A county public-health department wraps Claude Code in a small service that summarizes de-identified immunization-clinic notes. The dashboard team wants to show progress in the browser as the run proceeds, so they need machine-readable events emitted incrementally over the life of the run rather than one payload delivered at the end. Which --output-format value produces that behavior?

In headless mode the output format decides two things at once: the shape of what comes out and when it arrives. A live progress view is a timing requirement as much as a parsing requirement, so the choice has to land on a format that is both machine-readable and incremental.

stream-json emits newline-delimited JSON event objects over the life of the run, describing what the model and its tools are doing as they do it. A consumer parses each object on arrival instead of blocking until the process exits, which is exactly what a browser dashboard needs, and the same event stream gives the wrapper service intermediate steps to log and audit rather than only a final answer.

text does stream, but it streams prose written for a human reader and carries no event structure a dashboard can parse reliably; json is machine-readable yet delivers a single result object after the run completes, which is the batch behavior the team is trying to move away from; and markdown is not one of the available output-format values, and a progressively rendered prose format would in any case reintroduce the parsing problem structured events exist to solve.

Exam caveat: streamed events are a progress feed, not a result contract, and event shapes are richer and more incidental than the final result, so a dashboard that hard-codes assumptions about every event type is fragile. Operational check: consume the stream, assert that objects arrive before process exit, and have the dashboard read progress from events while taking the authoritative summary from the final result event.