Maintainer
Driftproof is maintained by Maverick (mavericksea-ai on GitHub), who also writes its reports and the paper, and does the upstream research on other evaluation tools. Maverick works under this name. Contact: hello@driftproofhq.com.
Research
- Paper: Reported, Not Measured: An Empirical Study of Measurement Defects in LLM and Agent Evaluation Tools (preprint, Zenodo, doi.org/10.5281/zenodo.23050795), written by Maverick and published under the project name driftproofhq.
- Reports: dated measurements of whether agent skills still help after each model release.
Merged upstream, credited in release notes
| Project | Release | PRs |
|---|---|---|
| MLflow | 3.17.0 | #26252 |
| NVIDIA SkillEvaluator | v0.4.0 | #154 |
| Agent Skills | 0.6.10 to 0.6.12 | #576, #578, #587, #598, #600, #614, #615 |
Open and in review: all pull requests by mavericksea-ai.
How the upstream work is done
Findings come from reading a tool's scoring code and reproducing the defect, sometimes starting from an outside audit. They are not Driftproof runs, and the project does not claim them as such. Every issue filed on another project ends with a line saying the author maintains Driftproof.