A team chooses a model with a large context window and plans to send its entire document archive for every question. Which concern should the architect raise?

1. Context limits, token cost, latency, and evidence relevance still require a deliberate retrieval and context budget.
2. Retrieval becomes an authorization boundary automatically without checking document permissions.
3. Input tokens are never billed when the model supports a large window.
4. A large context window guarantees perfect recall of every fact placed anywhere in it.