I’m writing a makefile.

I have an artifact data and I use gpt to process it. Due to token limits, I have to break the data into segments (rule segments), each processed by AI (rule ai), then assemble segments together (rule assemble).

segments : 
    break --size 20M ./data.txt seg-%d.txt

ai : segments
    for f in seg-*; do
                # outputfile is ai-output-%d
        gpt $f
    done

assemble : ai
    concat --output final.txt -- ai-output-*

This makefile works when I call make assemble.

However, gpt is not very stable. Sometimes I have to call it again to get a better output. Then, if I call make assemble, make will re-run rule ai, and overwrite my better output.

I would like to use this makefile in two cases:

1. When the folder is in a clean state, I call make assemble, and final.txt is created.
2. When the folder is not clean, I manually rerun chatgpt and get a different ai output. Then I call make assemble, and final.txt is created from my revised output.

How can I change my makefile to allow this two cases?