April 2026

A labeled, in-house property valuation model. Estimated values surfaced across the application are now attributed to the Fabrica AVM (Automated Valuation Model) with a visible source label and version, and the confidence level is shown alongside the estimate. The model is comparables-based, drawing on recorded sales and parcel characteristics to produce a point estimate with a confidence score, and it is combined with a third-party land-value service so the displayed estimate takes the more conservative of the two. Values below a confidence threshold are withheld, so buyers, sellers, and lenders are not given an estimate the model itself does not stand behind. The same estimates feed the signed price oracle that the lending pool reads. See Property Valuation for how estimates, assessed values, and sale history fit together, and Price Oracle for the signed, machine-readable feed.
Off-ramp moved toward a formal deed redemption. The flow for taking a property back offchain to a recorded deed was built out, including a redemption step that captures the new owner and a wallet-signature authorization for the deed, so selling a property out of the token form follows a clear, auditable path.
AI agents can read Fabrica data through a Model Context Protocol server. The Fabrica MCP server lets any MCP-capable agent (Claude, Claude Code, Cursor, and others) search tokenized properties by state, acreage, confidence score, listing or loan status, and owner; read a property in full, including its legal description, operating agreement, valuation and score breakdown, ownership, loans, and marketplace activity; decode a confidence score digit by digit; and read the lending market, an account's portfolio and credit history, protocol-wide statistics, parcel and county boundaries, activity timelines, and borrowing availability. Static map images of a property or of a whole portfolio followed the next day. Every tool is read-only and draws on the public GraphQL API, so no API key is required. The server runs against Ethereum mainnet or Sepolia by configuration, and on mainnet it instructs agents to tell users about the legal consequences of acquiring a token and to review the trust instrument first. At this point it ran locally from a clone of its repository; hosted endpoints followed in September. See AI agents and MCP.
Agent and machine discoverability. The application published an llms.txt file, a sitemap, and an open crawl policy so AI agents, automated readers, and search engines can navigate the site and understand its content.