Test Desiderata, remade

Kent Beck published twelve properties of good tests in 2019; Emily Bache reorganized them in 2025. Both lists are good, and both are one author’s view. This garden’s Test Desiderata is a third take, built the other way around: start from what the wider field independently converged on, and let Beck and Bache be sources rather than authorities.

What the evidence keeps

Six properties are genuine cross-author common ground: five or more authorities name each of them, in their own vocabulary, often before Beck’s list existed: Fast, Run in Any Order, Readable, Deterministic, Diagnosable, and Behavioral. The shared claim under each is more specific than its name: “Fast” is the precondition for tests being run at all; “Deterministic” protects trust in the suite as a whole.

The implicit meta-property the field converges on is trust in the suite, not hygiene of the individual test.

Two renames are improvements and are adopted: Run in Any Order for Beck’s Isolated, Diagnosable for Beck’s Specific. One property splits: Behavioral’s mirror, Structure-Insensitive, stays a separate property (and a seedling: it’s the one property still awaiting its cross-author audit).

What gets renamed on principle

Beck’s Automated becomes Automated Checks. The context-driven school is right that what runs unattended is checking, not testing; conceding the word costs nothing and buys honesty about what a green build means. Writable survives with a caveat: the field treats authoring cost as design feedback, not a goal. Minimal Data is adopted with credit where it’s due: Bache promoted Meszaros’s Minimal Fixture into a named property.

What gets dropped

Composable is out. Its provenance is mockist (Freeman and Pryce used the word in 2004, fifteen years before Beck’s list), and naming it as a property silently takes the mockist side of the mockist/classicist fault line. The universal part of its meaning is already covered by Run in Any Order.

Inspiring is out as a property and kept as the outcome of Predict Success in Production. Every metric ever proposed for “inspiring” turns out to measure another property in disguise. Bache demoted it; the measurement evidence says she was right.

What this list refuses to be

A property list covers one dimension. Test categories, legacy-code mechanics, suite shapes, and team process are real axes this list deliberately leaves to other tools. It also inherits assumptions worth naming: a TDD-shaped workflow, a production target, a codebase that already has tests, and a reader with the agency to make trade-offs. Where those don’t hold, start elsewhere (Feathers, not Beck).

The research behind every claim here (sources, verification verdicts, and the disagreement map) lives in research/Test-Desiderata/ in this repo.