Why I'm writing this

I have a folder of unfinished thoughts. Most of them are three paragraphs long, written in a burst of enthusiasm at eleven at night, and abandoned by the weekend. A few of them are genuinely good. None of them have ever been read by anyone, including me.

This site is an attempt to fix that, and the fix is embarrassingly simple: one small article, every week, published whether or not I think it’s ready.

The problem with keeping it in your head

There is a particular feeling — you’ve had it too — of reading an explanation and thinking yes, obviously, I understand this now. It’s a real feeling and it’s almost entirely unrelated to whether you understand anything.11 Psychologists call this the illusion of explanatory depth. People confidently claim to understand how a zip fastener works, right up until they’re asked to explain it step by step.

Writing destroys that feeling, efficiently and without mercy. The moment you try to put an argument in order, the gaps announce themselves. You discover that the step you were gliding over is the entire problem. This is unpleasant, which is precisely why it’s worth doing on a schedule rather than when the mood takes me.

There’s also the decay problem. Whatever I learned last March is mostly gone now. The standard model of forgetting has retention falling off exponentially with time,

where $S$ is a stability parameter that grows every time you retrieve the memory. The relevant point isn’t the exact functional form — it’s that $S$ only increases when you do something with what you learned. Writing it down is the cheapest available something.

Why weekly, and why small

The obvious failure mode is the ambitious one: announce a serious publication, plan a definitive 8,000-word treatment of some topic, spend four months on it, publish nothing. I’ve done this. It feels productive, right up until you notice a year has passed.

The alternative is to make each individual article so small that not finishing it would be absurd. A week is the right unit — long enough to have something to say, short enough that I can’t polish a draft into silence.

The arithmetic is the whole argument. If each article is worth some small increment $r$ of accumulated understanding, and they compound at all, then after $n$ weeks you have

which for a very modest $r = 0.02$ and $n = 52$ is a factor of 2.8 over a year. The intuition that matters: frequency dominates magnitude.22 This is not a rigorous model — understanding isn’t a scalar and it certainly doesn’t compound cleanly. But it’s directionally right, and the directional point is the one I need. Fifty small things beat one large thing, and it isn’t close.

def accumulated(rate: float, weeks: int) -> float:
    """Compounding at `rate` per week, for `weeks` weeks."""
    return (1 + rate) ** weeks


for r in (0.01, 0.02, 0.05):
    print(f"r={r:>5} -> {accumulated(r, 52):.2f}x after a year")

r= 0.01 -> 1.68x after a year
r= 0.02 -> 2.80x after a year
r= 0.05 -> 12.64x after a year

Even at one percent a week — an almost invisible improvement — a year of it is a two-thirds gain. The catch, of course, is that skipping weeks doesn’t just cost you those weeks; it costs you the exponent.

What I’ll be writing about

Three rough directions, though the boundaries are porous and the interesting questions usually sit between them:

What ties them together, for me, is that all three are fundamentally about systems with too many parts to track individually, where you have to find the right level of abstraction or drown. The honest reason I keep coming back is simpler: each one has moments where a short derivation suddenly makes a messy world feel tractable, and I want more of those moments written down before they fade.

How I want to write

I’ve spent a lot of time reading Distill, and what I took from it wasn’t the interactive diagrams. It was the seriousness about the reader’s time: real figures instead of decorative ones, notation defined before it’s used, and a willingness to say this part is unclear to me rather than papering over it.33 Distill stopped publishing in 2021, citing the unsustainable effort of producing articles to that standard. That’s worth remembering as a caution — the standard that produces nothing is not a good standard.

Concretely, three rules:

1. Figures earn their place. If a plot doesn’t change what you believe, it shouldn’t be there.
2. Uncertainty gets stated. “I think” and “I’m not sure” are load-bearing phrases, not weakness.
3. Length follows content. Some weeks that’s four hundred words. Good.

The best way to have a good idea is to have a lot of ideas, and throw away the bad ones.

The corollary is that you need somewhere to put them while you’re deciding.

What this isn’t

It isn’t authoritative. It’s a notebook that happens to be public, written by someone working things out. I will be wrong in public, on a weekly schedule, which is a strange thing to volunteer for but seems better than the alternative of being quietly wrong in private.

It also isn’t a funnel. There’s no newsletter popup, no cookie banner, no tracking beyond an anonymous page count. If you want to follow along there’s an RSS feed, which is the correct technology for this and always has been.

The failure modes I expect

Naming them makes them slightly less likely:

• Week four. The novelty is gone and nothing feels worth writing. The answer is to publish something short and bad rather than nothing.
• Scope creep. An article starts wanting to be definitive. The answer is to split it and publish the first half.
• Silence as evidence. No one reads it, and I conclude it isn’t worth doing. But the primary reader is me in six months, and he’s a captive audience.

So: one small article a week. Let’s see how far this gets.