Glossary¶

Words these docs use with a fixed meaning.

Analyser verb: One action the analyser offers to an assistant or a script: open, filter, graph, report and so on. Every way in uses the same verbs, within its own limits (Verbs); propose is offered only to Claude Code in the analyser. An MCP client sees each one as a tool named analyser_<verb>.
Approval: The dialog in which you allow or deny one call Claude Code in the analyser wants to make. Allow once covers exactly that call (Approvals).
Audit level: How much a node logs. A node appears in a record only if it wrote audit output at the level in force, so a node that logged nothing may still have run.
Audit log: The record a Fluxtion processor writes of each event it handled: one record per cycle, with an entry from each node that logged (Log format).
Bridge (MCP bridge): The small headless program (analyser --mcp) your MCP client launches. It forwards each tool call to the running analyser over its loopback listener (Connect your client). Claude Code in the analyser has its own private bridge, with a per-session credential.
Bundle (evidence bundle): One .fexp file carrying an investigation: the log, the graph, saved charts, reports and walks, and your notes. Verified on opening, unsigned (Evidence bundles).
Canvas: The analyser's shared view, which you and an assistant both work on: what is open, filtered, charted, flagged and lit.
Coverage: Which of the graph's nodes logged in a run, which did not, and which are excluded and why (Reading coverage results).
Endpoint file: ~/.fluxtion-analyser/rest-endpoint, readable only by you: the listener's address and per-run token, written while External MCP/REST access is on (External access).
Exchange folder: The one folder an assistant may write files into, and read external series from, once Allow assistant file exchange is on (Settings reference).
External MCP/REST access: The switch that opens the analyser's loopback listener to outside clients and scripts. Off by default (External access).
Finding: The note, and optional fix or assessment, written on a flagged record. A finding is fault or confirmation.
Flag: A mark on a record, with an optional finding. Flags belong to the session: opening another log clears them.
Focus (named focus): A saved set of topology nodes, with the reason it was saved.
Follow: Watching a local text log as another process writes it: new records are added as they complete (Open your own log ▸ Follow).
Full Claude Code: The assistant profile that runs your own Claude Code inside the analyser, with your settings, skills, hooks, MCP servers and tools, under your permission rules and the mode you choose (Claude Code in the analyser).
Glossary pointer: A project's pointer to its own domain glossary, a Markdown file in the repository whose text the analyser gives to the assistant. Not this page (Projects ▸ Vocabulary).
GraphML: The file the Fluxtion compiler writes beside a generated processor, describing its nodes and edges. The analyser reads it as the processor's graph.
In-app assistant: The analyser's own assistant: a chat with the model provider you configure, on your API key (The assistant panel).
Journey (conversation journey): A walk with a conversation shown beside it as it plays, labelled as simulated, recorded or edited.
Off: The assistant profile with no assistant in the analyser. Nothing is sent from it; outside clients are not affected.
Pairing: Whether a graph fits a log: the analyser checks, and says so, rather than assuming.
Posture: The name context uses for the work mode (context.handoff.posture: research or authoring). See Work mode.
Processor (event processor): The Fluxtion EventProcessor class that produced a log. The active one maps each node in the log to its source.
Profile (assistant profile): How the assistant works on this machine: Safe analysis, Full Claude Code, In-app assistant or Off, chosen with Assistant: … ▾ (Assistant profiles). Not the same as a project profile.
Project profile: The file <project>/.analyser/project.fluxtion-settings, holding a project's settings and context (Projects, profiles and sharing).
Prompt pack: A Markdown file of questions that replaces the start page's list for a window (Prompt packs).
Provenance: Which system a log came from, as declared by whoever opened it or matched from a project environment.
Reel: A walk recorded as one self-contained HTML page, ending on the evidence bundle behind it or on a statement that there is none (Send it as a reel).
Replay records: The recorded inputs of a run, packed into a format 2 bundle so you can feed them to your own build and compare the result with --replay-compare (Evidence bundle format).
Report (investigation report): A saved, ordered list of references to evidence, with labelled narrative, that re-renders against the log open now (Investigation reports).
Runbook pointer: A project-relative path to a runbook in the repository. The analyser stores where it is, never what it says, and never runs it.
Safe analysis: The default assistant profile: Claude Code that reads your evidence, with Read, Grep and Glob in the folder and no shell, edits or web; it changes what is open or writes a file only after you approve each call.
Saved analysis: A named sequence of analyser verbs, with its reason, kept in the project and run with Project ▸ Run analysis (Projects ▸ Repeatable analyses).
Spotlight: A temporary highlight with a caption, lit by you or an assistant to show where to look. Not saved, and not evidence (Spotlights, walks and reels).
Start page: What the analyser shows when nothing is open: What would you like to do?, Or open directly and Recent (The start page).
Walk (spotlight walk): A saved, named sequence of spotlight steps, played with ◀ ▶ on the analyser itself.
Work mode: Whether this session is for investigating or building: AI ▸ Work mode ▸ Investigate / Build, or a guess from what is open (Runbooks, glossary and posture).
Working copy: The disposable folder an evidence bundle is unpacked into, under ~/.fluxtion-analyser/bundles/. Claude Code is never started inside it: with a bundle open it works in an empty scratch folder, and the working copy's CLAUDE.md is never loaded.