vdbergh

Rather than stopping early the developers should be able to indicate the beta value they are happy with. For example if a developer routinely stops a test at LLR=-1.0 this...

I think we should remain focused. The facts are that many tests get stopped at `LLR~-1.0`. All I did in the OP is to indicate that stopping early drastically changes...

BTW I think the idea of picking a "promising" patch among related patches depending on how the LLR evolves is strongly flawed. The LLR is a much too noisy metric...

> If LOS is too noisy and LLR is too noisy, maybe we can fix one, or introduce a one that is not noisy? As I see it, noisy optimization...

I can explain why it is that way. Normally one would want to follow a test after creating it. The problem however is that the server may reject a new...

I will keep this issue in mind, but I am a bit hesitant to create new PRs at this time since there is already a sizable queue. I'd rather wait...

Actually now I remember that there is a more fundamental reason for the `?follow=1` hack. When we submit a new run form, the run id isn't known yet. So we...

That's not a bad idea (I am not too worried about the slight difference in functionality). We would need the name of the submitter of the current test. Since that...

Just thinking theoretically. I am trying to figure out how pyramid does this. I have the impression that it does not set any state on the server. As I understand...

So when the worker does an api call with username/password, in its reply the server should return a timestamp and a signature of the username/timestamp/ipaddress, both of which are stored...