# [Schema Benchmarks](https://schemabenchmarks.dev)
A project aiming to consolidate and standardise comparisons of schema validation libraries.
## Methodologies
**Runtime** benchmarks are run in sequence, on a GitHub runner. There are different types of benchmark:
- Initialization: Creating the schema itself. This is usually a one time cost.
- Validation: Checking if a given value matches the schema.
- Parsing: Checking if a given value matches the schema, **and** returning a new value. This will include any transformations.
Some libraries only support validation, or parsing. In these cases, we categorise them accordingly.
Each library is benchmarked against a set of data, both valid and invalid.
**Download** benchmarks are created by compiling example usage files with [Rolldown](https://rolldown.rs/), and measuring the size of the output, both minified and unminified.
**TypeScript inference** is measured with the TypeScript compiler API: whether the inferred input and output types describe the benchmark data exactly, how many type instantiations declaring the schema and reading both types costs, and the type an editor shows on hover.
## Structure
This monorepo is split into 5 main parts:
### `schemas`
The schemas and benchmark definitions for each library, and the data used.
_If you're looking to add a new library, this is most relevant._
### `website`
The website for displaying results, built with [TanStack Start](https://tanstack.com/start). Deployed at [schemabenchmarks.dev](https://schemabenchmarks.dev).
### `bench`
Node scripts to consume and run the benchmarks, and the results.
### `json-schema-tests`
A wrapper around [JSON Schema Test Suite](https://github.com/json-schema-org/JSON-Schema-Test-Suite) (gitignored, will checkout during postinstall), to allow us to run the tests against each library.
### `utils`
Simple utilities that might be useful in multiple places, that aren't worth depending on a specific library for.
## Contributing
See our [contributing guide](https://schemabenchmarks.dev/contributing/).