Aria Han

Active

llm-bench

Leaderboards were not answering my questions, so I made tests that did.

The problem

Leaderboards don't answer the only question that matters: will this model hold up on your actual work?

What I built

Real workflow tasks: extraction, code, planted bugs, email drafting, prompt injection, each graded by a programmatic verifier. Works with Ollama, Apple Intelligence, Claude CLI, Bedrock, any OpenAI-compatible endpoint.

Proof

21 tests, programmatic verifiers, published model comparisons including Opus 4.8 vs 4.7 vs Sonnet vs Haiku. 148 commits.

What I learned

A benchmark only matters if the measuring stick is explicit enough to argue with. Programmatic verifiers force an honesty that LLM-as-judge lets you skip.

Designing evaluation systems, the exact skill client evals and monitoring work needs.

Stack

Python · Ollama · Bedrock · Claude CLI

status
Active · Practical workflow benchmark
stack
Python · Ollama · Bedrock · Claude CLI
scope
Standard · Hard · Agentic · Adversarial · Messy
license
MIT

Links

A benchmark only matters if the measuring stick is explicit enough to argue with.

Connected work