Data problem fixed.
Root cause of failure: Autocorrect. Not even joking.
It sounded like a fun one. I was trying to figure out how you could disentangle the data in the shower earlier, like if the two data sets were dissimilar enough you could use upper and lower boundings or if it was periodic time series and both were out of sync could isolate them that way.
What we did was go though every single field, one at a time, checking where the data was being taken.
Two "gas meters" were actually pressure transducers ("Heat a massive building for $16 a month with this one simple trick!), and the rest were autofills that entered the location of the last location entered.
All the billable stuff is done. Next are the water treatment plants and the plant proper. Shouldn't take more than 4 months.