Question: assumptions around ProduceResponse → ProducerBatch matching

docs/producer-response-batch-matching.md

@@ -0,0 +1,20 @@
+# Notes on ProduceResponse → ProducerBatch matching
+
+While reading `Sender.handleProduceResponse`, I wanted to write down a few questions
+before making any code changes.
+
+The response-to-batch lookup currently relies on:
+- topicId when present (newer protocol versions)
+- falling back to topic name + partition otherwise
+- a snapshot of topicId → name mapping taken before send
+
+Given that metadata can change while a produce request is in flight (topic recreation,
+leader changes, protocol differences), I wanted to confirm the guarantees here.
+
+In particular:
+- Are we assuming that a response with a non-zero topicId can always be resolved
+  using the topic-name snapshot taken before sending?
+- In what cases (if any) can `batch == null` occur, and is that considered an
+  impossible state or a deliberate hard failure?
+
+This is just a note for now — no behavior change proposed yet.