Where do sufficiency records fit in a customer's AI training pipeline?
Sufficiency data maps to five linked record types that slot into supervised fine-tuning, preference training and agent evaluation, and resellers should describe those hooks to customers.
A sufficiency dataset is easiest to describe as five linked record types, delivered as JSONL or Parquet tables.
- `case`: `case_id`, `case_type`, `opened_at`, `status` - `document`: `doc_id`, `case_id`, `doc_type`, `issued_on`, `content_ref`, `quality_flags` - `policy_rule`: `rule_id`, `case_type`, `required_doc_types`, `recency_days`, `version` - `assessment`: `assessment_id`, `case_id`, `rule_ids`, `doc_verdicts`, `overall`, `rationale` - `request`: `request_id`, `case_id`, `items`, `message`, `due_at`, `lapse_outcome`
The joins are on `case_id` and `rule_id`. Keep them stable across releases so your customers can merge updates.
**Where customers use it:**
- **Supervised fine-tuning:** input is the case, documents and policy; target is the assessment and request. - **Preference training:** pair a specific request with a vague one for the same case. - **Agent evaluation:** expose the policy and documents through tool calls and score the final state.
If you pass the data on, include the schema file and a version string in every delivery. That lets your customers pin a version when they report results.
The data is generated. Confirm in a sample which fields are populated and which are optional.
Giving customers the record types and their joins up front shortens their evaluation and reduces support questions back to you.
- Audience:
- Licensed resellers
- Content type:
- Integration
- Buyer question:
- What it is and why it matters
Content release: r-a220eb6e099b243b
Related questions
- What does it mean for an agent to decide whether the supplied documents are enough?
- What does a sufficiency decision look like on a real loan application?
- How do I tell whether my process suits sufficiency-focused training or evaluation?
- Where does the sufficiency decision sit in an agent loop, and which tool actions does it need?
- Why not handle this with a fixed checklist or rules engine instead of an agent?
- How does a sufficiency decision map onto environment state and tool actions?