With the I2C sensors debugged & the mechanicals tested, it became clear that the pump needs a fuse.  The valves burn 200mA at 4.5V.  The pump burns 500mA at 4.5V.  The pump is air tight & doesn't need its own valve.  It's so air tight it stalls, so this would entail reading a 1R resistor with an ADC.  This would entail increasing the clockspeed to run the fuse, which would entail adding delays to the I2C code.  The best fan ended up being a laptop CPU fan at 5V, so everything would run directly on 5V instead of a buck converter.

The sensor generates temperature & humidity values.  The PIC computes dew point readings.  Quite a variation in humidity, even when the sensors are right next to each other. No-one outside China knows how the AHT20 works.  It only says it's a capacitive humidity sensor.  For most of the lion kingdom's life, humidity could only be measured by wet & dry thermometers.

The mane problem is determining when to replace the air in the container for the optimum drying.  The easy way is to use humidity.  When the humidity inside is higher than outside, turn on the pump.  This would only happen at night, when the temperatures were equal.

Daytime heating of the container is going to keep the inside humidity always below the outside humidity.  After taking water out of the filament, the hot, less humid inside air won't be as dry as it would be by ingesting cold, more humid outside air.

A dew point system would ingest air at night when the dew point was lowest, but would also ingest air during the day as water came out of the filament.  Ingesting air during the day would cool it down & decrease the rate of drying, but it might take more water out of the filament in the long term.  Depending on the pump rate, it may not cool down at all.  There might be a way to compare systems by measuring how long it takes to empty a glass of water.