Test data builders with defaults reduce test breakage when a domain object changes

Hypothesis

An object mother, in Fowler's description, is a factory that returns familiar canned objects ("John, hired last week") shared across many tests; he names as its fault the heavy coupling that arises because many tests depend on the exact data in the mother, which makes changing that data tricky. A test data builder takes the opposite stance: it starts from a valid default object and exposes one method per field, so that an_order().with_status("paid").with_lines(3).build() states exactly the fields the test depends on and inherits everything else. The hypothesis: when a required field is added to a domain object, a constructor signature changes or a validation rule tightens, builder-based suites need edits in one place (the builder's defaults), whereas suites that construct objects directly, load shared fixtures or depend on canned mother instances need edits proportional to the number of tests that touch the object. A secondary claim is readability: the fields named in a builder-style test are precisely those the assertion depends on, so a reader need not compare the test against a fixture file.

Prediction

In version history, commits that add a required field to an entity or change its constructor touch fewer test files, normalised by the number of tests that use the entity, in codebases where the entity is constructed through a builder than in codebases using direct construction, fixture files or an object mother. Reviewers asked "which fields does this test depend on?" answer faster and more accurately for builder-style tests.

Proposed test

1. Select repositories in which both styles coexist, or one codebase before and after builders were introduced for a set of entities.
2. Identify commits that add a required field or change a constructor; count changed test files and changed test lines per commit, grouped by construction style, and normalise by usage counts.
3. For the readability claim, give reviewers matched pairs of tests (same behaviour, both styles) and time the identification of the fields the assertion depends on.

Status

No result is claimed. Builders cost a class per entity and can hide which defaults a test silently relies on; object mothers keep their value for shared, named examples that appear in conversations with users. The hypothesis concerns edit counts on change and readability, not defect detection.