Overview and findings
How the Overview summarises a run, and how sections and findings are structured and labelled.
The Overview and Findings tabs hold what the agent concluded. This page explains how those conclusions are structured and how to read their statuses.
The Overview#
The Overview is the first thing you see:
- Archetype badge: the kind of report, for example Contract & security. See Report archetypes.
- Headline: one declared finding the agent treats as its answer, for example who controls a contract.
- Summary: the key findings in a short grid, each with its evidence count.
- Notices: anything that limits the result (partial result, bounded coverage, conflicting sources).
Fixed first-party agents add their own overview panels. For example, Token Researcher shows token identity and controls, and the transaction inspectors show the decoded actions and the simulation outcome.
Sections#
Findings are grouped into sections. Each agent declares its sections in advance, and some are required. A section has a status:
| Section status | Meaning |
|---|---|
| Complete | Every finding in the section was established. |
| Partial | At least one finding is unknown; each says why. |
A report is complete only when every required section is complete.
Findings#
A finding is one labelled value: a supply, an owner address, a count of transfers, a yes-or-no check. Each finding carries a status:
| Status | Meaning |
|---|---|
| Observed | Read in this run, with at least one evidence item behind it. |
| From input | Taken directly from what you entered, so it needs no evidence. |
| Unknown | The value could not be established. The reason says why (a failed read, a missing value, a type mismatch). |
| Not applicable | The question does not apply to this subject, for example the signers of a treasury that is not a multisig. |
| Unavailable | The step that would produce it failed or was skipped. |
Only observed and from input findings count towards a complete report.
Evidence links#
Every observed finding has an evidence button showing how many reads it rests on. Selecting it opens the Evidence tab filtered to exactly those reads. Findings derived from several reads link to all of them.
Signals#
Some tools also return signals: their own observations such as a risky configuration, an unusual permission or two sources that disagree. Signals are labelled by kind (risk, conflict, information) and cite the evidence they are based on. A signal is an observation to check, not a verdict.
AI-assisted summaries#
Some agents include a bounded AI analysis step. Its output is labelled as AI-assisted and is shown apart from the deterministic findings. It sees only the evidence this run captured, and every statement it makes must cite that evidence. If the analysis is unavailable, the rest of the report is unaffected. See AI analysis.