Base watch time on collected item metadata runtime if available, fallback to metadata providers runtime if not
I noticed this problem recently while watching a lesser-known web series: there can be drastic variances from the TVDB's runtimes to reality (seems like this will only get more prevalent as more and more content is created for non-traditional distribution).
The web series I was watching had a bunch of 2 or 3 minute specials. I noticed that after watching 10 of these (25 mins total), Trakt was totaling my watch time at 5 hours (stated runtime on TVDB is 30 mins).
It would be cool if a) I have the given media item collected, and b) scrobble or mark it watched, that Trakt would assume the runtime of the collected item instead of TVDB's standard runtime. If the item isn't collected of course, it would make sense to fall back to the best source.
Burke Davis commented
Just brainstorming here, given all the scrobbling data, etc. Trakt could probably impute / crowdsource much more accurate runtime data instead of relying on other metadata providers. Given it's just the runtime, it'd be a lot easier to crowdsource than other metadata and you could average the imputed runtime across all plays to make it more accurate over time.
I think this would help with Netflix / HBO episodes that can sometimes swing 30-60 minutes.