I've had really poor results with the 3-2-5, too. It is capable of winning 5-0, but is also capable of losing 0-5. It's all over the place. While I understand that Mr. L says this is the "randomness of the ME", I'm not seeing this level of problems with other tactics (including Mr. L's tactics). I even went so far as to create a custom team built around the unorthodox formation and even with a team of absolute world beaters, it was inconsistent. It plays some fantastic football at times, but it also can be completely terrible.
I'm curious how your custom database eliminates the "randomness". Fewer injuries?
And the bigger question is, what's the point of testing in a sterile database when that is not how the game is played? A "game breaker" tactic will be effective in actual match conditions, not just in a set of unrealistic circumstances. And if the ME is filled with randomness (as actual football is), you will not get accurate results from a single season, with or without a custom database. Only after running multiple seasons will you be able to determine the true effectiveness of the tactic, as a fantastic season may be followed by an awful one. If you really believe the ME is "incredibly random", you must accept that without regression analysis, you really have no idea what a tactic will do ever time.
TL;DR, Using a custom database over a single season to test the effectiveness of a tactic is not going to yield accurate results.
I agree that you need multiple seasons to get accurate results, and I have mentioned this earlier in the thread. The time issue is a problem for that though. I can't test all these tactics for multiple seasons each. You can for example look at mine and Jesaustralia's FM13 thread. This was the time before I got kids. You can see there that we tested all tactics in 76 matches in 3 different databases (228 matches tested per tactic). THAT gives you an accurate result. But this is not possible for me to do now, as I have a one-year-old, a two-year-old, a wife, a job at the university and I'm also writing my master thesis in linguistics.
However, it is clear to me that you haven't read my comments about my database either. It is no different than an actual game. There is nothing sterile about it, nor does it have unrealistic circumstances. I have merely fixed some settings that would have given one tactic an advantage over another.
As an example, take injuries. If in one test, I got 10 players injured, while when testing another tactic I didn't get a single injury. Would you say that was fair? (Please answer this question, I urge you)
Also, if you are curious about how I remove randomness, you CLEARLY haven't read my front page, where you can find mostly all information there is to say about the test itself.
That said, the MOST important statistics for me is the feedback from this forum, as that would give me "those extra seasons tested". It doesn't need to be constructive feedback, but because of the randomness in the game, this feedback need to include screenshots from at least 20 matches, preferably a full season. When enough results come in, I can know for sure, and then make changes. I would never change a tactic based on one comment saying "this tactic is ****", without any form of "proof", even though "proof" from one user wouldn't be enough. If 5 guys send me screenshots of a full bad season, then I would of course do something about it...