Montreal Condos is an independent, solo-run project by Alexandre Godard. It's not affiliated with Centris, DuProprio, Realtor.ca, any brokerage, or any government body.

Questions, corrections, or a data-removal request? Email contact@montoit.ca.

Listings are scraped from Centris, DuProprio, and Realtor.ca every 30 minutes, with deeper backfills through the day and a nightly re-derivation pass.

Each listing is joined to public reference data that we read but don't own:

• Municipal assessment roll — évaluation foncière: assessed value, building size, unit counts.
• Building permits from the boroughs' open data.
• Schools with Fraser Institute rankings.
• Transit stations and stops.
• Flood zones, heat islands, and contaminated sites (Québec GTC land-registry of terrains).
• SPVM crime incidents, aggregated per area.
• Bank of Canada 5-year mortgage rate (for the monthly-cost estimate).
• 2021 Census demographics from Statistics Canada.

Deal score. How far a listing's asking price sits below comparable value. Where a neighborhood has enough matched sold data (its ask→sold calibration passes the gate), we anchor on a sold-based estimate; otherwise it degrades to the weaker price-per-square-foot-versus-neighborhood-median signal, and the wording is hedged so we never imply a sold number we can't back.

Comps. Comparable units shown as evidence for a fair price — a range and a percentile, not an appraisal. We surface the inputs and let you read them your way (ADR-0002).

Close estimate. In neighborhoods with enough matched Land Register sales, we project a likely close price by applying the measured ask→sold gap to the listing's original ask. Thin-data neighborhoods get no estimate, and limited-data ones are flagged as such (ADR-0012).

Ask erosion. For units that left the market, how much of the original asking price they shed before disappearing — a lower bound on the real negotiation discount, since the actual sale price is at or below the last ask.

Building churn. How many distinct units in a building were listed in the past 12 months, against the building's unit count from the roll. Evidence of turnover, not a verdict on the building.

How fresh is the data?

Listings are re-scraped from Centris, DuProprio, and Realtor.ca every 30 minutes. A nightly redrive re-runs every listing through derivation so it picks up reference-data changes (municipal roll, permits, mortgage rate) even without a fresh scrape.

Why can a listing be delisted here but still live on the source?

The 30-minute scrape mainly heartbeats the newest results; the long tail is re-confirmed by a periodic deep backfill. A listing is marked delisted when we haven't seen it past a per-source threshold (30–36 hours). If it slipped out of our scrape window but is still on the source, that gap can briefly show it as delisted here before the next deep pass re-confirms it.

What does “possible flip” mean?

On the Land Register lookup, we flag a unit as a possible flip when the current owner bought it within the last 24 months — a short hold, not a judgment. It's a fact about timing, not an accusation or a prediction.

How does cross-source dedup work?

When the same physical unit appears on more than one source, we link those listings into one canonical group and show a single primary — chosen by a fixed source-priority rule, matched on normalized address and unit. The others are kept as siblings so nothing is lost, but you see one unit, not three copies.