I build software that makes its promises checkable: every guarantee has a test behind it, and every important test has been seen to fail when the code it protects is broken on purpose.

A testing toolkit for the bugs caused by duplicate, delayed, reordered and concurrent webhook deliveries. It turns "what if this event arrives twice, out of order, at the same time as its twin?" into a deterministic test that runs in the runner you already have.

• Seeded, replayable delivery plans, with assertions on the state your application ends up in rather than on HTTP status codes.
• Runnable recipes for Express, Fastify, NestJS and Next.js, including the raw-body signature check that is where frameworks really differ.
• Plans from every released version are kept as golden files, so a seed reproduces the same plan across releases, not only within one.
• Saved plans are fuzzed as untrusted input; the public API is a committed report that CI enforces.

A library for versioned application documents on the JVM and Android: a ZIP container that keeps structured data apart from streamed binary assets, migrations between schema versions, integrity checks, and atomic saves.

• A normative format specification that the build machine-checks.
• A mutation fuzzer for the reader, whose findings are committed as permanent regression tests.
• A compatibility corpus of files written by every released binary, which every build must still open.
• Android import and export tested at API 24 and 36; a 512 MiB asset streamed through a 192 MiB heap as a test, not a benchmark.

A narrative-first presentation editor for Android and Windows, now a maintained source archive with its architecture notes and engineering case studies. DocumentKit's persistence layer was extracted from it.

• A guarantee is stated with its limit. Both libraries' READMEs pair every promise with where it stops, and each row names the test behind it.
• Tests are falsified. Before trusting a test, I break the code it protects and watch it fail. That is how several tests that could never fail were found.
• Claims are tested, not assumed. Many of the bugs fixed in these projects were found by writing a test for something the documentation already promised — a teardown that could hang forever, a reader that trusted a size the file declared, a saved plan that crashed replay. The changelogs tell each story: EventLab · DocumentKit.