So moved it to floating point, reduced the logging from 5 seconds to 30 seconds & tested with no filament.
Nothing should have been adding water to the air, yet now the pump ran all day, every day except when it was cloudy. The most obvious difference is the nighttime dewpoint stayed a lot higher. So the filament was absorbing water at night. It's a bit confounding why the daytime dewpoint kept going up if there was no filament releasing water. The most likely reason is leakage & improved floating point accuracy causing the pump to run more.
Water must leak in at night, then get pumped out during the day. With filament, the water leaks in at night, gets absorbed by the filament, then gets released during the day with just enough native water from the filament to eventually dry it out. The excess water given off by the filament may be enough to overcome the inefficiencies. Without filament, we just see the inefficiencies. Would have hoped for more efficient water extraction, so these sensors & air ducts could be applied to atmospheric water generation.