RandJoy
Known-answer evaluation packs
Make small smoke tests with answers you can check exactly.
Latest result
Ready when you are
Make small smoke tests with answers you can check exactly.How it works
- Choose one task family and the number of cases. Use the seed to reproduce an evaluation pack.
- Generate and inspect the prompts. The main export contains IDs and prompts only.
- Export the answer key separately. Score returned text after trimming leading and trailing whitespace, using exact string matches.
Narrow tests with exact answers
Arithmetic cases add or multiply integers from −100 to 100. Sorting uses six integers with possible duplicates. Date cases transform valid 2024 dates to DD/MM/YYYY. Lookup cases request a named key from a small JSON object; extraction cases request one explicitly marked code. Inputs and expected answers share stable IDs. These tasks check constrained behavior and output formats.
Your inputs stay with you
Your inputs and generated results stay in this browser. Settings and five recent runs use local storage; clearing browser data removes them. Export a copy to keep it. Hosting may record standard page-request logs, but generated rows and participant names are not sent to the server.
Does a high score show general intelligence?
No. These authored templates are smoke tests, not a representative benchmark of reasoning, language, safety or real-world task performance. An exact-match scorer also rejects extra prose even when the underlying answer is right. Keep inputs separate from expected answers when sending cases to a model.
Repeatable test data
The research tools use the versioned xoshiro128** pseudorandom generator with named seed streams. They are intended for small examples and testing, not secrets or certified draws. A recipe records parameters and generator version. Floating-point transforms are rounded to six decimal places; keep exported files when exact cross-platform bytes matter.
Seeds are repeatable. Use the password generator for credentials.