Second opinion.

Any completed assistant reply has a "Second opinion" action in the actions row below it. Click it and LADLE runs a fresh Anthropic call to the OTHER tier — Haiku answers get reviewed by Sonnet; Sonnet answers get reviewed by Haiku.

The verifier's critique is stored on the message and rendered as a quiet block below the original reply. It typically takes 1-3 seconds and produces 200-800 words of critique focused on errors, omissions, or overconfidence.

**When it helps most:** - High-stakes factual answers (legal, medical, financial reasoning) - Long code where a bug could hide - Confident claims that don't cite sources - Decisions where a second read of the trade-offs is worth 10 seconds

**When it doesn't help:** - Creative writing (the verifier is optimized for factual review) - Casual conversation - Cases where the original reply already opened with uncertainty (the verifier will confirm the uncertainty, not resolve it)

**Cost:** The verifier turn counts against your usage cap. Typical cost is $0.005-$0.03 per second-opinion call. Sonnet verifiers use more tokens than Haiku verifiers.

**One per message:** You can only request one second opinion per assistant message. Re-clicking overwrites the previous critique with a fresh one from the same tier.

**Not a truth-check:** The second opinion is peer-review by a model with related training data. If both models share a training-data blind spot, both may confirm the wrong answer. For questions where ground truth matters and might not be in the training data, enable SEARCH first (Anthropic's server-side web_search tool) so the reply is grounded in live sources.