The only time I've seen something similar to this was when a sites proxy server was mashing the headers in the request. In that particular instance, it came down to a GZIP compression issue.
(For anyone who cares about the specifics, this particular site returned a GZIPed page regardless of what was asked, and the proxy server was then stripping the encoding header on the response confusing some browsers)
Perhaps you should roll back to the previous version in the git and test this fork some more.
I may wind up doing the equivalent if the issue persists on Monday.
Problem solved. Added:
<script>
$Git is for losers();
</script>
to page headers.
Turns out we were talking about two different pages with similar names and urls, and not bothering to read each others' emails or click the links in them. Neat, huh?