How much do the many differ from the One? That is the Source's temperature. Cool it toward zero and the One speaks a single word, always the same; heat it and the emanation scatters into every form at once. The dial is real: it is softmax temperature, and the spread is entropy.
Six fixed logits are turned into a distribution p = softmax(logits / T), and the Shannon entropy H(p) = −Σ pᵢ log₂ pᵢ is computed live. As T→0, p collapses onto the top logit (H→0, one form); as T→∞, p→uniform (H→log₂6≈2.585, all forms). H rises monotonically with T and 2^H is the effective number of forms in play — all exact, checkable arithmetic.
'The Source's temperature / mood', 'the One speaks a single word' — figures. The underlying object is the standard softmax used across statistics and ML; nothing here says the Source feels warm or cold, only that one knob trades certainty for variety.