St. John’s Properties gives every home in its database one permanent page, however many times it is listed. Listings come from CREA’s Data Distribution Facility, which describes each listing on its own. This post is about the step in between: deciding which listings belong to the same physical property.

Why listings are not properties

A house can be listed, withdrawn and relisted with a new MLS® number. A condo building has many units at one street address. A new subdivision has lots that share a builder’s description word for word. If every listing became its own property, a relisted house would lose its history; if similar listings were merged too eagerly, neighbouring homes would share one.

So the platform keeps three layers apart: the property (the physical home), its listing episodes (each period it is marketed) and its history (relisted, price reduced, details changed, removed from the feed). Entity resolution decides which property each new listing attaches to.

Step one: normalize the address

The same address arrives written in many ways. Before anything is compared, it is normalized: street suffixes and directions are folded to one form (“West” and “W”), “Saint” and “St”, “Mount” and “Mt” are made consistent, and every way of writing a unit (“Unit#403”, “#403”, “Apt 403”, “Suite 403”) becomes the same token.

The property’s ID is then a deterministic UUID of the normalized address and municipality. Deterministic means the same address always produces the same ID, on any run, without a lookup, so a relisting at an unchanged address lands on the same property automatically.

Step two: merge only on strong evidence

Addresses still change between listings: a typo is fixed, a unit is added, a street is spelled differently. When a listing’s ID has not been seen before, it is folded into an existing property automatically only when the evidence is strong:

A known alias

The address is a spelling the property has been seen under before. The first claim to an alias wins.

Same place, same door

An existing property within 30 metres with the same civic number and the same unit designator.

Never condo units

Units in the same building share an address and a location; they are never merged automatically.

Merges are permanent

A merged ID always resolves to the surviving property, so the next sync cannot re-create the duplicate.

Step three: send weak evidence to a person

Some pairs look alike but are not certain. These go to a review queue instead of being merged:

  • Identical long descriptions (120+ characters) shared by exactly two properties within 2 km with the same civic number and lot or unit designators.
  • The same civic number and designators on a similar street name, 30 to 150 metres apart.
  • The same address filed under a neighbouring municipality within 2 km.

A reviewer can merge the pair, which keeps the older property and moves every listing, history event and alias onto it, or mark them as different properties, which records the decision so that pair is never proposed again.

The 518 wrong merges

The limits in that list were not in the first version. The first live run without them proposed 518 pairs, and nearly all of them were different parcels. Builders reuse one description across a whole subdivision, so identical text is weak evidence on new construction; and “Lot 2” and “Lot 3” on the same street are neighbours, not the same home. The current rules add the civic-number, designator and distance conditions, and exclude shared subdivision remarks and neighbouring lots.

The lesson I take from it: a matching rule that looks reasonable on a few examples can be badly wrong at the scale of a whole feed. Run it against everything, read a sample of what it proposes, and only then let it merge anything.

Why not just use parcel numbers?

A parcel or property ID from the land registry would make most of this unnecessary. In practice the feed fills its parcel-number field on 11 of about 3,400 records, mostly with free text. Reliable parcel identifiers have to come from a municipal dataset, and the platform keeps a place for them on each property alias for when they do.

What it makes possible

  • One permanent URL per home; old listing links and merged IDs redirect to it.
  • A listing history that survives relisting: when it was first seen, every price change, and when it left the market.
  • Pages that stay useful after a sale, as a record of the home rather than a dead link.

Related