Discrepancy in Var of sigma in variance_matrix_ of LogNormalAFTFitter result

[vpport]

Hi,

I would like to start by commending the creators of this package. It has been really useful to my work. Now for my question, I am currently trying to calculate confidence intervals for estimated quantiles at use conditions for our devices using the LogNormalAFTFitter. I am trying to implement the method described by Meeker and Escobar in their book Statistical Methods for Reliability Data. I used a test dataset from the book for this as i could use the calculation in the book as a guide. For this calculation the variance matrix is used. Now when i do the analysis there is a discrepancy in the variance of sigma (result.variance_matrix_.values[2,2]) as calculated by lifelines and that in the book. In lifelines the var_sigma is 0.0294 while in the book it is 0.00287. I confirmed the result by doing the same analysis in Minitab and i got the value for 0.00287 there as well. All the other values in the variance_matrix_ are correct except that one. Do i need to transform the variance in some way or something as i sometimes need to do with the parameters as well? I've tried taking the log and exp and sqrt of the number but nothing works. Is it maybe some kind of bug or am i missing something? Here below is the test dataset if someone wants to replicate the analysis.
I used:
result = lifelines.LogNormalAFTFitter().fit(df = data, duration_col = 'seconds', formula = 'load')

seconds	load
123	2.74
704	5.47
184	2.13
113	1
94	0.32
76	0.31
78	0.51
98	0.29
240	0.96
110	0.6
213	2.1
284	3.1
317	5.86
142	1.18
127	0.57
96	1.1
111	1.89