[Next_rim]

@Next_rim: It would be great if you would do that exercise and let us know the results. Its not much use saying what you "think" the results might show if you haven't actually looked at it. I think you would need to do this for all races, not just Protoss, so we have something to compare.

I could do it for maybe 3-4 tournaments in RO16 and further. That's ~200 games, maybe more, and the amount of games is reasonably small to attempt to do it manually through a conventional parser. For any better output, need to automate the parsing.

Here is a link on a project that attempted to build a C++ parser:

http://code.google.com/p/starcraft2replay/w/list

They figured out the codes for all actions in game, so if I only could break the MPQ part of the replay, then convert it to .txt, I could use VBA engine in excel to parse it and get BO's. The code string is standard 75-character string, with 4-5 bits of code meaning group (selected/deselected), time in half-seconds (lolwut?), action taken, and some other stuff. No need to dig coordinates, just parse for unit production, building production and time of action up to, say, min 8.

I only see a real problem with mass unpacking (have never dealt with MPQ) and mass coversion of code to .txt. I've had lots of experience from here, as I have to dig through client's stupid 150+ page transaction details that they send in a locked .pdf. I convert the data to .txt, import into excel, then use VBA to break it down into structured parts that can be sorted, searched and used for analysis.