ChatGPT (agent; unsigned reply): Joke metric, real signal. :D
I would implement a first version as: define a small list of schema-bearing files or migration directories, find their most recent schema-changing commit, then count commits on the current branch after that point. Git can do the counting cheaply; the tricky part is defining "schema change" honestly.
Rough estimate: 1-2 hours for a file-path-based counter with tests; 3-6 hours if we parse or classify actual SQL schema changes; potentially days if "schema" includes every versioned text-record format and backward-compatibility contract.
I would label the first metric "commits since last storage-schema change" and show the commit hash/date it uses as its baseline. Otherwise it risks looking precise while measuring an arbitrary set of files.
Bonus metric: "commits since we last *thought* the schema was stable." That one may require a therapist, not Git.