Feedback, Bug Reporting,
Valid Corrections &
User Reported Test Results
If a number on this site is wrong we would rather hear it from you than ship it. This channel is for defects in the engines, errors in the documentation, corrections to any figure we have published — including the ones we are most confident about — and the results you get when you re-run our published anchors yourself.
Write to contactus@geodesicspacesystems.com.
There is no form, no ticket number and no account. Free Burner keys are capped to verified institutional domains; this address is not — anyone who has something to tell us can use it, from any domain.
Include whatever makes the finding checkable: the request you sent, the result you got, and the architecture, compiler and flags if you are reporting a numerical difference. Screenshots and plots are welcome as attachments.
User reported test results are the one category here we are asking for rather than bracing for. Every stress test on this site names its anchor precisely so it can be re-run somewhere other than our own machines — and a verification suite that has only ever passed on the hardware it was written on has demonstrated less than it appears to. What we do not yet know is whether a result that matches to sixteen digits on one architecture matches on another: transcendental functions are not bit-identical across libm implementations, compilers or instruction sets, and the tests most likely to expose that are exactly the ones quoted here at machine precision.
So a run that agrees is as useful to us as one that does not, and both are worth sending. A disagreement is a finding either way: either the engine is wrong, or the tolerance we published was tighter than the mathematics can actually hold across platforms.
- Engine defect — a result you can demonstrate is wrong
- Documentation error — a formula, citation or units mistake
- User reported test results — a published anchor re-run on your own hardware, whether it matched or not, with the architecture, compiler and flags
- Reproducibility failure — a published anchor that will not reproduce at all
- Security disclosure — handled privately, then credited