Methodology

A Property Intelligence Report answers one practical buyer question: is this listing priced well enough to pursue, and where should negotiation begin?

The report combines active inventory, municipal signal intelligence, neighborhood pulse data, and seller-flexibility signals into one decision surface. It is designed for asking-price discipline, not title diligence or legal underwriting.

The primary estimate is a k-NN weighted log-mean over the most similar active listings. Log-space averaging reduces the ability of a single outlier to drag the estimate around.

When the selected cohort is broad enough, the system can also use a log-linear OLS regression against size. If the comp base is too weak for either method, the report falls back to a broad active-cohort median with a visibly wider interval.

Confidence intervals are directional asking-price intervals, not appraisals. The weighted method uses bootstrap resampling; the regression method uses residual spread.

The generator starts narrow: same zone, same property type, and roughly similar size. If fewer than 10 active comps qualify, it relaxes size, then adjacent zones, then same-family property types.

Every relaxation is recorded in the report metadata and caps the confidence label. A relaxed cohort is better than pretending the data is tight, but the report makes that tradeoff visible.

Negotiation starts from the comp-adjusted fair value, then applies a factor model: base discount 5%, staleness up to 10% at 8% per year, observed price-drop history, and zone hotness.

The total discount is capped at 18% so the model does not invent unrealistic lowball guidance from noisy inputs.

The output is a suggested opening offer, probable close anchor, and walk-away line. Those are negotiation anchors, not guaranteed transaction prices.

The report pulls the highest-materiality municipal signals for the listing zone and property type. Signals are ranked for buyer relevance instead of shown as a raw database dump.

When deeper signal narratives exist, the report prefers those synthesized briefs; otherwise it falls back to the direct implications stored on the signal record.

The verdict is a bounded exposure read, not a legal opinion. It scores in-motion planning, zoning, environmental, and land-use signals tagged to the parcel zone or municipality-wide.

Negative signals carry the most weight, mixed signals carry partial weight, and positive signals still count lightly because rule changes can create execution uncertainty. Administrative commitments count more than funded or physically executing work because the shape of the final rule is less settled.

First-cut thresholds are HIGH at 8 or above, MEDIUM at 4 or above, LOW above 0, and UNKNOWN when no in-motion regulatory signal is tagged to the zone. The founder will tune those thresholds as more real reports are reviewed.

Pulse data tracks establishments, category composition, openings, closures, reviews, and phase movement at the neighborhood level.

This layer helps explain why two zones with similar listing comps can have different forward-looking risk. A maturing food-and-beverage layer, rising wellness mix, or elevated closure rate can change the context around the same asking price.

When a report can match the target listing to coordinates, it filters the public places intelligence file around the parcel instead of sending the full dataset to the browser.

The written list ranks operational establishments within 500 m using rating multiplied by ln(1 + review count), with a 20-review floor so one five-star review does not outrank a proven neighborhood anchor.

The interactive map uses the same score and distance helper over a 1.5 km radius, capped to the highest-quality 200 places so dense zones stay responsive.

The product refuses unsupported zone/type combinations before checkout. The active-inventory floor is 10 comps.

Some zones are excluded entirely until inventory or data quality improves: Sian Ka'an, Soy Tulum, Hotel Zone, Tankah, Tulum 101.

The refusal is intentional. A thin report is more damaging than no report when a buyer is making a high-stakes decision.

HIGH means the active comp base is strong enough and the interval is tight enough to treat the estimate as a real pricing anchor.

MEDIUM means the estimate is useful but should be paired with stronger human judgment, especially on unique properties or fast-moving zones.

LOW means the report found a way to remain honest, but the buyer should treat the numbers as directional and weight the analyst note more heavily.

Listing data comes from the cleaned Tulum inventory pipeline and is filtered to active URLs for the report cohort. Price history is reconstructed from listing snapshot events when available.

Municipal intelligence comes from the internal signal pipeline. Neighborhood Pulse comes from place-level composition snapshots and is refreshed on its own cadence.

The generator does not use closed transaction data today because coverage is too sparse and inconsistent for a broad buyer product.

The report uses asking prices, not verified sale prices. It cannot prove what a seller will accept.

We do not have parcel polygons, title status, ejido status (collectively-owned communal land, with restricted private-sale rights), or legal encumbrances. Land buyers still need local legal diligence before wiring funds.

Coordinate coverage is incomplete. When a listing cannot be matched to an exact in-database URL, proximity scoring and inset maps degrade gracefully instead of being approximated.

Scrape cadence is periodic, so very recent listings or price changes can lag the live market.