A useful first site record shows the evidence, its limits and what still needs checking.

Site judgement begins before concept design.

In the first hour, a professional weighs a small number of site questions that shape everything downstream. Errors become expensive to undo once the design has developed. In New Zealand that judgement also has a signature attached: restricted building work must be done or supervised by a named, individually accountable person, and a producer statement is recorded as a professional opinion rather than a guarantee.⁴⁵

The first output should be a decision record, not a completeness score.

A site summary is useful when the reviewer can see which records were checked, what each record says, where sources conflict and which questions remain open. A polished dashboard can still be unsafe if it hides the difference between a legal record, an indicative model and an untested assumption.

This changes the purpose of the first-hour exercise. The output is not a definitive site answer. It is a compact record that helps a professional decide what can be relied on now and what evidence must be commissioned next.

Public site evidence remains fragmented across several sources.

Many useful inputs are public, but they are scattered: parcels and terrain sit with LINZ, planning rules sit with each council, and hazards are spread across national models and regional registers in formats that do not line up. No single national source returns a property's complete planning and hazard picture, so the first review often depends on what can be assembled within the available time.⁹

AI can assemble the evidence into one reviewable site layer.

The first thing AI changes is reach. Much of the base is genuinely open and machine-readable: LINZ publishes parcels and titles under a Creative Commons licence with a public API, and national LiDAR now covers more than 80% of the country as a one-metre terrain model, free to use.⁶⁸ We have built a working version of this layer, which is what grounds the rest of this note (see the NZ site intelligence case study). The honest part is that the layers are not equal in quality, so the assembled picture has to carry that distinction on its face.

Firms can use general research or a source-controlled internal tool.

A general deep-research model can search the open web, while a source-controlled tool reads selected authoritative records directly. These methods have different strengths and failure modes, so the workflow should assign each one a defined role.¹⁴

The case for grounding load-bearing facts in controlled data is measurable. Using a verified source has been shown to cut hallucinated steps in structured output from around a fifth to under one in thirteen, while general web agents are documented as non-deterministic between runs even on identical prompts.¹⁶¹⁸¹⁹

A consolidated layer gives the reviewer four practical checks.

Bringing the evidence together exposes conflicts, gaps and interactions that are difficult to see while the data remains scattered. The prototype demonstrates these checks as a review method. Their effect on decision quality has not yet been measured.

A polished site layer can conceal uncertainty.

A clean, assembled site layer can look more authoritative than its sources justify. Much of the public record is indicative, modelled or out of date, so the interface must keep source quality, age and uncertainty visible.

The next study should compare decisions, not presentation quality.

The prototype shows that fragmented records can be assembled with provenance and uncertainty intact. It has not yet shown fewer late surprises, faster feasibility or better professional decisions. Those outcomes require a controlled comparison on real sites.