Thread by @federicolois

1/ What does this mean for research? For example, while my twitter followed has increased absurdly since early last year because of my work on data analysis on SARS-Cov-2, I was mostly known for my work in performance analysis.

https://twitter.com/federicolois/status/1359892199463395331

2/ What performance analysis teaches you is that you run experiments of the type 'If P then Q' every single day, several times a day.
For ex. "If I change this data structure then I will be able to obtain better performance by accessing memory contiguously"

3/ Now, when you enter into the realm of 'how the physical CPU is working' then becomes far more difficult. The reason behind that is that with that many moving parts noise is very difficult to separate from the signal.

4/ That's where Modus Tollendo Tollens comes to the rescue. Because now instead of trying to prove "If P then Q" most of the time proving "not Q" proves to be easier. When designing the @ravendb Storage Engine, most of the time was spent on proving "not Q".

5/ But the most interesting and overlooked use of that is that: If we though that P was the caused of Q. If we could find an instance where not Q, we immediately knew P (while related) is not the actual reason why Q happens. Therefore, we need to think and refine the hypothesis.

6/ That has immediate impact in how @LDjaparidze and me tackle our research. There are always million of competing hypothesis (we even have different ideas on certain topics) but our discussions revolve around "proving not Q" to rule out hypothesis.

7/ Overlooking the capability of Modus Tollendo Tollens to rule out hypothesis has lead several SARS-Cov-2 researchers into rabbit holes that is very difficult then to get out from.

8/ When analyzing data, this is probably the most effective and powerful method. And something that most are not using.