The model behind the Property Score.

Inputs

Twelve user-supplied fields and a small set of derived features. We deliberately keep the public version of the model interpretable — every weight visible, every input explained.

• Property type, year built, living area, lot size, beds, baths
• Walk Score, transit time to downtown, school rating
• List price, comparable rent

Weights

• Location 30% — walk + transit + schools.
• Value 20% — price-per-sqft vs. property-type benchmark.
• Yield 15% — annual rent ÷ list price.
• Age & build 15% — depreciation curve.
• Size 10% — log-scaled to a baseline.
• Lot 10% — log-scaled to a baseline (detached & townhouse only).

Bounds & calibration

Each sub-score is bounded to [20, 100]. We log-scale size and lot to reflect diminishing returns; we soft-clip the value score so a true bargain doesn't produce an unbelievable composite.

What's deliberately not public

On engagements, the model additionally consumes listing-level comparable sales, micro-location signals, strata health metrics, and a BC hazard layer. We will publish methodology updates as each module ships.