all models are wrong but some are useful 2

I always thought It's Always Sunny in Philadelphia was a weird show. The acting looked awful, but my friends (die-hard fans) insisted that was the whole point, that if I had started from episode 1 I'd understand. I wondered if that was really true, or that watching the show from the start would simply brainwash me into liking the show.

People say pickleball is actually really difficult. Putting aside the issue of what it means for a sport to be "difficult", I wonder if they say that just because they're the players who are accustomed to playing pickleball and are trained to feel that way. But pickleball is quite accessible and popular right now, so there are players from many other sports who encounter it and give more "objective" feedback. 

On the other hand badminton is a little less accessible and vastly less popular, so there's not much calibration with the outside. I wonder if my sport of choice is actually incredibly easy, not requiring much athleticism, despite having spent 10 years on it and not getting much better. If, hypothetically, the sport had multiple "traps" making you think it was really hard to get better, is it really hard? It's like gatekeeping a restaurant so hard and charging so much that when people finally get in they have no choice but to feel that it's amazing. If it's never realistically possible to test 

When you run a dataset through some model, you often don't care what the truth is. Often, whatever process generated that dataset is stable enough that as long as your predictions come out right, you don't mind that we have no idea about the causal mechanisms behind the data, don't even care that our linear regression coefficients have nothing to do with the reality. Accuracy is accuracy. 

You're not always going to be right, but in the moment you have to believe it, there is no alternative. The mind is only capable of drawing straight lines, as much as we'd like to believe otherwise. But we have to make a decision and tolerate error. In that instance we should have full faith in ourselves. Taking a decision more tentatively in the hopes that somehow you are hedging, does not optimize for result -- it optimizes for the appearance of safety. 

In a way this conviction is truth, your truth, in that instantaneous bubble of the present. Then you take these sequences of linear steps and you find some sort of success on the other side.*

Truth can be local. And in a way -- what else is there? We fawn over objectivity but even if we were to assume an objective universe, only a subset of the paths are well-traveled, statistically. 

A good functional definition of "truth" is then "well-calibration". How we want to define calibration afterwards is up to us. 

(Want to formalize this notion. Logic -- Model theory? Localization of truth, global verification -- application of sheaf theory here?  Tarski did it with the hierarchy -- truth is just an operator, something we model, not interested in philosophy.) 

Paraconsistent logic -- if locally there is no resources, no path, no facts available to disprove a proposition, then it is true, locally. The question is whether this definition of truth, this semantics, admits a Boolean algebra.

* How to deal with this? There's the notion of local truth which indicates a semantics but also a global "goal" here, and also a more global "policy" like "believe in oneself" -- is this a predicate, and the individual beliefs in oneself, are local propositions?