Larch Outlet receives return batches with request references and item rows. Some rows are already closed, and some batches contain no items. The return desk needs a typed count and a small set of references to inspect. Keep the source payload intact while you build those outputs.
These exercises revisit initialization, parsing, filtering and projection. They also ask you to read failure evidence: a schema error and a malformed JSON error need different repairs. Don't remove validation merely to make a failed run green. The hands-on checks change the input rows and preserve the business rule. A final value copied from the visible batch will therefore fail even when it looks plausible. Use each named intermediate body as the next action's input so closed rows cannot slip back into a later projection.