A controller is programmed once and runs; an entity develops. The clearest fingerprint of development is the U-shaped curve. A child says “went” correctly, then — having discovered the “-ed” rule — regresses to “goed”, then re-masters “went”. OIST saw exactly this in their curiosity-driven robots: 9 of 10 trained with swapped exception-mappings dipped and recovered on the exceptions (p = .0001). Below, at nano, a small net reproduces it on 8 of 8 seeds. The driver is not a trick: it is vocabulary growth (Plunkett & Marchman). Early, the learner knows few items and memorizes the high-frequency irregulars. Then the regular vocabulary floods in, the rule crystallizes, and — with no spare capacity — it overwrites the exceptions it had memorized: the dip, “goed”. Continued exposure re-carves them: recovery, “went”. The tell that this is development and not random forgetting is decisive — the very moment the exceptions dip, the regulars leap (accuracy ~0.3 to ~1.0). One event, opposite effects: the rule generalizing at the exceptions’ expense. You cannot get a U from a lookup table. You get it from a system reorganizing its own representation as it learns, which is what an entity does.