Summary
IRIS — Residential CMA — regenerated from canonical/guides/tools/iris.md. Do not hand-edit this page — update the canonical file and re-project. Status: beta. Primary domain: iris.stoicagent.com. Pipeline role: residential valuation + photo/condition scoring.
Content
Purpose
Desktop-appraisal accuracy for residential valuation. Three layers:
- CMA — comp pulls + weighting
- LENS — photo condition scoring
- ARGUS — 8-dimension risk analysis (also referenced in Athena — overlap to resolve)
Current state
- Status: Agent beta
- Accuracy: 20% variance (desktop-appraisal band)
- Target: 10% variance (field-appraisal band)
- Replaces: Cloud CMA (~$50/month) + Beacon AI (partial)
Where it lives
- Code:
Tools/IRIS/ - Deployed: iris.stoicagent.com (Netlify:
artemis-residential-cma.netlify.app) - Skill:
.claude/skills/cma/(if installed) — one-click CMA generator with MARIS comp filtering - Env vars: set per Infra Hardening April 2026
Integrations
- MARIS MLS — comp source (130K+ closed sales, weekly updates)
- ATTOM — property AVM, tax, owner, schools, demographics
- Regrid — parcel data
- Photo analysis layer (LENS) — condition scoring
- Risk engine (ARGUS) — shared with Athena (source of truth TBD)
Related
- Inventory row:
canonical/business-state/tool-stack-inventory.md→ "Iris" - Athena briefing:
canonical/guides/tools/athena.md(Argus overlap) - Memory: Infra Hardening April 2026
- Fair Housing compliance applies to all output (see vault CLAUDE.md rule 9)
Typical development touchpoints
- Comp filtering rules (FSBO exclusion, DOM outliers, school district, brokerage concentration)
- Weighting algorithm tuning (distance, condition, age, size, DOM)
- LENS photo scoring — ML model or rules-based?
- ARGUS risk dimensions — which 8, and how weighted
- Accuracy tracking — how do we measure 20% → 10%?
Active WIP location
working/drafts/tools/iris/ (create as needed during active development)
Open questions / gaps
- ARGUS overlap with Athena — which tool is the source of truth for risk scoring?
- Accuracy milestone plan — concrete path from 20% to 10% variance not yet documented
- Agent beta feedback capture — where does bug/suggestion data land?
- Fair Housing compliance in all LENS/ARGUS outputs