There are things like CONSORT[1] which kind of do this. Statisticians like Fisher[2] have a ton of good general advice of the design of experiments. (A plug for The Lady Tasting Tea[3] and the 7 Pillars of Statistical Wisdom[4] feels appropriate here.)
On the whole though, most of the things you should and should not do are so domain specific its very hard to give much useful advice at the level of "all science." Right now this seems to work because researchers are so eager to anticipate objections and
and avoid unnecessary arguments during peer review they stick slavishly stick to the same methods used by seminal papers in their field, and this has the same effect as running down a checklist.
There probably is a case to be made for using an actual checklist, though[5].