Vocabulary
StudioX provenance terms
When you download provenance from a mission conversation, you get a W3C PROV-JSON record. It uses the standard prov: terms, and a few StudioX terms with the prefix sx:, which stands for https://www.studiox-ai.com/ns/prov#. This page says what each one means.
Values are copied into the record up to 4,000 characters. Every copied value also carries sx:sha256 of its full text, so you can check a value you hold elsewhere against the record even when it was cut. See W3C PROV on the standards page.
Types
Values of prov:type: what kind of thing each activity or entity is.
sx:AgentRun- One turn of a mission conversation: a question and the work the mission did to answer it. A prov:Activity.
sx:ReasoningStep- One step inside a turn: a plan, an agent's work or a tool call. A prov:Activity informed by the step before it.
sx:HumanDecision- An approval or choice the mission asked a person to make. A prov:Activity associated with the mission (requested) and the person (decided).
sx:Question- The message that started a turn. A prov:Entity the turn used.
sx:ToolArguments- What a step sent to a tool. A prov:Entity the step used.
sx:ToolResult- What the tool sent back. A prov:Entity the step generated.
sx:Answer- The mission's reply at the end of a turn. Generated by the turn, derived from the question and attributed to the mission.
sx:DecisionOutcome- What the person chose, or the decision's final status. Generated by the HumanDecision.
sx:ProvenanceRecord- The exported record itself: when it was made, and of which mission and conversation.
Attributes
Attributes StudioX adds beside the PROV ones.
sx:sha256- SHA-256 (hex) of the full text of the value, so a copy held elsewhere can be checked against the record even when the value was cut.
sx:truncated- true when prov:value holds only the first 4,000 characters of the text.
sx:fullLength- The number of characters in the full text, when the value was cut.
sx:missionId- The mission's id in StudioX Missions.
sx:conversationId- The conversation the run belongs to.
sx:turnId- On a record of a single turn: the id of the question that started it.
sx:generatedAt- When the record was produced (ISO 8601, UTC).
sx:partOf- On a ReasoningStep: the AgentRun it belongs to.
sx:eventType- On a ReasoningStep: the kind of step the mission recorded (for example a plan, an agent call or a tool call).
sx:durationMs- On a ReasoningStep: how long it took, in milliseconds.
sx:tool- On a ReasoningStep: the tool that was called.
sx:server- On a ReasoningStep: the integration (MCP server) the tool belongs to.
sx:isError- On a ReasoningStep: true when the tool call failed.
sx:event- On a HumanDecision: the mission event that raised it.
sx:status- On a HumanDecision: where it stands, for example pending, approved or rejected.
sx:role- On wasAssociatedWith for a HumanDecision: requested (the mission) or decided (the person).
sx:notes- On a DecisionOutcome: the note the person added when deciding.
sx:email- On a prov:Person: the person's email address.
Identifiers
Every node in the record is named sx: followed by one of these and an id.
sx:mission_…- The mission, a prov:SoftwareAgent.
sx:agent_…- One of the mission's named agents, a prov:SoftwareAgent acting on behalf of the mission.
sx:person_…- A person who asked a question or made a decision, a prov:Person.
sx:sensor- A run started on a schedule or by a watcher, with no person asking: a prov:SoftwareAgent labelled "Scheduled or sensor-started run".
sx:turn_…- An AgentRun.
sx:step_…- A ReasoningStep.
sx:prompt_…- A Question.
sx:args_…- ToolArguments.
sx:result_…- A ToolResult.
sx:answer_…- An Answer.
sx:decision_…- A HumanDecision.
sx:outcome_…- A DecisionOutcome.
sx:record_…- The ProvenanceRecord.