Implementation details of --merge mode in StringTie

[Bathroomboss]

Dear Dr. Pertea,

I’m very interested in how StringTie implements the --merge functionality. I’ve read the original StringTie papers but couldn’t find a detailed description of the merge algorithm.
From my understanding, StringTie converts input GTFs into pseudo-BAM records, processes them locus by locus, builds a splice graph per bundle, and then derives a non-redundant set of consensus transcripts supported by the input structures.

However, I’m particularly interested in the low-level algorithmic choices during this process, for example:

1. How exactly are exon nodes merged when building the splice graph?

2. What criteria determine whether two similar isoforms are merged into a single consensus transcript?

I’d greatly appreciate any clarification or pointers to documentation/papers that describe the merge strategy in more detail.
Thank you sincerely for your time and for making StringTie such a robust and widely used tool.

Best regards,
Yu