Here is the thing about the reading. It's wrong. And it isn't close.

We ran the line again. Same probe. Same depth. Same nothing.

So we looked harder. One plot. One transect. Thousands of samples. No gradient anywhere in the record.

The estimator is blunt. On purpose. It sees totals. Nothing else. No order, no spacing, no structure.

It holds anyway. Four days out of five. That's the pattern.
