How Machine Learning Tracks the Same Builder Across Dozens of Legal Entities

A big homebuilder almost never shows up in court under its own name. The defendant in a construction-defect case is usually a small subsidiary — a company created just for that one project — while the parent and dozens of sister companies stay out of sight. Connecting those names into one picture of the whole enterprise takes two things: the right public records, and the right kind of software. This article covers both.

Why a name search isn’t enough

Subsidiaries often carry no trace of the parent brand, and no public record has a field that says "this company belongs to that one." So you can’t solve this with a search box. It takes machine learning — we explain why in Why entity resolution is a machine-learning problem, not a name-match problem. Here we focus on the inputs and the output.

What entity resolution does

Entity resolution is a technique for deciding whether records from different sources describe the same real-world company, then grouping them together. The software compares many fields at once: company names, registered agents, officers, addresses, state of formation, and related-party disclosures in licensing and corporate filings.

No single field settles the question. A registered agent may serve thousands of unrelated companies. An address may be a law office. So the model works on probability — it asks how many independent weak clues line up between two records, and turns that into a match score. It learns from ground truth, meaning relationships we already know are real, like confirmed parent-subsidiary pairs. Records that score above a tuned threshold get grouped into one enterprise profile, with a confidence score on every connection.

The records that feed the model

Different sources carry different clues, and they work best together. Contractor licensing records include officer names and license numbers that survive corporate reshuffles — the same licensed qualifier can tie together companies that share no words in their names. Secretary-of-state filings add registered agents, officer overlaps, and formation dates. Timing patterns are clues too.

County permit records add project addresses and license numbers that link companies across counties and name changes. Court dockets add defendant aliases and, sometimes, corporate-structure admissions made in the case itself. The model weighs each clue by how much it actually proves, and it is checked against verified relationships as new records come in.

What the unified view shows

Once the connections are made, you can see the builder’s full construction-defect footprint across every subsidiary — not just the one company named in the complaint. A project LLC with no history of its own may sit inside an enterprise with hundreds of records across a dozen states.

Patterns surface too. You can see whether the same enterprise has faced similar claims elsewhere under other names, whether defect filings rose across sister companies in the same window — a sign a problem may be systemic — and how the enterprise compares with other builders of similar permit volume. None of that is visible one subsidiary at a time.

Why this matters in construction defect cases

The defendant in a defect case is usually a project entity that finished its purpose years ago. Its own thin record says little about the resources and habits of the enterprise that owns it. What matters is the whole picture: the parent, the sister projects, and the people who appear across all of them.

Defense counsel already has that picture — they represent the same client everywhere and know how it has handled claims before. Entity resolution gives the other side the same structural view, built entirely from the public record.

Delivered responsibly

The resolved graph maps companies, not people. It is built from public and licensed records, delivered in aggregate, anonymized form. It does not surface individual claimant data, and it offers historical context — not a prediction about any case.