Howard Chung

The protos in GameTracking-Dota2 might be a good place to start. You might need to NetHook the client to see what calls are being made, and it might need to...

For performance reasons the rankings are done against a random sample on each request, so depending on the sample the results may vary. I can look into increasing the sample...

I doubled the sample size, is it more consistent now?

As discussed on Discord, this requires the death events to be extracted out to a new match level property: deaths_log. The current kills_log property in each player only tracks kills...

While we're on this, we should also clean up the public_matches/public_player_matches sample tables

https://github.com/odota/core/blob/master/svc/scenariosCleanup.js already implements a cleanup worker, we could probably just turn this into a generic cleanup worker and rename it

match_gcdata can also be cleaned, as replays are deleted after a while. We would want to keep pro games, though, so need to save the IDs in the matches table.

hitting this issue again, temp fix is `delete from hero_search where match_id < 5600000000`

Added to scenariosCleanup.js Clearing from match_gcdata `delete from match_gcdata where match_id not in (select match_id from matches) and match_id < (select max(match_id) - 100000000 from match_gcdata)` Clearing from hero_search `delete...

The data looks like: ``` "picks_bans": [ { "is_pick": false, "hero_id": 57, "team": 1, "order": 0 }, { "is_pick": false, "hero_id": 89, "team": 1, "order": 1 }, { "is_pick": false,...