I don’t know if this is a Stan output issue or an rstan issue . . .

I fit a model and here’s part of the result, inference for a 2x2 correlation matrix:

```
mean se_mean sd 2.5% 25% 50% 75% 97.5% n_eff Rhat
```

Sigma_eta[1,1] 1.00 NaN 0.00 1.00 1.00 1.00 1.00 1.00 NaN NaN

Sigma_eta[1,2] -0.19 0.03 0.49 -0.92 -0.60 -0.27 0.16 0.82 216 1.02

Sigma_eta[2,1] -0.19 0.03 0.49 -0.92 -0.60 -0.27 0.16 0.82 216 1.02

Sigma_eta[2,2] 1.00 0.00 0.00 1.00 1.00 1.00 1.00 1.00 4087 1.00

Sigma_eta[1,1] and Sigma_eta[2,2] are fixed to be exactly 1, hence the zero variation. That’s fine, but then: (a) why the NaN in the output for Sigma_eta[1,1], and (b) why the weird value of 4087 for effective sample size for Sigma_eta[2,2]? Seems like some bug.