Every listing, graded — and auditable.
Short version: every listing on marocain.investments ships with four signals — a Window-View Score, a Structural Quality score, a WC 2030 Catalyst score and an M-Value AVM. None of them are magic; all of them are auditable. Each carries the methodology label that produced it and a 0–1 confidence value that drops when the data is thin. Below: exactly what each score measures, how to read the bands — and, just as importantly, what each score is NOT.
Four scores on every listing.
Each grades a different thing, on a different method, with its own honesty clause. Read them together — none of them is the whole story on its own.
Window-View Score
A vision model scores the actual view from the primary living spaces — ocean, medina, skyline or garden — from the listing photos. Marketing-staged dusk shots don't earn a higher score than the underlying property warrants.
Structural Quality
Build era, finishes and visible condition, read from the photos. 90+ is a 2020+ build with designer or branded finishes; 70–89 is a 2010–2019 build or a restored heritage envelope; below 50 is cosmetic-tier needs-work.
WC 2030 Catalyst
A pure distance metric: how far the property is from the nearest WC 2030 host venue, the nearest TGV station and the nearest international airport. Closer = higher.
M-Value AVM
A median-of-comparables valuation built from the catalogue itself: listings in the same district + typology + ±20% surface, median price per m², applied to the target's surface. The band is 1.5σ on the comp dispersion.
The rubric, in full.
What each score measures, how to read the bands, and what it explicitly is NOT. This is an internal grading rubric — not an external market dataset.
| Score | What it measures | Bands / how to read it | What it is NOT | Range |
|---|---|---|---|---|
| Window-View Score | A vision model scores the actual view from the primary living spaces — ocean, medina, skyline or garden — from the listing photos. Marketing-staged dusk shots don't earn a higher score than the underlying property warrants. | 90+ = dominant Atlantic / Mediterranean frontage from multiple rooms · 70–89 = a substantial sea or skyline glimpse · 50–69 = garden or courtyard view. | Not a guarantee the view is protected — a vision read of today's photos, not a planning-permission check on what could be built across the street. | 0–100 |
MethodologyWhat it measures: vision-v1 (gemini-2.5-flash) · 2026.06 | ||||
| Structural Quality | Build era, finishes and visible condition, read from the photos. 90+ is a 2020+ build with designer or branded finishes; 70–89 is a 2010–2019 build or a restored heritage envelope; below 50 is cosmetic-tier needs-work. | 90+ = recent build, designer / branded finishes · 70–89 = solid 2010s build or restored heritage · below 50 = visible work needed. | Not a structural survey or a notary inspection. It is a triage signal — which listings warrant the flight in — not a verdict on the foundations. | 0–100 |
MethodologyWhat it measures: condition-v1 (gemini-2.5-flash) · 2026.06 | ||||
| WC 2030 Catalyst | A pure distance metric: how far the property is from the nearest WC 2030 host venue, the nearest TGV station and the nearest international airport. Closer = higher. | Higher = closer to a host venue, TGV station and airport. Two listings with the same score can still diverge on micro-market scarcity (Marshan vs Maarif). | It quantifies catalyst exposure — it does NOT predict appreciation. Proximity is not a price forecast. | 0–100 |
MethodologyWhat it measures: catalyst-v1 (deterministic) · 2026.06 | ||||
| M-Value AVM | A median-of-comparables valuation built from the catalogue itself: listings in the same district + typology + ±20% surface, median price per m², applied to the target's surface. The band is 1.5σ on the comp dispersion. | Renders only with ≥3 comparables — under that the chip simply doesn't show. The vs-asking delta is most useful past >8% above or below the comp median. | Decision-support, not a formal valuation. We would rather show nothing than a noisy number on a thin comp pool. | estimate + band |
MethodologyWhat it measures: m-value-v1 (median-of-comparables) · 2026.06 | ||||
Methodology + confidence.
No black box. Every score tells you how it was produced and how sure it is — so you can weigh it, not just trust it.
A methodology label on every score
Each score ships with the exact method that produced it — e.g. vision-v1 (gemini-2.5-flash) for the view, condition-v1 for finishes, a deterministic distance model for the WC 2030 catalyst, and median-of-comparables for the M-Value. If a score reads v1.deterministic, it was NOT model-graded.
A 0–1 confidence value that drops with thin data
Confidence falls when the source data is thin — fewer than three photos, missing coordinates, missing surface. A low confidence value is the model being honest about what it could and couldn't tell, not a hidden penalty.
Silence over a noisy number
When a signal can't be produced honestly — a comp pool under three rows, for instance — the chip simply doesn't render. We would rather show nothing than a number we can't stand behind.
What the scores tell you — and what they don't.
The honest ledger. The scores are a fast, auditable triage layer; they are not a survey, a valuation, or advice — and the page says so plainly.
- Which listings warrant the flight in — a fast, consistent triage across the whole catalogue.
- How a property reads on view and finishes from its own photos, on the same rubric every time.
- Its catalyst exposure and a comp-based sense of value — each with the confidence to weight it by.
- Not a structural survey or a notary inspection — Structural Quality is a photo-based triage signal, full stop.
- Not a formal valuation — the M-Value AVM is decision-support, and renders only with ≥3 comparables.
- Not investment advice, and not a price forecast — the WC 2030 Catalyst quantifies exposure, not appreciation.
A grade is only worth as much as the method behind it. So every score on marocain.investments shows its working — the model that produced it, the band you're reading, and how confident it is given the data. Use the scores to decide which listings warrant a closer look; use a surveyor, a notary and an agent for the things a photo-read can't settle. That's the whole contract: fast and auditable, never pretending to be certified.
Decision-support, not a certified appraisal or investment advice. The M-Value AVM renders only with ≥3 comparables; confidence drops on thin data.
How the grading works — straight answers.
AI listing grading on marocain.investments is an internal rubric that scores every property on view, structural quality and WC 2030 catalyst exposure, plus a median-of-comparables M-Value — each with a methodology label and a 0–1 confidence value, as decision-support rather than a certified appraisal.
How does marocain.investments grade a listing?
Every listing ships with four signals: a Window-View Score (0–100, a vision model on the actual view), a Structural Quality score (0–100, build era and finishes read from the photos), a WC 2030 Catalyst score (0–100, distance to the nearest host venue, TGV station and airport) and an M-Value AVM — a median-of-comparables valuation. Each score carries a methodology label and a 0–1 confidence value, so the read is auditable rather than a black box.
Does the AI score replace a structural survey or a valuation?
No. The Structural Quality score is a photo-based triage signal — which listings warrant a closer look — not a substitute for a structural survey or a notary inspection. The M-Value AVM is decision-support, not a formal valuation. Both are explicitly stated as what they are NOT on every listing.
What does the WC 2030 Catalyst score actually tell me?
It is a pure distance metric — how close the property is to the nearest WC 2030 host venue, TGV station and international airport. A higher score means more catalyst exposure. It does NOT predict appreciation: two listings with the same score can diverge on micro-market scarcity, so treat it as exposure, not a price forecast.
Why doesn't every listing show an M-Value?
The M-Value AVM only renders when there are at least three genuine comparables in the same district, typology and ±20% surface band. Below that threshold the comp pool is too thin to be honest, so the chip simply doesn't show — we would rather display nothing than a noisy number.
What is the confidence value next to a score?
Each score ships with a methodology label (for example vision-v1 (gemini-2.5-flash)) and a 0–1 confidence value. Confidence drops when the source data is thin — fewer than three photos, missing coordinates, missing surface. The number you see is honest about what the model could and couldn't tell from the available data.