Spatiotemporal Intelligence · v0 Manhattan

Drag time.
Watch value
move.

ValueAtlas is the prediction platform for real estate value across space and time. Our AI understands how the physical world evolves — starting with the value of every property in every neighborhood, past, present, and twenty years forward.

Observations
1.4B
NY Properties
846K
Horizon
2036
Live · Manhattan · Value Velocity
40.71°N — 40.88°N / -74.02°W — -73.91°W
2020
Observed
Δ value vs 2016
−20%+50%+150%+300%+
60 fps · GPU · WebGL
2016202020242026203020342036

Fig. 01 — Manhattan value velocity, 2016 → 2036. Historical (2016–2026) fitted from recorded deeds & MLS. Post-2026 is model forecast at P50 confidence.

Harlem ▲ +214% since 2016Washington Hts ▲ +268%Midtown ▲ +38%FiDi ▼ −6% then ▲ +91%Lower East Side ▲ +182%Inwood ▲ +301% forecast 2032Model calibration 91.2% within P80Next drop ATL · Q1 2027
Harlem ▲ +214% since 2016Washington Hts ▲ +268%Midtown ▲ +38%FiDi ▼ −6% then ▲ +91%Lower East Side ▲ +182%Inwood ▲ +301% forecast 2032Model calibration 91.2% within P80Next drop ATL · Q1 2027
The platform

An operating system for the physical world, beginning with real estate.

01Principle

The map is the product.

Not a dashboard with a map attached. The map is the primary canvas, time is the primary control, and AI is the primary guide. Every insight is geocoded, timestamped, and connected.

02Principle

Motion is the interface.

Weather taught the world to read patterns by watching them move. Value has waves, fronts, reversals and diffusion — we render them the same way, at 60fps.

03Principle

Every forecast is remembered.

Predictions are written to an immutable registry with model version, inputs, horizon, and confidence. As reality arrives, calibration becomes a proprietary dataset.

04Principle

Confidence over certainty.

Forecasts always shown as ranges. Historical facts visually separated from modeled estimates. Data gaps lower confidence visibly. Trust is the product beneath the product.

The science

AI has learned language. It must now learn place.

ValueAtlas · Spatiotemporal Model v3
L01 · Input
Observations
Deeds, listings, permits, transit, hazards, satellite — every event tied to (lat, lon, t).
L01 · InputObservations
Live model trace·Fwd pass on 12,864 real estate parcels·Green = positive contribution·Red = negative contribution
L01
Observation layer
1.4B geocoded events · deeds, listings, permits, transit, hazards, employers, retail, satellite
L02
Entity resolution
846K real estate parcels reconciled across 12 authorities with time-aware edges & source scoring
L03
Property knowledge graph
Every property connected to schools, transit, zoning, comparables, hazards — validity-dated
L04
Event engine
Rumored → proposed → approved → built → operational. Markets react before completion.
L05
Spatiotemporal ensemble
Repeat-sales · hedonic · geo-weighted · graph neural nets · temporal transformers
L06
Forecast registry
Immutable log of every prediction. Calibration measured by market, horizon, and regime.

The core unit is not a document or a prompt. It is an observation tied to coordinates and time. The platform answers four linked questions — what happened, where, when, and what likely happens next.

Language models are useful as interfaces and reasoners. The durable intelligence comes from structured spatial relationships, temporal sequences, event histories, and observed outcomes.

Backtest · NY residential · 2019–2026
91.2%
P80 hit rate
4.6%
Median 12-mo MAPE
0.87
Directional AUC
The atlas

Depth in a few cities beats shallow coverage of many.

NYC · Live model
live
New York — Manhattan
Value velocity · 2016 → 2036+184% Harlem
Coverage
98%
BRK · Live model
live
New York — Brooklyn
Rent yield surface · rolling 12mo+141% Bushwick
Coverage
94%
SFO · Live model
beta
San Francisco Bay
Employer-linked demand · 2020 → 2030−12% SoMa office ring
Coverage
78%
LAX · Live model
beta
Greater Los Angeles
Fire & insurance-adjusted price+63% Inland Empire
Coverage
71%
CodeQueued marketStatus
MIASouth Floridaqueued
ATLAtlanta Metroqueued
AUSAustin — Central TXqueued
SEAPuget Soundqueued
The trajectory

From one map to the operating system of place.

  1. 2026

    Manhattan launch

    Consumer product, animated heatmap, property + neighborhood cards, forecast registry v1.

  2. 2027

    Outer boroughs + Bay Area

    Rental demand & affordability layers. Enterprise portfolio monitoring beta.

  3. 2028

    National residential

    Top-25 MSAs. Insurance & climate exposure layers. Embedded maps for partners.

  4. 2030

    Beyond residential

    Commercial, retail demand, mobility. The Bloomberg Terminal for the physical world begins to take shape.

Early access

Understand any place
in five seconds.

We're onboarding a small group of homeowners, investors and enterprise partners for the Manhattan launch. Consumer waitlist opens in weeks.