Originally Posted by Kinetic
Or I could choose the boss and MT healers as actors, and MT as target...to see why MT died because he didnt get enough heals, or boss got lucky crushing streak..
|
I've thought about programming a similar filter back in the day, and several things that may or may not be issues came to mind, so I put it on hold:
1. The logs have no indication of non-actions, including location, facing, and movement. So how will you know if the tank took a couple of steps backwards and stepped out of healing range at a critical time in the fight? All you'd see is that the healer took 4 or 5 extra seconds to get the heal off... a second or two to realize that something was wrong, a second or two to move forward, and then cast again. The log will not indicate movement or anything, just an actionless time gap -- healer's fault.
Not just actor locations, but mobs and things like totems and AoE. What if the Shaman dropped the totem 3 yards short and the MT gets feared? What if the OT parked his mob 5 yards the wrong direction, and that had a cascading effect?
2. How do you avoid the "blame the last domino to fall" syndrome? Most wipes, I'd guess, are not a single player's fault. (OK, I wiped a raid single-handedly with a stupid mistake, but it was while everyone was buffing before the pull.) That's why the typical PUG blame is either "poor tanking" or "not enough heals", when in fact the problem was more substantial than that. And I'm not talking something as simple as "group-wide DPS was low, meaning healers had problems, meaning the tank died", but longer chains of events.
It's not an excuse to not have a tool, but it's like a novice using R, SAS, or SPSS and "proving" something because statistical test XYZ indicates that ABC is statistically significant to the 95% confidence level. Did they use the correct test, given the type of data and ABC? Were the methods used to get the data correct? There's a very long chain and it's easy for an amateur statistician to really screw things up, but with authority, because the numbers come out of a Name Brand statistical package using a Name Brand statistical test.
In like manner, even a clever, well-informed Raid Leader could fail to account for all of the what-ifs and things like #1, and #2, above, and could come to an "irrefutable, based on the logs" proof that it was A's "fault" when in fact it was B's. (Or more probably, the logs don't give the proper information to tell who started the death spiral.)
3. How do you account for real-time judgement? The healer's making a bunch of real-time decisions in terms of locations, spell usage, spell rank, mana burn rate, etc. Looking back at only the recorded numbers won't give any feel for the issues that someone had to tradeoff in a half-second decision which might turn out to be "wrong". ("Wrong" in the sense of being the last in a series of decisions by multiple actors -- the one that the tool user happens to see and says "Aha!" and stops looking.)
Yes, you can see that the Shaman is dropping totems 3 seconds too early and wasting mana. You can see that the Druid is throwing a Wrath into his rotation when it should be an extra Starfire instead. Things that are cyclical, repeated through a fight. Things that are averaged together over 30 or 40 actions. It's like statistics, really only working with masses of data, not individual data points.
But will it be evaluated that way?
Perhaps this is interesting, but not really applicable... Give people a tool and while most will misuse it, many will not and after all "It CAN be done, so therefore SHOULD be done." Just my personal worries that I've not been able to resolve.