A LINQ query produces different results on a second loop and repeats expensive work. Why can that happen, and how would you fix it?

Materialization is not automatically better: it consumes memory, delays the first result, and can turn a bounded streaming operation into a large allocation. Some operators buffer even when the outer API is deferred; OrderBy, grouping, and set operations may need substantial state before yielding results. I keep side effects out of selectors because execution count and timing should not control business writes. In production I would inspect the concrete source, make one-pass ownership explicit, and add tests with an iterator that counts enumerations. For IQueryable, the same syntax has provider translation and remote execution concerns, so I keep that boundary close to data access rather than treating it as ordinary in-memory LINQ.