Repository navigation
Deep Dive
Claude edited this page Nov 16, 2026
·
2 revisions
The Home page is the summary for engineering leaders. Everything below is the working detail, in two tracks. The tracks share pages; each landing page tells you which parts matter for you.
Start at For Engineers.
| Page | Why read it |
|---|---|
| Architecture | How the harness drives the real Claude Code tool and what it records |
| Request Logging | How to know which model answered each request, and what a run costs |
| Early Findings | What reading the plugin and the documentation turned up |
Start at For AI Evaluators.
| Page | Why read it |
|---|---|
| Experimental Design | The hypothesis, the arms, the measures, and what the pilot can and cannot show |
| Corpus | The benchmark chosen, the alternatives, and its limits |
| Request Logging | Attribution and contamination control |
| Page | Why read it |
|---|---|
| Results | Results of the first study, with every chart and the limits. |
| Learnings and Next Steps | What we learned about method, and what comes next |
FrontierEval is a project of Side of My Desk. First study published November 2026. Code under the MIT License; paper and results under CC BY 4.0.
FrontierEval
Deep Dive
For engineers
For AI evaluators