Benchmark
Supabase Evals
Evaluate agents that build, deploy, investigate and repair Supabase applications.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
A public benchmark spanning the Supabase developer journey, with separate regression scenarios for specific failures.
Methodology
Agents use containerized Supabase environments with the CLI and MCP server. SQL checks, client requests and generated files provide deterministic evidence; an LLM judge handles semantic assessment. The launch protocol permits one retry after failure.