# Diagnosis

## Symptom

Attempting to retrieve every vector in an index; query requires a vector or id and top_k caps results.

## Likely cause

Pinecone is a similarity engine, not a table scan. There is intentionally no return-everything query.

## Confirm it

Confirm what you actually need: a live iteration (migration, audit) or a static copy (backup). The answer differs.

## Fix

1. Live iteration over a serverless index: `list` record ids in the namespace (paginated), then `fetch` them in batches.
2. Static copy of the whole index: create a backup (collections on pods; export paths on serverless).
3. For ongoing sync, track ingested ids in your own store at write time instead of dumping the index later.

## Verify the fix

After a list-plus-fetch pass, compare the fetched count to `describe_index_stats` totals for the namespace.