Nobody Abandons Error Analysis Because It Does Not Work
Ask a team why they stopped doing and you get a version of "it does not scale". Reading traces by hand is fine for ten and impossible for a thousand. So they automate the reading, and nobody on the team looks at an actual failure again.
Watch what actually happened. Somebody opened a JSON file with a few hundred traces in it. They read one. They scrolled. They lost their place. They read another. They tried to remember which codes they had already used and typed a slightly different phrase. After ten traces they closed the file, and the experience became a belief about the method.
One of those is a claim about qualitative research. The other is an afternoon of work that nobody scheduled.
The real problem is the tool, not the method. This lesson builds the tool, and it genuinely is about a hundred lines with no dependencies and no server.
That is the whole thing, running. One trace, in order, with a numbered list of the codes already used and an invitation to type new ones.