The map is public record, assembled

One national dataset instead of three thousand websites

Property ownership in the United States is recorded county by county. Each county assessor values the land and improvements for tax purposes, each county recorder registers the deeds that transfer them, and each publishes some version of that record, on some schedule, in some format. The information is public, and it is also scattered across thousands of separate systems with no common field names, no common geometry, and no common idea of what a parcel identifier looks like.

Regrid does the reconciliation. It collects those county records, normalizes them into one schema, attaches the surveyed boundary geometry, and assigns each parcel a stable identifier that Plotbook can cache against. Plotbook consumes that dataset rather than rebuilding it, which is a deliberate choice: parcel aggregation is a full business on its own, and doing it badly is worse than not doing it.

What arrives per parcel is the boundary polygon, the name on the deed, the address the tax bill is mailed to, the county's assessed value, the use description, the parcel number, zoning, property class, and a coordinate. Those fields are the raw material for everything else in the product. The map colors by the assessed value. The owner label prints the deed name. The research agent starts from the deed name and the mailing address. None of it is inferred.

• Boundary geometry, so parcels are shapes rather than pins
• Deed owner name, printed on the parcel and used as the research starting point
• Assessed value, which drives the color bands and the value filter
• Mailing address, which is how absentee ownership is detected
• Use description, parcel number, zoning, and property class

Vector tiles through a proxy that holds the key

Parcels are delivered as vector tiles rather than pre-rendered images. A tile is a compact packet of geometry and attributes, so the browser receives the shapes and the data together and draws them itself with WebGL through Mapbox GL. That is why the boundaries stay sharp at any zoom, why hovering a parcel can highlight it instantly, and why changing the value filter repaints the view without another round trip to a server.

Those tiles do not come straight from the tile vendor. They pass through tiles.plotbook.io, a Cloudflare Worker Plotbook operates, and the credential for the upstream tileserver lives only in that worker's secrets. A browser session never holds a key that could be lifted and reused, and the worker caches tiles at the edge so a market that somebody already explored loads from a nearby datacenter rather than from origin.

Zoom controls how much the map is willing to tell you at once. Parcels start drawing at zoom 14, roughly a neighborhood. Every parcel gets its deed owner label at zoom 16, close enough that the names have room to sit inside their boundaries. Between those two levels, an overlay labels only the large holdings early and sorts by value, so when two labels compete for the same space the more valuable parcel keeps its name.

Six bands, so a region reads in one glance

Each parcel is filled by its assessed value in one of six bands: unknown, under $500K, $500K to $1M, $1M to $5M, $5M to $10M, and over $10M. The bands are deliberately coarse. The purpose is not to read a number off a color, it is to make the shape of a market visible from altitude, so waterfront, the ridge behind town, and the one street where the older estates sit separate themselves before you have clicked anything.

The same six bands drive the filter. Five quick presets narrow the view to a tier, and filtered parcels are painted out rather than removed: they keep a dimmed outline and stay clickable, so you can see that the filter is hiding something and still open it. Both the filter and the legend state persist between sessions, because most people work one market repeatedly and should not have to rebuild their view every morning. A palette control turns the color off entirely when you want to look at the ground instead.

Three basemaps sit under all of it, Mapbox Satellite plus outdoor and hybrid styles from MapTiler, which matters more than it sounds: satellite imagery is how you tell a large lot from a large house, and a hybrid view is how you keep street names while doing it.

• Unknown, under $500K, $500K to $1M, $1M to $5M, $5M to $10M, over $10M
• Five filter presets matching the bands, toggleable and persisted
• Filtered-out parcels stay visible as dimmed outlines and remain clickable
• Colors can be switched off entirely from the legend

232 hotspots that are curated, not computed

Zoomed out, parcel geometry is useless: at state scale there is nothing to see but a solid wash. So above the parcel layer sits a separate dataset of 232 hand-curated high-net-worth ZIP codes, the Fisher Islands and Athertons and Palm Beaches, rendered as a heatmap at country and state zoom, as clickable clusters through the middle zooms, and as individual pulsing badges with area names as you come down toward a neighborhood. A side widget lists wealthy areas near wherever you are currently looking.

It is worth being precise about what that layer is, because this is exactly the sort of thing the category tends to dress up. It is a curated list of postal codes with density tiers, not a computed measure of household wealth, and no figure shown on it is derived from the individual households inside those ZIPs. Its job is orientation: it tells you where to zoom in when you are working an unfamiliar state. Everything below it, the parcels and the names, is the actual record.

What a single parcel will tell you

Clicking a parcel opens a drawer with the assessed value, the address, the name on the deed, and the use type. That is the county's view of the property and it is where public record stops. Contact rows appear in the drawer but stay masked, because a phone number is not a county record and has to be retrieved from somewhere that charges for it.

From there the drawer offers the two ways forward, and they are different in kind. An instant owner lookup queries residential records for the people at that address, which is the right move when the deed names a person and you want their household in a few seconds. Deep research hands the property to the AI agent, which is the right move when the deed names an LLC, a trust, or a partnership and there is no person to look up yet.

Once a profile exists for that property it links back: the parcel keeps a green marker on the map, and the drawer shows the owner's name, wealth estimate, and unmasked contacts instead of the research buttons. The map stops being a lookup surface and starts being a record of what you have already worked.

Cached against a stable id

Tiles are streamed and not stored. Parcel detail is different: when you open a property, its record is written to Plotbook's Postgres database keyed by Regrid's stable parcel identifier, along with the full original response, so a field that is not surfaced in the UI today can be surfaced later without re-fetching anything.

Each cached parcel also stores its location as a PostGIS point with a spatial index. That is what makes the map's saved layer possible: as you pan, Plotbook can ask which of your saved profiles have properties inside the current viewport and return them as markers, in one indexed query rather than by scanning your whole prospect book.

The practical effect is that repeat work is fast and does not re-spend anything. Reopening a parcel you have already opened serves from cache. Re-running research on a property you have already researched returns the profile you already own rather than starting a second billable run.

Assessment cycles, not a live feed

Property data moves at the speed of county government, and that speed varies. Assessors reassess on their own cycles, some annually and some every few years. Deeds appear in the record after they are filed, which happens after closing, not at it. Regrid refreshes counties on a rolling basis and Plotbook shows what Regrid has. Nobody in that chain is streaming, and any product that claims a live view of ownership is describing something that does not exist.

This matters most for the case people care about most: a property that recently changed hands. A sale shows up once the deed is recorded and the county publishes it, which is a matter of weeks in most jurisdictions rather than days. If your workflow depends on reaching new owners within days of closing, that is a genuine gap and worth knowing before you build around it.

The other freshness question is assessed value against market value. An assessment is a tax figure produced on a cycle, and in a moving market it will lag, sometimes by a wide margin, in either direction. It remains an excellent relative signal: within one county, a parcel assessed at $6M and one assessed at $400K are not in the same conversation regardless of what either would actually fetch. Read it as a tier, not as a price.

Deeds are public by design

Land ownership is recorded publicly because the system depends on it being verifiable. A deed only protects a buyer if anyone can check who holds title, which is why assessment rolls and recorder indexes are open to any member of the public. Looking up who owns a property requires no consent, and building a map of it changes the convenience, not the legal status.

What Plotbook adds to public record is aggregation and visualization. What it does not do is scrape private information. Contact details are a separate layer entirely: they come from licensed data providers under their terms, they are masked until you deliberately spend a credit to retrieve them, and they are never part of the parcel record itself. That separation is intentional, and it is the line worth understanding when a compliance team asks what the map actually is.