Data format of unrestricted baseline in the design matrix for survival data with binary encoding I am reading Tutz & Schmid "Modeling Discrete Time-to-Event Data" (2016) section 3.4 Estimation [of basic regression models]. It presents specification of the data structure for expressing survival models as models with binary dependent variable. That makes them amenable to the widely available fitting routines for such data.

I wonder if the identity matrix in red should not be appended by some columns of zeros to the right, so that the total number of columns is $q$ where $q$ is the maximum time. Otherwise, the rows of the data matrix would be of different length for different individuals, depending on when they experienced the event. E.g. according to the current setup, there would be a single column with a 1 in it if an individual experienced the event in time period 1. Meanwhile, another individual that experienced the event in time period 2 would have an identity matrix of dimension 2, etc etc.
 A: I don't find the displayed data format to be helpful, because of the reason you cite and because that's not the format ultimately used for binomial regression.
In that format, the subscript $i$ represents an individual. The "augmented data matrix" examples are for single individuals, each under a scenario with a single final event time or right-censoring time, indicated as $t_i$.
If you wanted to stack such "augmented data matrices" among individuals in this format, I suppose you could proceed as you indicate. I haven't thought that through carefully, as that data format isn't what's used in practice. In the R discSurv package, a companion to this book, I don't see any data examples in the format that you show, or any tools for handling that format.
To combine such "augmented data matrices" among individuals, the matrices first are converted to the format displayed lower on page 134, with $t_i$ rows for each individual and the observation time explicitly included in each row:

The top is for an observed event at time $t_i$, the bottom for a right-censored observation. That's the format ultimately used for modeling discrete survival data with binomial regression.
