Benchmark foundation v0.1
Bench
Each track uses versioned prompts and a fixed publication contract. Source may vary; every runnable artifact ends as an isolated static bundle.
5 Benchmark tracks0 Published runs
Catalog · Benchmark tracks
Published runsDevelopment only
Published runs
0No benchmark result has been published yet.
This is intentional. The first public result must include a real model identifier, immutable prompt version, source artifact, and evidence.
Development only
Architecture fixture
Method
What makes a run publishable?
- 01The prompt and system instructions are versioned and hashed.
- 02The exact provider model identifier and run time are recorded.
- 03The unedited source and every failed attempt remain available.
- 04The build produces a static bundle with an HTML entrypoint.
- 05The bundle passes isolation, runtime, and evidence checks.