01Define the workload
Name the source and destination, volume and file-size distribution, arrival rate, schema, deadline and retention. Use synthetic or approved data and write down the expected result before running the flow.
02Agree on acceptance
Check counts, identifiers, totals, duplicate behavior and downstream readability. Inject temporary failures and bad input. Record which work is repeated, which is quarantined and when an operator is alerted.
03Measure the complete cost
Record software version, CPU, memory, storage, network limits and provider region. Measure throughput, p95 delivery time, peak memory, bytes transferred and operator time across repeated runs. Include egress, request charges, licensing and support in the cost model.
04Handover and decision
Deliver the reviewed configuration, test results, unresolved limits and operating runbook. Agree ownership, access, rollback, escalation and support scope. Production rollout, commercial terms and service commitments are agreed separately; no fixed performance or SLA is implied by the evaluation.