Uniform Discrete Diffusion models (UDMs) usually incorporate explicit time $t$ as a conditioning variable, yet this step is often unnecessary in practice. We first show that the population‑optimal UDM predictor generally depends on time: time controls how much the model should trust the observed context, which can be expressed as $$\hat{y}_t = f(x, t)$$ where $t$ modulates a weight $w(t)$, making the predictor rely more on the context at low noise and more on the prior at high noise. We then argue that in finite‑data regimes typical of language tasks this dependence becomes negligible. When a corrupted training sequence $\tilde{x}$ remains much closer to its original clean sequence $x$ than to any competing training sequence, the empirically optimal predictor is almost insensitive to $t$ over most of the diffusion trajectory, with sensitivity only increasing near the high‑noise endpoint. Empirically, trained language UDMs exhibit limited time sensitivity across most of the trajectory, and time‑agnostic predictors remain competitive with, and often outperform, time‑conditioned models across datasets and training objectives. Review: These results suggest that although the theoretical optimum depends on time, explicit time conditioning is frequently superfluous for practical language modeling, allowing simpler architectures and reduced computational cost.