Every AVM is only as good as the data feeding it. Transaction prices, listing signals, and property attributes — and how fresh they are.
An automated valuation model estimates property value using data instead of a physical inspection. The quality of the estimate depends entirely on the quality, completeness, and freshness of the data feeding the model.
Every AVM, regardless of methodology, needs three categories of input: confirmed transaction prices (what properties actually sold for), property attributes (size, type, age, location), and market context (what comparable properties are currently listed at, how long they take to sell, and which direction prices are moving).
Confirmed sold prices are the ground truth of any valuation model. Without knowing what similar properties actually transacted for, a model is estimating in a vacuum.
In Canada, transaction data is fragmented. Land registry records (Teranet in Ontario, land title offices in other provinces) capture legal transfers but often with delays and without the listing context that preceded the sale. Listing-based sold records capture the transaction price alongside the full listing lifecycle — original asking price, price changes, days on market, property details — but require a continuous capture pipeline to maintain.
BrightCat maintains 899K+ sold events matched to their full listing lifecycles, including 194K+ verified repeat-sale pairs. Repeat-sale pairs — the same property sold at two different points in time — are the highest-signal records for AVM training because they isolate price appreciation from property-level variation.
A model trained only on past sales is always looking backwards. Active listings represent current market conditions — what sellers are asking, how long properties sit before selling, and whether prices are rising or falling in a specific micro-market.
Weekly listing data provides the freshness that assessment-based or registry-based AVMs lack. A property assessed at $500,000 two years ago in a neighbourhood where current listings average $620,000 tells a different story than the assessment alone suggests.
BrightCat tracks 5.8M+ Canadian residential properties weekly, capturing every listing event, price change, and status transition. This gives AVM builders a current-market overlay on top of historical transaction data.
AVMs need to know what they are comparing. Square footage, lot size, property type, number of bedrooms, year built — these attributes determine which properties are genuinely comparable and which are not.
In Canada, property attribute data comes from multiple sources: municipal assessment authorities (MPAC in Ontario, BC Assessment, etc.), listing records that include agent-entered property details, and land registry records that capture legal descriptions.
A model trained on ten million records that are all two years old will underperform a model trained on one million records that are all from the last 90 days. Property markets move. Neighbourhoods gentrify, correct, or stagnate. A model needs to see what the market is doing now, not just what it did historically.
Weekly data refresh is the minimum cadence for an AVM that needs to track market shifts. Monthly is usable. Quarterly is market commentary. Annual is a census.
BrightCat Core provides the integrated dataset for AVM construction: twelve years of continuous listing history, 899K+ lifecycle-matched sold transactions, 194K+ repeat-sale pairs, weekly refresh across all ten provinces, and persistent property identifiers that survive relisting cycles.
The repeat-sale pairs are particularly valuable for model calibration. Because they track the same property across multiple transactions, they allow the model to measure actual appreciation at the individual property level rather than inferring it from cross-sectional comparables.
See the signals for yourself — real data, updated weekly.