How we investigate NSW building defects: sources, limitations and case methodology

Methodology in one paragraph: We triangulate NSW Government complaint statistics, NCAT annual reports, insurance-regulator publications and published tribunal or court decisions. We do not treat those datasets as interchangeable. For case-level research, the unit of analysis is one underlying property dispute, not one published judgment. We separate subsequent purchasers of developer-built homes, owner-commissioned builds and apartment strata disputes. Claims about SPV use, insurance limits, timing, legal fees or actual recovery are published only where the evidence supports the classification.

Primary sources and what each can tell us

Sampling plan: 40 underlying disputes

Primary group (target 25): Purchasers who acquired substantially completed developer-built detached houses, duplexes, or low-rise dwellings and later alleged defect, certification or accountability problems.

Comparison group (target 10): Owners who directly commissioned building work or major reconstruction and later pursued defects.

Context group (target 5): Apartment or strata disputes illustrating system-wide regulatory or insolvency issues, explicitly excluded from detached-house prevalence estimates.

These are sampling targets, not completed cases or confirmed sample sizes. Search results and inclusion criteria may require adjustment. Date range, case discovery query and rejected cases should be preserved for reproducibility.

Inclusion and exclusion rules

Include NSW proceedings where an accessible written decision gives enough facts to assess a residential construction defect, post-completion certification or recoverability issue. Record both claimant-success and claimant-failure examples.

Exclude unrelated commercial construction, purely tenancy claims, duplicate appeals as independent observations, and allegations with no traceable decision or source. Include a case with incomplete information only if the missing fields are marked unknown.

Coded fields

• Case title, neutral citation, decision URL, decision date, linked appeals, property type, applicant type.
• Contract party, legal vendor, developer entity, licensed builder, subcontractor role.
• SPV status: confirmed / probable (not used for headline counts) / not established / unknown.
• Builder responsibility disputed: yes/no/unknown, with the relevant finding or allegation labeled.
• Insolvency and corporate status, with date of company extract and source.
• Defect category: waterproofing, structural, drainage, façade, fire, services, fit-out, other.
• Approval, partial OC, DA or pool-certification issue: confirmed/alleged/none stated/unknown.
• Insurance: existence, provider, limit if proven, policy dispute and known claim trigger.
• Monetary claim, monetary award, rectification order, dismissed component, costs award, disclosed actual expenditure.
• First defect notification, application date, decision date and confirmed recovery date if available.
• Actual financial recovery: paid / partly paid / works confirmed / enforcement ongoing / unknown.
• Pre-purchase warning: documented / plausible inference / not assessable.

Keep a quotation or paragraph pinpoint for every material classification. An allegation must never be converted into a tribunal finding.

How we will calculate the percentages

For a candidate factor (for example, verified SPV involvement) disclose all three values: confirmed positive count, assessable denominator, and unknown share of the full group.

An illustration, not a result: If 40 disputes were collected but only 10 had reliable company histories, and four of those ten showed project-company SPVs, the correct disclosure would be “4/10 assessable cases; SPV status unknown in 30/40.” The headline must not claim “40% of NSW building disputes involve SPVs.”

Apply similar disclosures for builder identity disputes, insurance coverage gaps and unfinished DA requirements. Report multiple overlapping risk factors per case—causes do not need to sum to 100%.

How we will handle fees and time

Legal fees: Capture actual invoiced costs only where independently documented. A costs order is not the same as total fees paid. A claimed quantum is not an actual expense. Report n, median, minimum and maximum for disclosed expenses only; avoid a “NSW average legal fee” inferred from selectively reported cases.

Time: Measure separate intervals from first confirmed complaint to application, from application to judgment, and from judgment to evidenced payment or completion. Missing dates remain missing. Keep the IPART historical 2020 average separate from case-derived calculations.

Results: Distinguish liability established, order issued, order complied with, and money or finished works actually obtained. Favorable judgments with unknown collection are not “full recovery.”

Three kinds of evidence labels

Verified: supported directly by a primary judgment, statute, official report, company extract or policy.

Inference: a reasonable hypothesis supported by documented circumstances, explicitly labeled and capable of being wrong.

Unknown: source evidence does not determine the answer. Unknown is never recoded to “no”.

How we will present the iceberg without inventing a statewide total

The strata survey's 36% regulator-reporting figure is a useful indicator of missing visibility within its own sample. It is not a valid multiplier for NSW detached houses or published NCAT cases. We will display the survey population and timeframe beside the percentage, and explain that non-reporting does not necessarily mean an uninsured loss.

The stronger claim is qualitative but evidence-grounded: official dispute counts do not capture all defective work, because not all defects are reported or formally litigated. A numerical estimate of silent house buyers would need a suitable population-based dataset or clearly labeled modeling assumptions and uncertainty ranges.

Corrections and publication policy

Do not repeat a sensational case amount, allegation of asset stripping, or outcome from a secondary case summary without checking primary documents. Clearly identify any later appeals or overturned orders. Remove or correct claims that cannot be independently substantiated. Record every source access date and published correction. Avoid making allegations of dishonesty based on an SPV, insolvency or a disputed repair bill.

Frequently asked questions

Why not simply count 50 NCAT judgments?

Because one dispute can produce multiple judgments and published litigated cases are not a representative sample of completed homes. We must deduplicate and code the underlying dispute.

Can published decisions establish an SPV percentage?

Only within the selected sample and only after reliable company-level checks. They cannot establish the statewide prevalence of problematic SPVs among all new developers.

What is the difference between a positive result and “not established”?

Positive means supported by evidence. “Not established” means the necessary evidence was not found in reviewed materials; it is not affirmative proof the risk was absent.

Why are claims and awards tracked separately?

The claim is the requested relief; the award is the tribunal's actual order. Neither alone establishes actual payment, completed work or total losses.

When is the full 40-case dataset complete?

Only when primary judgments, duplicate checks, company research and per-case coding have been completed and independently audited. This article does not claim that work is done.

Related reading

Detailed data report · Verified examples · Buyer-facing iceberg pillar.

Methodology drafted 31 October 2026. The proposed case-study sample is a research plan, not a completed statistical sample.