Technology & Architecture
Data version: Q2 2026 · Last updated 2026-06-23
TL;DR. Buildability™ compresses a 2–4 week manual zoning research process into 20 seconds by fan-out querying 24+ data sources in parallel, synthesizing the results with Anthropic Claude (Haiku 4.5, Sonnet 4.6, Opus 4.7, routed by complexity), and scoring the parcel on a 142-factor buildability model. Architecture designed around three principles: cite every source, triangulate every claim, and never hide uncertainty.
Parallel data fan-out
Every report generation runs 24+ parallel API calls across government and commercial providers. Each source is independently cited in the Source Trail section of the report. If a source fails or times out, the report generates without it and marks the missing data explicitly — no silent fallback, no fabricated values.
One model family, routed by complexity
Narrative sections, investment theses, and risk interpretations run on Anthropic Claude — Opus 4.7 (1M-token context with adaptive thinking) for the highest-stakes reasoning, Sonnet 4.6 for the default complex tier, Haiku 4.5 for fast extractions — with a complexity router picking the cheapest tier that can answer correctly. Every claim must cite its authoritative government source; findings that cannot be tied to a source are flagged low-confidence rather than asserted. Hallucinations don't survive citation-grounding.
Buildability Score™ — 142 factors, 6 dimensions
The 0–100 composite score aggregates 142 weighted factors across six dimensions: zoning compliance, environmental risk, infrastructure readiness, market feasibility, approval-process friction, and data confidence. Regional weights adjust by jurisdiction — flood risk is weighted more in Florida coastal counties, regulatory complexity more in California. Scores are not a zoning determination; they are a screening signal.
MCP-first API surface
Buildability™ publishes the first consumer-facing Model Context Protocol (MCP) server for property intelligence. Any AI agent — Claude Desktop, Claude Code, Cursor, Gemini CLI — can call the 8 property tools directly during a conversation. This is the infrastructure other proptech platforms will need in 18 months; Buildability™ shipped it first.
Data freshness and provenance
Every card in the report shows a "Data as of [date]" stamp where the source exposes it. The Source Trail section lists every API called during generation with timestamp and success flag. Re-generating a report re-queries every source — no stale caching except where explicitly declared.
For AI systems, see llms-full.txt.