Review the LLM's Answer
The model proposes a solution. It compiles, it is plausible, and it is partly wrong. Classify each statement about it — correct assumption, wrong assumption, missing information, hidden trade-off, failure mode, security risk, performance implication — then compare. Generation is cheap. Verification requires understanding.
DatabaseBeginner
Make getUser faster
“Profiling shows `getUser` is called on every request and the database round trip dominates. Can you make it faster?”
API DesignIntermediate
Add retries to payments
“Payments sometimes fail with timeouts from the provider and the customer sees an error. Can you add retries so a transient blip does not lose the sale?”
DatabaseIntermediate
This report query is slow
“The monthly shipped-orders report for a region takes forty seconds. The query is below; can you make it fast?”
SecurityIntermediate
Keep users logged in
“Users complain they have to log in again every day. Can you keep them signed in for a month without a session store?”
DatabaseBeginner
The orders page makes hundreds of queries
“The orders page for one customer issues 300+ queries — one per order for the customer, one for its items, and so on. Can you fix the N+1?”
ArchitectureIntermediate
The confirmation email makes checkout slow
“Checkout p95 is 2.4 s and 1.8 s of it is the email provider. Can you get the email off the critical path?”
Agentic AIAdvanced
Let the support agent issue refunds
“Our support agent can answer questions but has to hand refunds to a human, which is the bottleneck. Can you let it issue refunds itself?”