Rstan throwing inappropriate error message

Summary:

Rstan throws the error Error in if (any(efbmi_per_chain < threshold)) { : missing value where TRUE/FALSE needed, which apparently, should not be the case (ref).

Description:

It happens when I try to fit the model with some data, but not others.
I am trying to model count C with two covariates: Treatment and Mice.

Reproducible Steps:

Code to get the data from dput

structure(list(Treatment = c(1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 5, 
5, 5, 5, 5, 6, 6, 6, 6, 6, 4, 4, 4, 4, 4, 3, 3, 3), C_error = c(6227605.5, 
5844779.25, 1338474, 10365561.75, 6349542.75, 9831545.25, 10968084, 
15513791.25, 2564861.25, 6725652.75, 1609213.5, 716698.5, 991467.75, 
822666, 521778, 11940, 345215.25, 42237.75, 7611.75, 10746, 447.75, 
447.75, 149.25, 597, 149.25, 14477.25, 1492.5, 7910.25), C_good = c(1008.15, 
216.45, 8583.9, 5557.5, 3956.55, 2195.7, 16155.75, 9490.65, 8712.6, 
3281.85, 290.55, 651.3, 146.25, 165.75, 347.1, 144.3, 95.55, 
42.9, 50.7, 72.15, 152.1, 87.75, 95.55, 101.4, 101.4, 183.3, 
95.55, 801.45), Mice = 1:28), row.names = c(NA, -28L), class = c("tbl_df", 
"tbl", "data.frame"))

My stan model

data{
    int C[28];
    int Mice[28];
    int Treatment[28];
}
parameters{
    vector[6] bT;
    vector[28] bA;
}
model{
    vector[28] lambda;
    bA ~ normal( 0 , 5 );
    bT ~ normal( 5 , 10 );
    for ( i in 1:28 ) {
        lambda[i] = bT[Treatment[i]] + bA[Mice[i]];
        lambda[i] = exp(lambda[i]);
    }
    C ~ poisson( lambda );
}

Current Output:

The model runs when I use data = list(C = d$C_good, ....)
But when I change the count to data = list(C = d$C_error, ....), the model samples and throws the error

Running MCMC with 1 chain, with 1 thread(s) per chain...

Chain 1 Iteration:   1 / 1000 [  0%]  (Warmup) 
Chain 1 Iteration: 100 / 1000 [ 10%]  (Warmup) 
Chain 1 Iteration: 200 / 1000 [ 20%]  (Warmup) 
Chain 1 Iteration: 300 / 1000 [ 30%]  (Warmup) 
Chain 1 Iteration: 400 / 1000 [ 40%]  (Warmup) 
Chain 1 Iteration: 500 / 1000 [ 50%]  (Warmup) 
Chain 1 Iteration: 501 / 1000 [ 50%]  (Sampling) 
Chain 1 Iteration: 600 / 1000 [ 60%]  (Sampling) 
Chain 1 Iteration: 700 / 1000 [ 70%]  (Sampling) 
Chain 1 Iteration: 800 / 1000 [ 80%]  (Sampling) 
Chain 1 Iteration: 900 / 1000 [ 90%]  (Sampling) 
Chain 1 Iteration: 1000 / 1000 [100%]  (Sampling) 
Chain 1 finished in 6.2 seconds.
Warning: 444 of 500 (89.0%) transitions hit the maximum treedepth limit of 10.
See https://mc-stan.org/misc/warnings for details.

Error in if (any(efbmi_per_chain < threshold)) { : 
  missing value where TRUE/FALSE needed

RStan Version:

‘2.26.13’

R Version:

"R version 4.1.3 (2022-03-10)"

Operating System:

Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 19044)

Start by reproducing the issue with the supplied R data and Stan model in RStan 2.26.13, comparing the C_good and C_error inputs. Trace the post-sampling diagnostic path that evaluates efbmi_per_chain, then determine what diagnostic or error should be reported when the sampled result is invalid. Done means the reproduction no longer produces an inappropriate missing-value error and the relevant behavior is covered by a test.