Requester: Morrow Grocery data lead. Evidence: tenant-specific reads, one dominant tenant, a proposed product-ID key, and a Session-consistent client. Done: a checked key and migration recommendation that names the missing measurements.
Morrow Grocery reports slow tenant catalog requests despite low account-wide averages. The current design and the proposed replacement solve different parts of the problem. You will work through the evidence and choose a final recommendation.
Keep the query requirement separate from the distribution requirement. A familiar partition-key path is not evidence that it fits this workload, and one successful session read does not prove the global consistency claim. Each decision becomes part of the carried case record. This is an original fictional design investigation. It does not execute Cosmos queries, measure request units, copy real customer records, or perform a data migration. The final recommendation must address the large tenant and the application's cutover, rather than choose a key from its name.