Task histories with context, actions and outcomes
Histories of operational work items, such as service requests, approvals, claims and back-office tasks, recording the context each person had, the actions they took in which system, and the outcome. Labs use them to train and evaluate agents on multi-step business processes with known results.
What a record contains
One work item: the request and its context, every state change and action (who, what, when, in which system), attachments, decisions with reasons, and the outcome.
| Field | What it holds |
|---|---|
item_id, process | For example refund approval or claims intake |
context{} | What the worker could see at the start |
events[] | Actor role, action, system, time and note |
decisions[] | Decision and stated reason |
outcome | Approved, rejected, reworked, completed |
{
"item_id": "W-18822",
"process": "refund_approval",
"context": {"order_value_band": "100-250", "customer_tenure_years": 3},
"events": [
{"actor": "agent_role", "action": "open_order", "system": "oms", "at": "10:02"},
{"actor": "agent_role", "action": "request_approval", "system": "workflow", "at": "10:05"},
{"actor": "lead_role", "action": "approve", "system": "workflow", "at": "10:31"}
],
"decisions": [{"decision": "approve", "reason": "carrier damage photo attached"}],
"outcome": "approved"
}How AI labs use it
- Agents on real processes
- Multi-step work across real business systems.
- Rewards from outcomes
- Approved, rejected and reworked items as signals.
- Process evaluation
- Replay histories and score an agent’s next action.
Typical preparation requirements
Agreed with the supplier before any work begins. Typical requirements include:
- Customer and employee identities pseudonymized
- Free-text notes redacted for personal data
- Regulated records such as health and credit data reviewed separately
Every dataset has a documented owner and confirmed licensing rights. See data governance on sourcex.si.
What makes a strong package
- Reasons recorded for decisions
- Outcomes linked to each item
- Consistent process definitions over time
Compared with public datasets
Public sets such as Mind2Web and TheAgentCompany are useful references, but limited as enterprise training data. The operations and workflows category page compares them with licensed data.
Who typically holds it
- Insurers
- Healthcare administration teams
- BPOs
- Logistics companies
- Finance operations teams
Know a company like this?
Introduce the company to SourceX. If its data deal closes, you can earn up to $100,000 in referral fees, paid after the buyer accepts the data and SourceX receives payment.
Refer a companyQuestions
How is this different from ticket data?
Tickets capture conversations. Task histories capture the actions and decisions taken in business systems to complete the work.
Need this data for a model?
Describe what you need: domain, volume, history, format and licensing terms. SourceX looks for companies that hold it and can license it.