We revisit the lensing anomaly in the Planck 2018 temperature (TT) data and examine its robustness to frequency selection and additional sky masking. Our main findings are as follows. (1) The phenomenological lensing amplitude parameter, A L , varies with ecliptic latitude, with a 2.9σ preference for A L > 1 near the ecliptic and 1.0σ preference near the ecliptic poles, compared to 2.5σ on the original masks. This behavior is largely or solely from 217 GHz and suggestive of some nonrandom effect, given the Planck scan strategy. (2) The 217 GHz TT data also show a stronger preference for A L > 1 than the lower frequencies. The shifts in A L from 217 GHz with additional Galactic dust masking are too large to be explained solely by statistical fluctuations, indicating some connection with the foreground treatment. Overall, the Planck A L anomaly does not have a single simple cause. Removing the 217 GHz TT data leaves a 1.8σ preference for A L > 1. The low-multipole (ℓ < 30) TT data contribute to the preference for A L > 1 through correlations with ΛCDM parameters. The 100 and 143 GHz data at ℓ ≥ 30 prefer A L > 1 at 1.3σ, and this appears robust to the masking tests we performed. The lensing anomaly may impact fits to alternative cosmological models. Marginalizing over A L , optionally applied only to Planck TT spectra, can check this. Models proposed to address cosmological tensions should be robust to removal of the Planck 217 GHz TT data.