Accuracy

Validated against closed transactions, not vendor claims

Plotgleam accuracy figures come from comparing our valuations against subsequent closed sale prices across 18 months of Colorado transactions. No hedonic models, no self-reported estimates.

94%
Within 5% of final sale price

2.8%
Median absolute error

18 mo
Validation dataset window

Validation Methodology

How we measure our own accuracy

For each property in our validation set, we ran a Plotgleam valuation using only data available on the day a hypothetical lender would have ordered it - before the property closed. We then compared that valuation to the actual closed sale price recorded in county records.

This forward-looking validation design prevents the hindsight bias that plagues many AVM benchmarks, where the model is tested on data it was also trained on.

Data cutoff discipline

Every validation run is conducted with a strict data freeze: no comp sales, permit updates, or MLS data dated after the valuation order date are permitted to influence the estimate being tested.

Dataset scope
12 Colorado counties. Single-family residential only. Properties with known data integrity issues (duplicate records, missing square footage) excluded. 18-month window ending 60 days before publication.
12 counties, 18 months
Sample weighting
Stratified by price decile, county, and property age to prevent metro-heavy sampling from masking accuracy gaps in rural and mountain communities.
Stratified by 3 dimensions
Update cadence
Validation runs quarterly. Published figures reflect the most recent completed quarter. Model updates are tested against a held-out validation set before deployment.
Quarterly revalidation

Performance Summary

Key accuracy metrics

94%
Within 5% of close price
2.8%
Median absolute error
98%
Within 5% at 90+ confidence
88%
Mountain market accuracy

Comparison

Plotgleam vs. standard AVM approaches

Published accuracy comparisons for national AVMs reflect their self-reported figures. Plotgleam figures are from our own prospective validation methodology described above.

Feature Plotgleam National AVM (typical) Desk Review Full Appraisal
Turnaround time Under 8 min Seconds 1-2 days 10-14 days
Within-5% accuracy 94% 72-82% 89% 96%
Median absolute error 2.8% 4-6% 3.2% 1.8%
Block-level comp scoring Partial
Condition adjustment Partial
Auditable comp table
Confidence score Some
Mountain / rural market support Limited Depends

Confidence Scoring

What the confidence score tells your team

90-100: High confidence

Strong comp availability, clean data, stable micro-market. 98% of these reports land within 5% of final close price. Safe to act on in most underwriting workflows.

70-89: Moderate confidence

Adequate comps but some data gaps or market volatility. 87% within-5% accuracy. Consider alongside other underwriting signals for borderline loans.

Below 70: Escalation flag

Thin comp market, significant data gaps, or high market volatility. Report is flagged for human review. Plotgleam will not deliver a low-confidence estimate without an explicit escalation notice.

Run a test on your own properties

Request a demo and bring a list of recently closed properties from your market. We will run Plotgleam retroactively and show you the accuracy figures directly.