Boss Nomad.6487:

Kills of wvw matches seem to be cached for 5 minutes. I think the rest of the data is cached for 1 second now so is it an oversight that kills are still cached for 5 minutes or a limitation of the architecture? They also seem to be cached asynchronously by map (on request kills of one map could be updated but the rest not).

Lawton Campbell.8517:

So, the kill/death counts are cached in the exact same blob as the map state — however, they’re collected using a different mechanism than the map state. The way I implemented it, there’s actually an API-specific backend server that watches logs coming off of the map instances. Every skirmish (I think? it might be on an interval), each WvW map instance flushes some transient state to logs, including kill/death counts; those logs are picked up by the API backend server and aggregated.

I have some minor changes that might be ready for the next release that watch for individual on-kill notifications (in addition to the periodic aggregated one) — that should improve the latency a bit.

tl;dr architectural, but I’m working on it.