Though it does seem obvious, it's pretty dangerous to go looking for patterns in the data. More scientific to define the criteria before knowing the results so you can have a hypothesis to test, not just an observation to make.
The danger is that you will come to a conclusion that is objectively wrong.
P-hacking is a known issue in many scientific fields. So is drawing conclusions from over collection of data without repetition of the study... As the number of data points you collect approaches infinity, the chance of finding at least one meaningful-seeming correlation approaches 100% because having an unlimited number of data points to bash against each other makes you more likely to observe an improbable correlation.
No, looking at the data is observation or analysis by definition...whether or not you make a prediction about it. Experimentation is when you make a hypothesis first (after observation) and then test it. Looking at data is not a test--it's either observation or analysis.
