shipping past them. This page is held by one of them.

A number is never printed next to another without both being named. “up to 30 site checks on one address in Queensland, across 90+ datasets” reads like one bigger claim, and it is two different ones.

What the count leaves out

Listing exclusions is the part that makes the rest checkable — anyone can inflate a number by quietly widening what qualifies.

Data we hold but do not read. Ingested datasets with no read path are excluded by name in our census, not silently omitted.
Non-government sources. OpenStreetMap, SoilGrids and commercial weather services are used in places and are not government datasets, so they are not in this count.
Region-scoped datasets. Our US, UK and NZ layers are wired and working, and cannot be read at an Australian address — so they are excluded from the figure Australian pages state.
Layers we have not built. Declared-but-unbuilt sources count for nothing until a report can read them.

When each dataset was last refreshed

The datasets below are the ones we store and refresh on a schedule. Read live from our refresh log, not typed by hand. Many other sources are queried from the publisher at report time; the response may still carry publisher-defined currency and limitations. That is why this table is shorter than the count above.

Reading the refresh log…

We show overdue refreshes rather than hiding them. A date here is when we last pulled a copy, not when the publisher last released one — a row past its refresh window may well be behind the custodian’s current version, and we would rather you knew that than assume our copy is the live one. Some rows read not yet ingested: those are datasets we have listed and do not yet hold, so there is no copy to be stale and no refresh due. They are planned, not held.

Every finding in a report names the dataset it came from, so you can trace any single statement back to the publisher who issued it. The count above is only a summary of that — the citation on the finding is the real receipt.

Every check we run, named Read a real report

Dataset figure as at August 2026.