Test impact analysis for .NET

Run only the tests your change can break.

TestDetta reads your diff and runs just the test classes that can observe it. When it cannot be sure, it runs more, never less.

TestDetta · pull request check Passed

93% less recorded test time

All tests 5 m 59 s · 274 classes
TestDetta 22.4 s · 8 classes
  • 8 classes run they run or reference the changed code
  • 266 skipped no path from the change reaches them
  • 229 tests passed 0 failed
A real change to OpenTelemetry .NET. See the whole run →

Measured on 53 open-source .NET repositories, including

  • Polly
  • Newtonsoft.Json
  • AutoMapper
  • Spectre.Console
  • Markdig
  • Serilog
  • Refit
  • Autofac
  • NodaTime
  • FastEndpoints
  • Scriban
  • FluentValidation
  • AutoFixture
  • OpenTelemetry

Independent measurements. These projects are not affiliated with TestDetta and do not endorse it.

Measured on real code, not on demos

We change one member at a time and check that every test class that fails is one TestDetta selected.

53repositories, each with its own frameworks and build quirks
3,071changes that broke at least one test
35%of its test classes ran per change in the median repository
5misses in the latest round, each a known limitation: code that runs once or on another thread credited to another class, and a source generator whose output was not read; earlier real misses were fixed with a test that fails without the fix

The 0.2.0 release round of the evaluation, 2026-10-13 to 2026-10-15, judged strictly: a test run that crashes or hangs outside the selection counts as a miss unless the same run without the change shows it is not the change's.

Share of test classes run per change, one bar per repository The latest round of the evaluation corpus: for each of 53 open-source repositories, the mean share of its test classes TestDetta selected for a change to one member, sorted from smallest to largest. Median 35%. 0% 50% 100% polly: 3% spectre: 6% opentelemetry-dotnet: 9% communitytoolkit: 10% swashbuckle: 10% unitsnet: 10% nsec: 11% guardclauses: 13% mimekit: 13% autofixture: 14% nodatime: 15% aspnetcore-odata: 17% awesomeassertions: 18% fluentvalidation: 20% vogen: 20% fluid: 21% carter: 22% fastendpoints: 23% serilog: 24% managedcode-communication: 25% libgit2sharp: 26% roslynator: 28% castle-core: 30% open-xml-sdk: 30% alba: 33% nlog: 34% bunit: 35% heron-mudcalendar: 38% nager-date: 38% markdig: 39% yamldotnet: 39% lamar: 41% newtonsoft-json: 43% refit: 44% mediator: 48% scriban: 48% autofac: 49% graphql-dotnet: 50% immediate-handlers: 51% devarchitecture: 53% efcore-projectables: 53% nsubstitute: 54% scrutor: 55% aicentral: 56% argu: 56% protobuf-net: 58% cliwrap: 62% efcore: 68% fscheck: 69% grpc-aspnetcore-validator: 70% automapper: 86% orleans-syncwork: 86% mapperly: 97% median 35% pollymapperly
Each bar is one repository: the share of its test classes TestDetta ran for a change to one member, smallest to largest. In 20 of 53 it was a quarter or less.

How it decides

One real change to OpenTelemetry .NET, followed step by step. Each step narrows the selection only when it has evidence.

1 · The change

…ConfigurationExtensions.cs
-bool differs;
+bool overridden;
-differs = new FilePath(…) != filePath;
+overridden = new FilePath(…) != filePath;

A rename that changes no behaviour. A normal pipeline runs every test project: 274 test classes.

2 · What TestDetta works out

  1. Projects2 of 30 test projects affected; the other 28 are not even built
  2. Symbols1 member changed: WarnIfConfigFileEnvVarDiffers
  3. Coverage map8 test classes ran it on main
  4. NamesNo new code to trace
  5. FallbacksNone needed

3 · Tests to run

8 of 274 classes run, 266 skipped

Exactly the 8 that fail when the method breaks. How we checked

When a step cannot trace a change, the whole test project runs. What runs everything →

Built for teams that can't afford a missed test

TestDetta runs inside your CI and never stands between you and a release. If it cannot be sure, it runs more.

Security and data →

0
lines of your code leave your CIAnalysis, coverage maps and test results stay on your machines.
0
network calls of its ownTestDetta sends nothing anywhere. Your CI fetches the tool and restores the maps like any other package or cache.
5·2
test frameworks and runnersxUnit v2 and v3, NUnit, MSTest and TUnit, on VSTest and Microsoft.Testing.Platform.
0
license keys, accounts or sign-upsApache-2.0 for every repository, public or private. Install it and it runs.
FrameworksxUnit v2 · xUnit v3 · NUnit · MSTest · TUnit
RunnersVSTest · Microsoft.Testing.Platform
CIGitHub Actions · GitLab CI · Azure DevOps · Jenkins · Bitbucket
PlatformsWindows · Linux · macOS

Two steps in GitHub Actions

Your default branch records once. Every pull request then runs only what it affects.

  1. on push to main

    Record a coverage map

    Every test runs once, with coverage. The map is saved in your Actions cache under its commit.

  2. Actions cache testdetta-coverage-<sha>
  3. on pull request

    Run what the change affects

    The newest map within 50 commits before the merge base is restored, only the affected test classes run, and the job summary says why each one did.

    Passed 8 of 274 classes · 93% less recorded test time

.github/workflows/ci.yml
on:  push:    branches: [main]  pull_request: permissions:  contents: read  actions: read jobs:  test:    runs-on: ubuntu-latest    steps:      - uses: actions/checkout@v7        with:          fetch-depth: 0       - if: github.event_name == 'push'        uses: TestDetta/action@8750db18099a2af0436788d20dc591e94b550c07 # v0.5.2        with:          mode: record       - if: github.event_name == 'pull_request'        uses: TestDetta/action@8750db18099a2af0436788d20dc591e94b550c07 # v0.5.2        with:          mode: test

For AI coding agents too

Claude Code, Codex and other agents run only the tests their changes can affect, and check them before they hand back. In Claude Code: /plugin install testdetta --marketplace selcukgural/TestDetta

Free and open source, for every repository.

Apache-2.0. Public or private, on any CI and on your own machine.