Understand the terms behind the results.

The source state and relevant configuration represented by an analysis. This does not imply immutable graph storage; inspect run and freshness information when reusing results.

An addressable code element, such as a function, method, or class; not necessarily a whole file.

A link between code components resolved by the supported analyzer. Inspect its type, resolution method, and confidence.

Components and paths the analyzed graph suggests may be affected by a change. Coverage and unknowns limit the conclusion.

How much of a defined relationship set the analyzer resolved. Read the run's denominator and exclusions; this is not an accuracy or safety score.

A retained placeholder where a target was not resolved. It marks an unknown, not automatically a defect.

An item identified by static detection for review, such as a pattern match or a structural anomaly. It is not an AI-generated discovery or automatically a confirmed defect.

A forward slice follows what a component depends on. A backward slice traces what depends on it. Queries may be bounded by depth or result limits; neither guarantees runtime behavior.

A group of related components inferred from graph relationships. A display grouped by folders is a different view and should be labeled as such.

An interface through which a compatible coding agent queries stored relationships, findings and source context. Setsumei’s local query tools are read-only; the agent’s other tools, model and data paths remain independently controlled.

The source references, relationship context, detector information, and recorded reasoning available to support or challenge a conclusion. Check its scope and unresolved information.

An item flagged for engineering review. Inspect the candidate evidence and any investigation outcome before deciding whether it needs action; a reported item alone does not establish a defect.

The person accountable for choosing the next action and ensuring that the agreed review or verification work is assigned.

The recorded confirmation, dismissal or unresolved result of optional model-assisted investigation into a pre-identified candidate. Review its evidence and investigation run before deciding the action.

Structural signals that help prioritize reading and investigation. Consequence measures how much a component sets in motion; foundation measures how much other code rests on it. These are not measures of business criticality or runtime availability.

A component through which many analyzed graph paths pass. This identifies structural concentration, not a proven runtime performance bottleneck.

An analysis that updates changed files and affected dependents. Dependency completeness is best effort; periodic full analysis can be needed. It is not automatic live updating.

Information about the analyzed run and whether relevant source files have changed. Check findings provenance and in-progress-analysis warnings as well as file freshness before reusing an answer.

The browser interface opened from the scan’s generated HTML. It presents the analysis for human review. MCP is a separate query path into stored analysis; there is no manual report-import step.

A reviewable presentation of analyzed relationships, candidates, source context, coverage and any available investigation outcomes. The GUI brief and architecture document are distinct ways to use analysis; a report is not a shared team decision history.

Analysis of source and code relationships without executing the application. Setsumei’s deterministic core analysis builds the graph and candidates without requiring an LLM or model API key. Optional AI-assisted steps, such as investigation, call a language model.

A separate step that uses relevant source and graph context to examine candidates static analysis already identified. It uses the provider configured for that workflow; it is not required to obtain the static analysis.