Reasoning & Review
See Lobster's reasoning, tool calls, and scientific review verdicts.
Overview
Omics-OS Cloud shows you how Lobster thinks, not just what it says. Every assistant response can include a chain-of-thought block (reasoning steps and tool calls) and an optional scientific review verdict. Both are collapsible so they stay out of the way until you need them.
Chain of Thought
A dashed card appears above the assistant's reply whenever reasoning or tool calls occurred. The header summarizes what happened (e.g. "2 reasoning, 3 tool calls"). While the agent is still working, a spinner and "in progress" label are shown.
Click the card to expand it and see:
- Reasoning blocks -- the agent's internal thinking, displayed as italic text.
- Tool calls -- each tool invocation (data loading, plotting, computation) rendered with its name and parameters.
The card collapses smoothly with an animation so you can inspect reasoning without losing your place in the conversation.
Scientific Review
After certain analytical responses, a color-coded review card appears below the message:
| Verdict | Meaning | Color |
|---|---|---|
| Analysis looks good | No issues detected | Green |
| Acceptable with minor gaps | Minor methodological notes | Amber |
| Issues found | Errors that need attention | Red |
Expand the review to see individual issues, each tagged with a severity (error, warning, or info). A footer row shows the reviewer model, confidence score, and latency.
Review verdicts are advisory. Always verify critical statistical claims and p-values against the underlying data before publishing.
Tips
- Chain-of-thought blocks only appear when the model uses extended thinking or tool calls. Simple text answers have no block.
- A wrench icon with a count badge on the collapsed header gives a quick tool-call tally without expanding.
- Reviews with zero issues still show the green "Analysis looks good" card so you know the check ran.