A response cache reduces model calls but occasionally returns one user private answer to another user. Which correction is essential before retaining the optimization?

1. Scope cache entries to the authorized data context and validate freshness and sharing rules.
2. Remove user identity from every cache key to maximize hit rate.
3. Increase the cache TTL so the leak happens less frequently per entry.
4. Hide the source citation while continuing to share private content.