I’ve been having a lot of fun vibe-coding with marimo notebooks recently. It also turns out that AI is extremely helpful in collecting data when I’ve got some random questions – stuff that is publicly available but usually too much of a pain in the ass to spend an evening researching.

Surprisingly, AI is quite effective at identifying public data sources and visualizing trends. I did find it still struggles with more complex topics – anything finance especially is a no-go – but it is particularly good at generating plots and doing some basic comparisons.

I’ve included a few random experiments I’ve done over the past few months, just for reference.

Airline Delays#

So news often publishes articles on the worst airlines for delays, but I’ve always found the analysis a bit limited. I don’t really care about a 30 minute delay, but I definitely care about overnight or 6hr+ delays.

Notably Southwest seems to be the best when it comes to avoiding longer delays – while American, ALlegiant, Frontier and American Eagle are the worst:

Long delays are also surprisingly most common in the summer, which may have to do with the frequent extreme weather in the south and southeast:

We can also check delay data specifically by airport:

Film Gender Representation#

There was a UK study recently that found that among the 100 highest-grossing films, it was more common for a leading actor to be named Chris than for it to star a woman over 60. I was curious about wider trends if we were to focus this on popular US Hollywood films (>=10k votes on IMDB between 2015-2025) and looking at the top two billed roles on each film. Interestingly, there is an overall gender gap in hiring:

What’s more interesting is the difference in age distribution by gender. You’ll see that women tend to be significantly younger, with women younger than 40 being much more common than older. Meanwhile for men, the two biggest buckets tend to be from 30-50, and men 50-59 are actually slightly more frequently than those younger than 30.

I also did a breakdown by the IMdB-classified genre to see if there are genre-specific biases.

Interestingly the fre genres that seem to defy the gender age distribution for women tend to be Action, Biography, Crime and Drama.

Cheaper Gas Purchasing Strategies#

In an AskReddit post on “dumbest things you’ve ever heard,” I saw a comment that said something like:

My friend told me that they weren’t affected by fuel prices because they only filled up to $50 at the pump.

Which is a pretty ridiculous claim. But it got me thinking: Could that actually save money? Logically, it could potentially “reduce the hurt” of expensive fuel and creating more opportunities to visit when it’s cheaper. The question is… how much does it really save?

I found a blog post that found gas prices could be modeled as a random walk, something not unexpected for a commodity market. If we make a generic assumption that price at the pump follows this mostly with a premium, we can test our refueling strategies in a simulator to determine the magnitude of effectiveness.

For this simulation, I came up with three strategies:

  1. Fill up: When at 1/4 tank, fill up the tank to full at the current price.
  2. Budget fill: When at 1/4 tank, fill up the tank with $50 of gas.
  3. Dip Buyer: Sort of a mix of the two. Follow the budget fill strategy by default, but if gas prices are cheaper than 10% from last purchase, do a fill up.

I ran this on 5000 simulations, assuming a 15-gallon tank, a 24 MPG gcar, a budget fill of $25, over 5 years, with miles driven per day. Results are below:

Strategy Avg $/gal Avg Stops
Fill Up $4.465 57
Budget Fill $4.458 115
Dip Buyer $4.456 113

Interestingly it did actually save money – I did a paired t-test to confirm that budget filling is at least different from the fill up average – but by less than $0.01, and that’s under very favorable conditions as well.

I found that if I increased daily miles or increased the budget fill number to $50, the difference approached closer to ~$0.003 in savings. In order to realize an approximate $0.01, you need to visit the pump much more frequently.

Because this one is mostly a table, here’s a cool plot the agent generated of what the random walk oil prices look like: