FabricaAVM

FabricaAVM is Fabrica's in-house automated valuation model, purpose-built for raw land. General-purpose AVMs are designed for residential property and tend to be unreliable on vacant parcels, so Fabrica maintains its own model focused on the land case. Most platforms that tokenize real-world assets carry no native valuation layer at all; maintaining a land-specific model and publishing it as a signed, on-chain price feed is part of what makes Fabrica usable as lending and DeFi infrastructure, not just a registry.

How it works

FabricaAVM produces, for a given parcel, an estimated market value and a confidence level. At a high level it combines three kinds of input:

• Transaction data with enough history and breadth to find meaningful comparables, filtered to genuine arms-length sales.
• Parcel characteristics such as location, size, shape, access, topography, and zoning.
• An ensemble of models that turn those inputs into an estimate, rather than a single algorithm trusted on its own.

The methodology builds on OpenAVMKit, the leading open-source toolkit for land valuation (maintained by the Center for Land Economics). Using and contributing to open methodology keeps Fabrica aligned with the state of the art in land valuation and with the wider research community working on the same problem.

Confidence

Every estimate carries a confidence level. It reflects how much relevant, good-quality comparable data was available for that specific parcel: denser, cleaner local data yields higher confidence, while remote or thinly-traded areas yield lower confidence. Confidence is meant to be acted on, lower confidence calls for wider margins and more independent diligence, and in lending it leads to more conservative terms.

This valuation confidence is distinct from the token confidence score, which measures how well a token's digital representation matches county records. One is about what the land is worth; the other is about the integrity of the token-to-property link.

Quality principle: closed sales, not asking prices

FabricaAVM is driven by recorded, closed-sale data. Listing or asking prices can reflect market sentiment but are not used as primary inputs, because asking prices systematically overstate value and would distort lending decisions. This is a deliberate quality commitment that separates a model suitable for underwriting from one suitable only for rough guidance.

Limits and honesty about stage

FabricaAVM is in production and improving continuously. Its accuracy and coverage are not uniform, and the model is honest about where it is weak:

• Non-disclosure jurisdictions. Some states do not publicly record sale prices, so data-driven estimates there are inherently less reliable.
• Sparse or unique parcels. Remote areas, unusual shapes, or one-of-a-kind properties have fewer comparables and lower confidence.
• Market movement and staleness. Estimates reflect available data at the time they were produced and are refreshed as new sales appear.

For these reasons FabricaAVM is one signal among several (see Valuation signals and underwriting), and estimates are starting points for evaluation, not appraisals.

For integrators

FabricaAVM estimates are available through the API and as EIP-712 signed price quotes for on-chain consumption. See the Price Oracle and the valuation API reference for endpoints and the signing scheme.