What decides whether behaviour pays: movement, not the signal it follows

The previous article of the series reported that local resource-greedy movement helps in a patchy pond world but that its advantage shrinks as the rich zone fills. This article asks what decides whether a movement rule pays at all, and two of the three expected answers turn out to be wrong. The sign of the advantage depends on the resource recovery rate: with slow recovery the greedy rule loses to a non-moving control (gap -0.078 to -0.205, all three seeds negative), with fast recovery it wins (+0.036 to +0.101). The cause of the loss is a trap of inaction - the step rule fires only when the cell's prediction exceeds the actual resource, and the prediction tracks the falling resource, so the cell stops noticing that it is in a bad place. Removing the adaptation flips the sign sharply (+0.338 instead of -0.174), but with the number of moving cells held equal the two references are indistinguishable (0.238 against 0.246): what matters is how much a cell moves, not what it compares with. A continuous diffusing signal field was tested in four designs; only the memory variant gives a gain (+0.038 to +0.045), and plain smoothing of the resource matches it (0.299 against 0.297). A cell moves 0.14 cells from its birth cell in a whole lifetime, so long-range information cannot be used. Every printed number is re-derived from the deposited artifacts by a machine check (26 checks, 0 failures).