Andrew Lee @candrewlee.com · Dec 3 The fact that this bogus “9.5% ghost engineer” study came from a student at Stanford should cure your imposter syndrome. The results of a small ML model trained to score PR difficulty is unreliable *at best*. But to extrapolate that across the whole industry? And all that to plug consulting…
1 likes 2 replies
Replies
Sometimes the most complicated bugs come down to a single line of code. I seriously doubt this model could capture the context on how complicated debugging a distributed system, a storage corruption bug in a database, etc. is. IMO you’d only analyze this way if you’ve never written code.
Using git commits per month is a totally flawed methodology too - Are they merging or squashing PRs? - This biases towards doing many tiny changes, rather than larger pieces of work (sometimes necessarily large) - Many at senior & staff level have more meetings and spend more time mentoring