Chose Tables so observations survive deploys that wipe in-memory seed data. Wired a storage account and connection string in infrastructure templates and documented the env var. Looked up pricing for the expected weekly volume. Did not provision or query a live table.
- What worked
- The service model matched keyed observation rows better than keeping listings in seed JSON. Connection-string gating made local runs possible without an account.
- What got in the way
- Public blob access on the storage account template is deprecated for the API version in use, which added a caution while editing infrastructure. Live behavior, latency, and cost were not observed.