Data / Florida title strings

Florida high-value home title strings, preliminary 2026

Last updated October 24, 2026

A preliminary algorithm labeled 25,783 public assessment records across 66 Florida counties. Its accuracy review failed. These counts describe algorithm labels, not verified ownership or buyers.

What did the calculation count?

The edition includes 18,932 records with just value from $5 million to under $10 million, and 6,851 at $10 million or above. The bands do not overlap.

The source covers 66 counties. It mixes 56 final-folder and 10 preliminary 2026 NAL submissions. Jackson is excluded because its listed archive contains SDF sales data instead.

The calculation uses DOR codes 001 for single-family property and 004 for condominiums. Counts describe assessment records, not unique owners, households or purchases.

Classification interpretation has low confidence. The published counts are verified outputs of the stated algorithm and inputs.

Preliminary algorithm labels by assessment-value bandTwo bars show the share of assessment records receiving each unvalidated title-string label. Exact counts are in the table.$5,000,000 to under $10,000,000 · 18,932 recordsIndividual-like. 12,496 records. 66.00 percent of algorithm labels.Trust-like. 1,956 records. 10.33 percent of algorithm labels.Company-like. 3,283 records. 17.34 percent of algorithm labels.Trustee named. 520 records. 2.75 percent of algorithm labels.Other or unresolved. 677 records. 3.58 percent of algorithm labels.$10,000,000 and above · 6,851 recordsIndividual-like. 4,192 records. 61.19 percent of algorithm labels.Trust-like. 745 records. 10.87 percent of algorithm labels.Company-like. 1,553 records. 22.67 percent of algorithm labels.Trustee named. 152 records. 2.22 percent of algorithm labels.Other or unresolved. 209 records. 3.05 percent of algorithm labels.

The bar segments run left to right in table-column order. Shares describe algorithm outputs only.

Just-value band, USDAssessment recordsIndividual-like algorithm countTrust-like algorithm countCompany-like algorithm countTrustee named algorithm countOther or unresolved algorithm countInterpretation and validation statusSource coverageSource and check date
$5,000,000 to under $10,000,000
Source
189321249619563283520677Preliminary title-string algorithm counts. Classification validation failed. These are not verified ownership or buyer statistics.Combined 66 counties excluding Jackson. 56 final-folder and 10 preliminary 2026 NAL files.Go Low Profile calculation from Florida Department of Revenue public NAL files
Checked 2026-10-24
$10,000,000 and above
Source
685141927451553152209Preliminary title-string algorithm counts. Classification validation failed. These are not verified ownership or buyer statistics.Combined 66 counties excluding Jackson. 56 final-folder and 10 preliminary 2026 NAL files.Go Low Profile calculation from Florida Department of Revenue public NAL files
Checked 2026-10-24

Why is this a preliminary release?

The unchanged classifier uses the primary-owner text in each record. It assigns individual-like, trust-like, company-like, trustee-named or other labels.

A label does not establish a legal owner type, beneficial ownership or anonymity. The source name field can contain at most 50 characters.

Three disjoint samples each contained 500 strings. An independent AI review assigned comparison labels without seeing classifier predictions. The first two samples informed calibration.

The original target required precision and recall of at least 0.95 in every class. The final 500-string holdout failed that requirement. Company recall was 0.9265. Other-label precision was 0.8182 and recall was 0.5625.

Those are sample metrics, not calibrated error bounds for this population. No human validation is claimed. The classifier was not tuned on the final holdout.

Which records and values are outside the measure?

The DOR guide defines just value as the appraiser's market-value opinion after statutory adjustments. It is not sale price or independently verified current value.

Assessment values relate to January 1 of the assessment year. Owner strings observed in the roll do not establish the buyer at a sale.

The public files omit confidential records, including records covered by section 119.071. This omission creates a coverage limit. The study cannot describe those excluded records.

The stream read 11,069,299 records and excluded 56 blank just-value fields rather than replacing them with zero. A small selected-use missing-value count remains withheld.

Archive sizes, hashes, schemas, row widths, county/year fields, file type and record-key uniqueness were checked. No qualifying record had an invalid numeric just value.

Final-folder status does not establish that every value-adjustment-board change has been incorporated. This is not a complete statewide final-roll estimate.

Read the official 2026 DOR guide and DOR Data Portal.

How does the release protect individual records?

The release combines all 66 included counties. It provides no county breakdown, owner names, addresses, record identifiers or lookup.

All ten published class counts exceed 10. The smallest is 152. Every positive complementary sum of the ten disjoint cells also exceeds 10.

These checks cover this combined table. They do not authorize later geographic or value subdivisions.

How can you reuse the calculation?

Download CSV · Download JSON

Machine-readable method and limitations. Original compilation licensed under CC BY 4.0.

Cite version 0.1-preliminary, checked October 24, 2026. Preserve the failed-validation notice, mixed vintages, excluded county and confidential-record coverage limits.

  • 2026-10-24 Published combined 66-county descriptive algorithm counts with failed holdout metrics and missing-value exclusions disclosed. No validated ownership estimate is claimed.

Public source repository · Methods · Changes · Corrections