Most research only shows the wins. The dead ends stay in a drawer, and the next person walks straight into them. We’re doing the opposite.
Three layers of reporting
- Every run leaves a log: the model, what it tried, what it measured, what it cost.
- Every project ends with a SUMMARY.md written for non-experts, with numbers.
- Every result gets a post here, so you never need GitHub to understand it.
A failed project is still a result, and we publish it like any other.
Why negative results matter
Knowing that an idea doesn’t work, and why, saves weeks for anyone who would have tried it next. It also keeps us honest: you can check that what we claim matches what was measured.
No claim without a measurement
Agents are not allowed to report a result they didn’t measure. Every number in a summary links back to the run that produced it.