Valuation

Model p5v4-2026.07. This is an automated estimate, not an appraisal. The current per-segment calibration artifact permits ranges or suppression only; no segment is approved for an actionable point. Consequently exact offers, profit, NOI, DSCR, cash-on-cash, loan limits, cashflow projections, and property strategy scores are withheld on current range/area/suppressed results.

Canonical subject

Every artifact carries a subject block: listing/geo key, coordinates, tract/county, price and unit provenance, beds, GLA, year, property type, and modeled condition. Unknown facts remain null. A nearest sale may not supply subject GLA, beds, year, neighborhood, or township. The current identity layer is still incomplete: PIN-level resolution and independent assessor reconciliation remain Phase 2 work.

Renovated-comp evidence

Expansion ladder: 0.5 mi/180 days → 0.75 mi/270 days → 1.0 mi/365 days. Coarse year-only records receive documented recency slack; actual sale dates are still a known data gap.
Arms-length and broad property-type gates run first. Renovated/new condition labels are modeled, not inspections. New construction, style, and lot controls remain incomplete.
When subject GLA is known, comps must fall within ±20%; beds must be within one when both sides are known. Log-price-per-square-foot MAD trimming and a 35% adjusted-price dispersion ceiling reject unstable pools.
Three qualifying comps can produce evidence; fewer than five is range-only. At most 12 are retained and eight feed the weighted median.
Time indexing is capped at 8% per year and a total factor of 0.75–1.30.
GLA adjustment currently uses median market $/sf capped at 15% of comp price. That is not yet a calibrated contributory/depreciated adjustment and remains a published limitation.

The empirical comp band is replaced for serving by the segment’s split-conformal 80% interval. The artifact retains included/rejected comps, adjustment fields, evidence count, method, calibration segment, confidence fields, source chain, and versions.

Hedonic and area context

A transparent size/age/location/time hedonic model can cross-check a comp range and widen it when the approaches disagree. If no qualifying comp result exists, an eligible Cook segment may return a hedonic range; hedonic output is always non-actionable and a county-fallback location effect is never served as value.

A neighborhood-premium calculation may be returned as area_context. It is explicitly not property-specific, never appears in the ARV field, and never drives tax, offer, return, debt, or strategy-score math. When evidence does not clear the required gates, ARV is suppressed with a machine-readable reason.

Income and reproducibility

Income never sets ARV. NOI divided by the stated cap assumption is retained only as a cross-check when multifamily units are verified. The artifact also carries an input hash, but the current hash is a calculation key rather than anonymization and does not yet cover every upstream asset/version that affects output; complete reproducibility hashing remains open.