Dream runs

Preview, then run

memcal dream --dry-run   # bundles, size, estimated cost — calls nothing
memcal dream             # run the pass
memcal dream --rounds 3  # repeat until the spool drains
memcal dream --no-sweep  # skip the final cheap sweep
memcal dream --mode nightly --model gpt-5.6-luna

--limit caps the item budget per pass; --redo un-claims processed items for reprocessing (merge-on-key, so corrections are not duplicated).

Retry a pass that failed

The Runs view shows whether each pass was ok, partial, failed, running, or priced only. Retrying releases the lines that pass claimed and runs them with the provider and model configured now:

memcal dream --retry 29

Only failed or partial runs with bundles are retryable. Lines beyond the model horizon stay marked as read.

Inspect what happened

memcal trace             # prompt, reasoning, reply of a real model call
memcal review            # what it wrote, what it wants to know, what it can't resolve
memcal stats             # post-gate volume and run history
memcal gatecheck         # what the gate is passing and rejecting

Schedule

memcal schedule          # when the pass last ran, whether one is owed
memcal schedule install  # nightly launchd job
memcal schedule status
memcal schedule run      # run now regardless
memcal schedule due      # just the answer

See Scheduling for the launchd layout and Monitoring for the Runs view.