Context:
We're building an application where users create a case and upload documents. Backend is hosted on Railway, frontend on Vercel, and Supabase Cloud is our DB.

Summary of Issue:
In production, we intermittently face a visibility issue where newly inserted case records are not immediately retrievable post-transaction. This happens ~1 in 5 times.

Key Observations:

1. No Transaction Errors:
   All case creation flows complete without error. We’ve tested various isolation levels (including SERIALIZABLE) and ensured proper transaction wrapping across related inserts (e.g., case + structured analysis).

2. Delayed Read Visibility:
   Reads immediately after the insert sometimes return no data. The same record appears on the second or third retry — indicating potential read replica lag.

3. Local vs Production Behavior:

• Local (Direct Connection): No lag; reads are consistent.

• Production (Pooled Connection): Read issues appear intermittently — likely due to Supabase’s connection pooler routing reads to replicas.

1. Structured Logging & Audit:
   Logs confirm the insert occurred, but follow-up reads failed. Even with forced retry + backoff, the issue persists.

2. Merged Transaction Experiment:
   Merging case, structured_analysis, and steps into a single transaction avoided the error — but sometimes the inserted data itself was missing, and post-insert logs weren't generated, suggesting visibility issues post-commit.

Ask:
Do you also think that the issue is because we are using three different cloud setups a network lag ? as it happens 1 or 2 times in 5 to 10 transactions.
We suspect the read-after-write lag due to replica reads in Supabase’s pooled setup. Would appreciate any official insights, recommended config changes, or ways to ensure read-after-write consistency in this environment.