Batches may take up to about 24 hours, but the clerk SLA is 30 hours end-to-end. What must submission cadence account for?

Batches may take up to about twenty-four hours, while a clerk SLA of thirty hours end-to-end leaves limited slack. Submission cadence must therefore be planned against the SLA given the batch processing window: enqueue early enough that a worst-case window still leaves time for post-processing, exception handling, and clerk consumption. Treating the batch window as free latency ignores how close thirty hours sits to a full day of deferred compute. Submitting once per month regardless of the thirty-hour SLA abandons the service level entirely. Ignoring the batch window because SLAs never include processing time is operationally false—processing time is part of end-to-end delivery. Assuming every batch always returns in under one minute imports interactive expectations into a deferred API and produces late surprises when jobs run long. Exam caveat: SLA math should use the documented maximum window plus local queue and validation overhead, not optimistic medians alone. Operational check: back-calculate the latest safe submission time from clerk-needed-by minus max batch window minus reconciliation buffer, alert when enqueue slips past that cutoff, and keep an overflow interactive path for urgent packets.