METHODOLOGY
PUBLISHED BY THE LISTING.
How the Market Signal is calculated, the per-listing prediction-vs-actual on closed listings, and the accuracy standards the Service is held to.
How First Access Signals works
First Access Signals tells you when to act on a home, not just what a home is worth. The Service combines three things into one app:
1.
A Market Signal, a 0 to 100 score for any GTA home that estimates the probability the home would sell within 90 days at a competitive price, and the verdict on whether now is a good moment to list.
2.
An Automated Valuation Model (AVM) estimate of the home's market value, with a 90 percent prediction interval that tells you the range within which the actual sale price is expected to fall.
3.
A Buy Signal for neighbourhoods, which ranks active listings by timing, value, and lifestyle fit so a buyer can focus on the homes most likely to be a good move right now.
First Access Signals is licensed under the Toronto Regional Real Estate Board (TRREB) Virtual Office Website (VOW) Datafeed Agreement and is operated by Keith Godding under THE AGENCY brokerage.
Data sources
First Access Signals uses the TRREB VOW data feed, which provides a 24-month rolling history of every listing in the GTA, including final sale prices on closed deals. The Service combines the listing data with leading-indicator macroeconomic and neighbourhood data:
•
Bank of Canada policy rate.
•
Statistics Canada housing-starts and building-permits data at the Census Metropolitan Area level.
•
Canadian Real Estate Association (CREA) Home Price Index for the GTA.
•
Toronto building-permits and crime data at the neighbourhood level.
•
TTC and GO Transit schedule data for commute distance calculations.
•
Statistics Canada 2021 Census demographics at the forward sortation area (FSA) level.
•
Fraser Institute school ratings.
First Access Signals does not buy data from credit bureaus. The Service does not use any data it is not authorized to use under the VOW Datafeed Agreement.
How we measure accuracy
This is the part that makes First Access Signals different from other Canadian home-value tools.
For the Market Signal (a probability)
First Access Signals measures accuracy with the Brier score and the Brier Skill Score (BSS).
In plain English:
•
The Brier score measures how far off the model's percentage predictions are on average. Lower is better. A perfect score is 0.0; a score of 0.25 is the score of a model that always says "50 percent."
•
The Brier Skill Score (BSS) is the percentage of prediction noise the model removes compared to a model that always says "yes" (the base rate of selling). A BSS of 0 percent means the model is no better than the base rate; a BSS of 100 percent means perfect predictions. The published floor is 12 percent.
•
First Access Signals also publishes reliability, which checks that when the model says "60 to 70 percent chance of selling," the actual rate falls in that band. The target is a maximum deviation of 10 percentage points across all probability bands.
Current published values (90-day Market Signal):
•
Brier score: 0.1771 (HEALTHY threshold ≤ 0.20).
•
Brier Skill Score and reliability max-deviation: refreshed weekly after retrain.
Current published values (365-day Market Signal):
•
Brier score: 0.1885 (HEALTHY threshold ≤ 0.20).
•
Brier Skill Score and reliability max-deviation: refreshed weekly after retrain.
For the AVM estimate (a dollar value)
First Access Signals measures accuracy with the Median Absolute Percentage Error (MDAPE) and a conformal prediction interval.
In plain English:
•
MDAPE is the median percentage by which the home-value estimate differs from the actual sale price for closed sales not included in training. Lower is better. A 5 percent MDAPE means half of the estimates are within 5 percent of the actual sale price.
•
A conformal prediction interval is a 90 percent confidence range around our estimate. In 9 out of 10 cases, the actual sale price falls inside the band. Conformal intervals are the gold standard in machine-learning uncertainty quantification because they hold their accuracy guarantee in real-world conditions.
Current published values (Sale AVM):
•
Test MDAPE: 3.34% (HEALTHY threshold ≤ 4.00%).
•
90 percent conformal prediction interval empirical coverage (target 90 to 92%): refreshed weekly after retrain.
Current published values (Lease AVM):
•
90 percent conformal prediction interval empirical coverage (target 90 to 92%): refreshed weekly after retrain.
Where the model performs well, and where it performs less well
First Access Signals publishes both ends of the accuracy distribution by neighbourhood. Honesty about weakness is part of the competitive moat. The lists are refreshed every Monday after the weekly retrain.
PENDING WEEKLY RETRAIN
Top 10 best-performing FSAs and bottom 10 worst-performing FSAs publish on this page after the next weekly retrain export wires through. If your home is in a segment where the model underperforms, the app displays a low-confidence badge on the result. The recommendation is to treat the estimate as a starting point for a conversation with Keith rather than a precise number.
How the model remains honest over time
A real estate model that was built on 2024 data and never re-checked would drift further from reality every month. First Access Signals employs a five-layer drift management strategy:
1.
Leading-indicator nowcasts. First Access Signals uses Bank of Canada and Statistics Canada releases to anticipate market shifts before they show up in closed sales.
2.
HPI-deflated targets. Where it improves accuracy, sale prices are deflated by the CREA Home Price Index so the model learns the property-specific signal, not the macro tide.
3.
Time-decay sample weighting. Recent listings get more weight in training than older ones.
4.
Change-point residual detection. If the model's residuals shift systematically week over week, the system flags it for retraining.
5.
Weekly retrain cadence. Every Monday morning the model retrains on the latest 24-month window, automatically.
If published accuracy values fall below the HEALTHY thresholds in two consecutive weekly retrains, First Access Signals publishes a notice on this page describing what changed and what actions are being taken.
VOW compliance and the bona-fide-consumer commitment
To use the parts of First Access Signals that show specific listing details, you must be a registered user, which means you have signed in with Apple or Google and acknowledged the following statement, required verbatim by the TRREB VOW Datafeed Agreement §6.3(k):
The information provided herein must only be used by consumers that have a bona fide interest in the purchase, sale, or lease of real estate and may not be used for any commercial purpose or any other purpose.
If you are not a bona-fide consumer (for example, you are a researcher building a competing data product), please do not register.
The brokerage of record is THE AGENCY, and the designated Realtor for First Access Signals is Keith Godding.
Plain-English glossary
Brier score
How far off the model's percentage predictions are on average. Lower is better.
Brier Skill Score (BSS)
The percentage of prediction noise the model removes compared with a model that always says yes. The published floor is 12 percent.
Reliability max-deviation
When we say 60 to 70 percent chance of selling, the actual rate falls within 10 percentage points of that band.
MDAPE
The median percentage by which our home-value estimate differs from the actual sale price. Lower is better.
Conformal prediction interval
A 90 percent confidence range around our home-value estimate. In 9 out of 10 cases the actual sale price falls inside the band.
Walk-forward backtest
Predicting each calendar quarter using only data from before that quarter. If our error grows quarter over quarter, we know the model is going stale and we retrain.
Drift detector
If the model's accuracy is drifting more than five percent per quarter, we automatically retrain.
Bona-fide consumer
A real person looking to buy, sell, or lease a home, not a researcher or a competitor scraping data.
VOW (Virtual Office Website)
TRREB's data-feed program that lets registered Realtors and their clients see listing details.
MLS attribution
Every listing card on our app shows the listing brokerage's name and the MLS source.
FSA
Forward sortation area. The first three characters of a Canadian postal code (for example, M5V).
Contact
This page is the canonical methodology disclosure for First Access Signals. It is updated quarterly or when the model changes materially.