Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
matm
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
18 ms
·
151.
▲
by
matm
14y ago
Very keen observation. There are a number of ways to solve this problem. Machine learning helps (there are a few simple heuristics we currently use to sort the Event Feed). But the right interface for defining events helps even more. Sorry
152.
▲
by
matm
14y ago
Thanks for the heads-up, we'll fix this asap. We've run into this issue before, and it seems to have something to do with the font's Unicode values being "out of range" on Windows, whatever that means.
153.
▲
by
matm
14y ago
Unfortunately, we're not tracking those events now, mostly because they don't fit neatly into the event-based analytics model (i.e. they're not as "graphable"). But that doesn't mean they aren't important. Agreed the data can be very useful
154.
▲
by
matm
14y ago
You don't see Heap on there?
155.
▲
by
matm
14y ago
Good point. In terms of network performance, we try to batch events and minimize server requests. For link tracking, our method is similar to how other analytics products track link clicks - we delay the page navigation the min of: 1) 250ms
156.
▲
by
matm
14y ago
Our bottom tier (which includes a 1 month free trial) lets you track 2,500 users/month. That said, pricing is really hard to get right. There's basically zero chance we nailed it on the first try.
157.
▲
by
matm
14y ago
It's essentially imperceptible. We measured it, and our client-side script introduces a 0.1% CPU overhead.