Introduction · Methodology v1.0
How MARA works
What MARA is, what it calculates itself, and what it asks an AI model to do.
MARA (Montreal AI Realtor Assistant) is an analysis tool for the Greater Montreal Area. It combines three different things, and it is worth understanding which is which, because they have very different reliability.
| Layer | What it does | Reliability |
|---|---|---|
| Your data | The listing facts you enter or import: price, taxes, condo fees, size, year built. | As reliable as the source it came from. MARA labels the source of every field. |
| Application logic | Deterministic maths: weighted comparison scores, ranking, renovation subtotals, contingency, totals, all currency formatting. | Exact and repeatable. The same inputs always produce the same output. |
| AI analysis layer | A general-purpose AI model, driven by MARA's Montreal-specific prompts, that reads the property information and produces category scores, cost estimates, risks and written explanations. | Interpretive. It can be wrong. Everything it produces is labelled as an estimate. |
The rule MARA follows
If a number can be calculated, the application calculates it. The AI is asked to assess and to explain — it is never asked to do the arithmetic that decides a ranking.
Scoring methodology version
All scores in MARA are produced under Scoring Methodology v1.0. Saved analyses record the methodology version they were generated with, so older reports stay interpretable if the methodology changes later.