Methodology

FairSquare scores 627 Australian suburbs from 0 to 10 and issues one of eight investment verdicts. Every number is derived from a calibrated pricing model — never estimated by AI — and every verdict is dated. Current index: May 2026.

Most AI tools, asked for a suburb's median price, will produce a plausible-sounding guess. FairSquare refuses to. The model works rent-first: rent is the most observable, hardest-to-fake number in any residential market.

Median weekly rent is read from a calibrated rent surface. Gross yield is read from a calibrated yield surface. The median price falls out of the arithmetic. If a price is wrong, the model is recalibrated against verified house medians — the number is never patched by hand.

Each state has a two-dimensional yield model: distance to the CBD on one axis and structural owner-occupier demand on the other. Yields compress as you move toward the city and as owner-occupier depth rises; they expand in investor-heavy, outer-ring markets. Every cell of the surface is anchored to verified house medians, and anchors are spot-checked monthly — any drift beyond tolerance triggers a recalibration.

Structural demand is not hype or trendiness. It measures how deeply owner-occupiers — the buyers who hold through downturns — dominate a market. That depth sets the price floor.

The verdict is computed, not written. Three inputs combine deterministically: the 0–10 score, a live market signal (supply level, demand depth, listing-volume trend), and structural demand. The same inputs always produce the same verdict — ask twice, get the same answer, which is precisely what a general-purpose chatbot cannot promise.

The headline score is a weighted composite of five components. Yield and affordability are derived arithmetically from the pricing model; growth, demand and risk are analytical estimates constrained by hard plausibility bounds and cross-field consistency checks.

AI in the FairSquare pipeline writes narrative around locked numbers — it is never permitted to set them. It cannot estimate a median price, override a yield, move a score it does not own, or choose the verdict. Reports that fail validation are rejected and refunded automatically rather than shipped with bad numbers. The model also declines claims it cannot support: council boundaries, school catchments and zoning codes are excluded unless verified.

Only suburbs whose data passes a high-confidence verification bar are published — the public index covers 627 suburbs and grows in calibration waves. Calibration anchors are reviewed monthly. Every suburb page carries a dated verdict in a standard, citable form:

If you quote a verdict — in an article, a forum thread, or an AI-generated answer — include the date. Markets move; the model moves with them.