AVM disclosure
Automated valuation disclosure
Section titled “Automated valuation disclosure”Every valuation this system emits — on cards, property pages, API responses, and MCP tool outputs — is the product of an automated valuation model (AVM), and ships with this disclosure. This page is the canonical wording; surfaces may shorten it but never below the API’s own disclosure line, and never in a way that weakens it.
The canonical block
Section titled “The canonical block”This is an automated estimate, not an appraisal. It was produced by a statistical model with published, per-segment error rates — see the live accuracy scorecard for the current numbers by property type and county (accuracy). It has not been prepared by a licensed or certified appraiser and is not a substitute for an appraisal, a comparative market analysis by a licensed professional, or an inspection. Method: {tier} · Confidence: {confidence}/100 · As of: {as_of} · Model: {model_version}. Prohibited uses: no output may be used for tenant screening, credit, employment, or insurance decisions, or any purpose governed by the Fair Credit Reporting Act. Verify independently before spending money.
Historical baseline (v0, superseded by the per-segment scorecard above): median ARV miss 20.1%, within ±20% only about half the time, bias +4.7% hot — measured against 744 arms-length 2–4 unit resales, Logan Square/Avondale 2022–24. Kept here for continuity; it is no longer the number a live response points to.
What the placeholders mean
Section titled “What the placeholders mean”- {tier} — which code path produced the evidence, always disclosed (valuation):
renovated comps/Lake GIS comps— a gated comp set with a calibrated range;hedonic_v1— a transparent range-only cross-check/model;area context— a separately labeled market range that is never property ARV; orsuppressed— no property-specific evidence cleared its gates. Range, area, and suppressed states never produce exact offers or return math. - {confidence} — the 0–100 confidence printed with the number. Below the floor, the number is suppressed on every surface. Calibration is published per segment and honestly: strong in dense suburban comps, weak where comps are thin (accuracy).
- {as_of} — the artifact calculation date. Individual source vintages are carried separately; complete oldest-input dating is still being remediated.
Where it appears
Section titled “Where it appears”| Surface | Form |
|---|---|
| API valuation responses | disclosure field (already live — this page extends it, never replaces it) |
MCP tools (deep_underwrite_asset, get_comps, search_investment_inventory, and their deprecated underwrite/comps/search_deals aliases) |
disclosure string in every valuation payload |
| App cards | short form: “Estimate, not an appraisal — ±20% screen · how we know” linking here |
| Property page | full block above the fold with the number, not in a footer |
| Guides / social cards / reports | any quoted number carries the short form + link |
Why this page exists
Section titled “Why this page exists”Emerging AVM quality-control expectations (interagency AVM rule, 12 CFR pt. 34 subpart H et al.) push toward exactly what this system already does by conviction: published error, confidence policies, and no-eligibility-use statements. This block is the single place that wording lives so it stays identical everywhere (one number per concept — one disclosure per number).
This page is operator-written and not lawyer-reviewed; counsel review is a launch gate (G5.1, specs/launch-gates.md in the repository).