I'm not sure it's good to have these points, here's why:
Low-cardinality columns
It depends. WHERE deleted = TRUE will use an index if the vast majority of rows has deleted=FALSE.
Small tables (< a few thousand rows): full scans are faster than index lookups for tiny tables.
Not true in the case of covering indexes. But the real problem is: Claude may run on the developer's laptop, where ALL tables are probably very small. Even in staging, all the tables might be very small. But in production they can be big.
Write-heavy columns
This might confuse LLMs. They don't seem to me very good at inferring which operations will run often.
I'm not sure it's good to have these points, here's why:
Low-cardinality columns
It depends.
WHERE deleted = TRUEwill use an index if the vast majority of rows hasdeleted=FALSE.Small tables (< a few thousand rows): full scans are faster than index lookups for tiny tables.
Not true in the case of covering indexes. But the real problem is: Claude may run on the developer's laptop, where ALL tables are probably very small. Even in staging, all the tables might be very small. But in production they can be big.
Write-heavy columns
This might confuse LLMs. They don't seem to me very good at inferring which operations will run often.