Reproducibility asks whether the number comes out of the code. Sciveris asks whether the number is real.
Sciveris — Latin, "you will have known." (say shee-VER-is)
Sciveris re-executes a data-driven analysis, reproduces its headline finding, then adversarially
stress-tests whether that finding holds up — checking for data leakage, batch / site confounds,
effect-size inflation, and p-hacking — and issues a Robustness Report Card
(CERTIFIED · FLAGGED · INCONCLUSIVE · ERROR), each verdict backed by a reproducible exhibit.
It certifies and strengthens findings — it doesn't scold.
Tested by fire. Known by evidence.
🚧 Public release coming soon. This repository is a placeholder while we prepare the first public version.
Developed under the working name Crucible.
Built with Claude.