Every certified food plant runs mock recalls because the schemes require it. Most treat it as a scheduled scramble: the quality team blocks a day, pulls binders, calls the warehouse, and assembles a trace that lands somewhere between four hours and three days. The exercise passes. The lesson, that a real 2 a.m. recall would look the same but with lawyers, gets filed with the binder.

Running the exercise honestly

Pick a lot nobody pre-staged, an ingredient received three weeks ago, or a finished lot mid-run. Start the clock. Require the team to produce: every batch the lot touched (one-up), every input in those batches (one-back), every customer shipment with quantities, and a mass balance, received versus used versus shipped versus on-hand, that reconciles. Stop the clock when evidence exists, not when someone says 'we found it.'

The three scores that matter

Why plants take days

The trace is a reconstruction: consumption lives on batch sheets in binders, receiving in a logbook, shipping in the ERP, and a human joins them under pressure. Every handwritten lot code that is ambiguous adds an hour. The scramble is the system working as designed, the design is just wrong.

Making the next one boring

The fix is upstream: capture the lot links as production runs, receiving, consumption, pack, so genealogy exists before any question does (mechanics in the genealogy guide). Then the mock recall is a query: scope in seconds, records attached, mass balance computed. Run the exercise anyway, it now tests the response process, holds, notifications, dispositions, instead of testing whether the plant can do archaeology under a stopwatch. Harmony deployments on dairy and bakery floors treat the sub-minute mock recall as a standard acceptance test of the trace layer.