In my panel data set analysis, I interact the treatment variable with the running variable to see whether the effect of the treatment differs for different values of the running variable. However, I get a highly significant and large in value effect for the value of the running variable for which treatment is never applied. How to explain this?

I investigate the U.S. House, 1980-2021. Control group are one-district states, their numberofseats=1. Treatment group are all the other states, their numberofseats depends on the year but is always >1. census takes value 1 in redistricting years (1982,1992,2002,2012), 0 otherwise. Cregions is a control for the region in which a state is.

outcome variable is a valuation of representative's extremism, measured in the scale (0;1) (but no actual 0's or 1's).

Code:
 
gen treatm=1 if census==1 & numberofseats>1
replace treatm=0 if treatm==.

reghdfe extrem1 treatm##numberofseats i.Cregions, absorb(congress) vce(cluster state_icpsr)
margins numberofseats, dydx(treatm)
in the margins output, the marginal effect for numberofseats=1is huge and highly significant (and is identical to the coefficient of treatm in the reghdfe output). but treatm is always 0 for numberofseats=1. could someone please explain how to interpret this? or if i'm doing something wrong?