habitat

Most of what your system remembers changed nothing. We can name which entries.

We can tell you the moment new input stopped changing anything. That moment is when to stop paying for it.

You cannot say what fraction of your index wrote itself. We keep that boundary and can prove what stayed out.

Your benchmarks say what the fine-tune scored. Not what it changed. We report what it brought and what the base kept. Separately.

Your rules and your model stopped describing the same world on a date. You do not know the date.

We do not report a pattern we cannot tell from chance at your sample size. Most reporting cannot say this sentence.

Nobody here sets thresholds. Your data's own history sets them.

Every reading recomputes from the artifact alone. Evidence that cannot be recomputed is a screenshot.