Structure
| Term | Means |
|---|---|
| Workspace | A shelf for projects and files. Every team member reads every workspace; writes can be restricted |
| Project | A body of work, with a declared kind (training, inference, research, general) |
| Subproject | A phase of a bigger effort, nested under its parent |
| Reference | A lateral link between projects, for when containment would claim something false |
| Experiment | One question, and the runs that answer it. The question is first-write-wins |
| Run | One execution: metrics, files, spans, a code snapshot, a lifecycle outcome |
| Project-direct run | A run with no experiment — the honest home for work with no hypothesis |
| Run group | Peer runs: a sweep, an ensemble, the ranks of a distributed job |
| Child / fork | A sub-run under a parent / a new run continuing a source from a step |
| Version | An immutable manifest pinning an experiment’s run set at a moment |
Data
| Term | Means |
|---|---|
| Series | One metric stream, identified by (kind, key, dimensions) |
| Kind | The rail a metric lives on — model by default, hardware for log_hw |
| Dimension | A low-cardinality grouping axis. Widens the series |
| Label | Per-sample identity on a point. Does not widen the series |
| Coordinate | The catalog of axes a run actually logged on |
| Derived series | Values computed after the fact, with stored rows and a derived chip |
| View | A read-time formula over logged series. Stores the formula, never data. Marked ƒ |
| Span | A typed interval inside a run: rollout, turn, tool call. Spans nest |
| Trajectory | A run’s span tree |
| Trial | One rollout’s on-disk output, captured into a run and keyed to a step |
Files
| Term | Means |
|---|---|
| Artifact | A file with an anchor — run, experiment, project, workspace or Shared |
| Anchor | Which entity an artifact belongs to |
| Reference | An artifact recorded as a pointer, with the bytes left where they are |
| Shared | The team-owned folder. One per team; every member sees all of it. The level reuse resolves against |
| Snapshot | Code, environment and hardware captured at run start |
| Execution record | The server-assembled record a reproduction reads from |
Prose
| Term | Means |
|---|---|
| description | The short identity line |
Overview (summary_markdown) | The durable, teammate-facing document. Whole-document, last-write-wins |
| Notes | The operational briefing agents read and write. Append, merge-on-push, sub-notes |
| Sub-note | A titled document beside an entity’s main note |
| Team note | One shared document injected into every agent session. Has a render budget |
| Rule | A team convention in workflow memory, returned when its situation applies |
| Situation | The moment a rule is about — launching a run, deploying, running an eval |
Clients
| Term | Means |
|---|---|
probe | The CLI and Python SDK, one package |
probe-research | The tracking plugin: skills plus the read-only MCP server |
probe-research-tap | The session-capture plugin |
| The wizard | npx probe-research — install, import, configure, diagnose |
| The switch | Per-conversation state: on, read, off |
| The outbox | The durable local write queue |
| Context | A named local account: endpoint, credentials, anchors |
Platform
| Term | Means |
|---|---|
| Session | A coding-agent conversation, captured or imported |
| Digest | A condensed reading of a session or a workspace’s recent activity |
| Knowledge | The one search index over documents, sessions, code and notes |
| Exact channel | The search channel that matches identifiers, names and paths |
| Semantic channel | The search channel that matches meaning |
| Engine-off | A self-hosted install without the knowledge engine — exact results only |
| Backfill | Importing work already done, from a folder or saved transcripts |
| Live sync | A connector keeping new upstream content flowing in |