Method

Every material claim traces to a named source

Facts that drive a guide, calculator or tool — thresholds, rates, grant amounts, legal rules, process descriptions — are stored as individual, dated claims in a versioned graph, each linked to the specific government, regulator or authoritative page it came from. Guides read from these claims rather than having numbers typed directly into prose, so a figure a reader sees links back to where it came from and when it was last verified. See the knowledge base for the full source catalogue, organised by subject.

How a source gets in

A source is fetched directly from the issuing authority's own site — a government department, regulator or official register — archived with a content hash, and a specific claim is extracted from it in plain language. A source's tier, retrieval date and document type are recorded alongside it. A weekly monitor re-fetches every archived source and compares a stable hash of its cleaned text (not raw bytes, which change on every request from cache-busting and tracking parameters with no real content change) against what's stored — a genuine change is flagged for human review, never republished automatically.

Claims are never silently overwritten

When a claim needs updating, the previous version is marked superseded and kept, with the date it was superseded and what replaced it — visible on the relevant knowledge-base subject page, not deleted. A reader can see what Delora previously stated as well as what it states now.

What the build itself enforces

Delora's site build validates every active claim before publishing: a claim with no source, or with signs of a formatting corruption (a stripped currency figure, a leftover template placeholder, or a missing value printed as text), fails the build rather than shipping. This isn't a manual checklist — it runs on every single regeneration of the site.

Facts, estimates and opinions are kept distinct

A suburb's median house price is a fact from a named dataset. Delora's hedonic price model is an estimate, explicitly labelled as modelled context with its assumptions and confidence range stated, never presented as a valuation or a prediction. An editorial recommendation, where Delora makes one, is labelled as guidance, not fact. If a claim can't be traced to a public source, Delora doesn't present it as fact. See our fair-value model's public track record for how well those modelled estimates have actually performed against real subsequent sale prices — wins, misses and all.

Some suburb-level figures are also published as open data: documented CSV files of Victorian suburb figures, each with its sources, licence, data date and a recommended citation. A dataset goes there only when every source behind it has had its licence read on the publisher's own page and that licence allows reuse.

Use of AI

Delora's guides and site code are drafted with AI assistance, working from sources fetched and registered directly rather than a model's general training knowledge. Every material claim is checked against its cited source before publishing — AI assistance speeds up drafting, it doesn't replace sourcing a claim or disclosing a limitation honestly.

What Delora won't publish

Corrections

Found something wrong? See the corrections page for how Delora handles errors, or get in touch directly.