Note: I try to have a lot of respect for my subscribers’ time, and mostly limit my articles to high-effort, high quality explanations, including some novel research component when possible. Going forward, promotional-ish posts like this one will be posted to Substack Notes, and not sent by email. But I don’t think anybody knows about Notes yet, and I want to make sure this makes its way to those readers who might find it helpful.
Want to chat about an idea? Need help with a project? Confused about Kalman filters? Just want to argue with me about inflation or AI risk? Have questions about your power electronics homework, debugging your 3D printer, or getting advice on an engineering career?
I’m not actually a full-time Substack writer; I work as a professional engineer for a living. (As is probably no surprise from my slow pace of publishing). I’ve opened some time slots for free, informal, 45-minute consultation sessions.1 Any topic of your choice, one-on-one (or invite your friends and co-founders). This coming Wednesday & Saturday.
Book a time slot here and let me know what you want to talk about: jbaylessconsulting.ca/booking
If you were hoping for a proper These Are Systems article — I’m sorry to disappoint! But if you’re a new subscriber, consider whetting your appetite with one of my other articles, linked below. They’re all free to read in full.
Causation does not even imply correlation: Everybody knows that just because X and Y are correlated doesn’t mean there’s a causal relationship between X and Y. But you might be surprised to learn that when feedback is involved, a causal relationship X→Y can actually eliminate any observable correlation between X and Y.
Inflation, Part 1: On Shelter Futures: In which I apply a system dynamics lens to the Canadian consumer price index, analyzing the rising cost of shelter and to what degree its impact on inflation is predictable well in advance. Published one year ago this month, but still quite relevant to the future, and so far the shelter model seems to be tracking quite well against reality — especially the predicted increases in rent. Also see Part 2 for somewhat more up-to-date graphs and a comparison with the USA. I still plan to write more in this series.
Reading Jaynes: Fear not the Unknown Unknown. Donald Rumsfeld famously cautioned about “unknown unknowns”. But no matter how uncertain you are, you should never be afraid to calculate probabilities: in fact, the less you know, the more probability theory can help you. And the next article in this list is an example of putting this philosophy into practice:
Task Estimation: Conquering Hofstadter’s Law: Frustratingly, everything takes longer than you expect, even when taking that knowledge into account. If you struggle to estimate how long your tasks take, or think time estimation is a hopelessly doomed and futile exercise, you might enjoy reading this.
Voiding the Warranty: How the Kalman Filter Updates on Data. Part of an in-depth technical series that starts with raw Bayesian estimation and later introduces dynamic-model-based prediction. The Kalman filter is well-known as an optimal linear predictor for systems with good-old-fashioned Gausssian noise, but its usefulness goes well beyond that. Here I demonstrate that — with slight improvements — a Kalman filter can work well even with “pathological” Cauchy noise, which is so random that it’s not even possible to calculate the mean or variance.
Thanks for your patience with today’s non-article, and I hope you’re enjoying These Are Systems!
As of this publishing, two slots are still available Wednesday and three on Saturday, but I’m not saying this in the “sign up quick before they’re all gone!” sense, because if there’s a lot of interest I’ll make more time available. You’ll just have to wait a bit.