Insurance and Odds

"But driving slower and within the law reduces your odds of getting into an accident", they say. While this may be true, there's an inconsistency in the model. Let's take me as an example:

1. I've been driving for just about 18 years now.
2. I've been in 1 accident, which was entirely attributed to the guy who rear-ended me.
   2a. In other words, I've caused zero accidents, ever
3. I've been given 3 speeding tickets
   3a. The first was dismissed by my taking a defensive driving course
   3b. The second one stuck on my record for 3 years, costing me about $1000 in added insurance
   3c. The third was dismissed under Washington state's deferral program.

The argument goes that since I have been confirmed speeding, my odds of getting in a wreck are higher. I would argue that since I've had almost 18 years of fault-free driving, my odds of getting in a wreck are no higher than if I'd never been [caught] speeding.

The core issue is that the argument only holds if I phrase it as:
"If I drive within the law, my odds of getting into an accident are lower than if I drive faster". It can only be compared by keeping the driver the same person. In a giant population model, certain factors are correlated because, without keeping the driver constant, they correlate with more accidents. However, after a large enough sample size is collected on a particular driver, the best predictor of future accidents is their past record of being in accidents.

The now-false statistical argument goes that "every time out has an independent odds of getting into an accident, and riskier driving habits increase those odds". However, the past track record establishes a likely upper bound on said risks each time out, which would then be applicable to future trips as well. These independent odds are now the same as anyone else's who has been driving without faulted incident for a similar period of time, regardless of their driving style. And yet, if they are a woman, or older, or drive a Cadillac, their rates are lower than mine. And that's just bad math.

Demoing the house

It turns out that demolition is not as awesome as it might seem. There aren't always sledgehammers.

We recently wanted to remove our old trim and floors so we could put in new wood floors and update our trim. Turns out it's a bitch. Removing trim is a pain in the ass and it's pretty easy to mess up the existing paint, which then leads to needing to do fixup work. Removing the floors is more straightforwards: prybar, circular saw, etc. All in all, there's a lot of trash and it takes three times as long as expected.

Kota postulate #2

Many one-hit-wonder bands have a barely-known song that is as good or better than the one they are famous for.

Examples:
A-Ha - Take On Me, The Sun Always Shines on TV
Lipps, Inc - Funkytown, Rock It

... and probably lots of others, but I'm bad at coming up with "examples of topic X".

Kota postulate #1

(at the bars, clubs, ... )

The number of "whooOOOOoooooOOOOOOOOoooooo" are inversely proportional to the amount of fun people are really having. People really having fun sounds different. The end.

Crepes, Hungarian style

or as we call them, palacsinta.

3/4 cup regular flour
1/4 cup gluten
1 egg
~1 cup milk
~ 1 cup soda water

Mix flour, gluten and egg. This mix will still be quite dry.
Add milk slowly, while stirring. The goal is to get to a smooth, but fairly thick consistency. The batter should flow, but slowly, like a thick paint.

Let sit until you are ready to go.

Stir up the batter in case it's settled. We always want to keep it nice and smooth. Now add soda water, while mixing, until the batter is quite runny. It should be runnier than oil.

Place a thin skillet/griddle on medium-high heat. The ideal temperature should be right around the smoke point of olive oil. When you add a half teaspoon of oil you should see some smoking. Ladle in some batter, maybe 100mL? Tilt the griddle this way and that to let the batter spread. Once the batter cooks and comes unstuck from the griddle (~20 seconds?) give it a flip. Let finish for about 10 more seconds. The first side will be goldened, the latter should have small charred spots.

Add sugar, jam, cinnamon, whipped cream, or whatever else you like asap while it's still hot. Roll up and stick a fork in it.

Perfection!

Cycling out the bottom

Microsoft's performance model has made the news a number of times. Recently, there was the perception that those getting the lowest reviews are insta-fired. I'm certain this isn't completely true, but there's some validity to the statement. This, of course, pisses off the "all-star degenerate case" supporters. In reality such teams don't really exist, but let's say, for the sake of argument, that everyone on the team is really good, and the bottom guy out of, say, 30, gets axed and replaced.

Naturally this sucks for the last-place really good guy. He needs to go find a new job. But remember, he's really good, right? So he should have no problem, really. Unless he's not, in which case perhaps the replacing was warranted.

Let's consider this from the company's standpoint.
There's really no way to grade/verify 'absolute ability'. In fact, this is a ridiculous concept. You can have a collection of very good engineers, generally speaking, but their relative strength will depend on the particular projects/work the company does. They have no other alternative than to stack them by "what impact do they provide me". Even if you could assign an absolute score, you can't just say "everyone over 85 stays" because you may not be able to get enough 85s, or you can find them in droves.

Consider a football team. There are 53 guys on the roster, the last 10 or so of whom are consistently in flux. These guys are all elite players on an absolute scale, but the bottom 10 don't make nearly the same impact as the top set, or are limited in up-side. It's clear that roughly that level of quality is easy to find. Since these bottom 10 guys aren't making an impact, your odds are better by taking a shot at someone else. Based on your demonstrated hiring ability, the replacements should be at least about as good and there's a chance they'll be better. And no fan has an issue with this because football's all about results.

Now back to Microsoft. They have a demonstrated ability to screen hires to a level where 29 engineers are better than the worst 1 (this is simple math, but putting it this way proves that 97% of their hires are better than the guy about to get fired). So they let one guy go, and have a 97% chance of replacing them with someone better. And oh yeah, Microsoft (and every other company) should be (and is) all about results. Why should this be different than football?

The real trick is to find the proportion to let go each year, based on the rate of hiring and growth.

The All-Star engineer squad

Hey you, get a team of 20 people to work on the next great piece of software. You want the 20 best people you can find, right?

I don't think so.

Given a project of sufficient size, there will always be areas that "just need to get done" and have relatively straight-forward solutions. It's best to have people who don't wish to be the top dog on these: they will do solid work, not cause turmoil with egotistical design discussions, and simply just get things working.

The simple truth is: there isn't enough interesting/novel stuff to go around to warrant an All-Star at every spot.