Start with a common time axis
Record machine states, inventory and accepted output over the same interval. Mark cleaning, planned stops, raw-material changes and operator actions. If observations use different periods, an apparent cause-and-effect relationship may be a coincidence.
Keep competing explanations open
A full buffer may reflect low downstream demand, a failed transfer, a route being unavailable or deliberately held product. A starved machine may be waiting for a batch, utilities or a permitted restart. Locate the event sequence before selecting a change.
Close the loop with a testable proposal
State what evidence supports the candidate change and what would disprove it. Define product conditions, sustained observation period and disturbance tests. The upgrade workbench keeps the evidence request beside each suggestion; it does not calculate a promised efficiency gain.
