Skip to content
MirrorFirm

What should I specify when requesting a generated booking dataset?

A checklist of the choices a buyer can make when scoping generated booking material: product type, system landscape, exception mix, formats and answer-key detail.

Decide these points before asking for a sample, so the generated material matches your agent's job.

- **Product and domain:** business term loans, equipment finance, vendor finance or premium funding, each with different rules. - **System landscape:** which systems your agent will touch, and how many separate records it must reconcile. - **Condition forms:** structured fields only, or also memos, amendments and email threads. - **Exception mix:** which categories to include and what share of cases should be blocked rather than bookable. - **Tool surface:** the action names and payloads your agent framework expects. - **Answer-key depth:** end-state records only, or also acceptable action sequences and required handoff facts. - **Format and layout:** JSONL tasks, Parquet tables, or an environment image with a reset command. - **Difficulty controls:** document noise, date gaps and conflicting sources.

Treat the first sample as a way to refine this list. Look at a few cases end to end, then tighten the specification.

Specifying these choices up front keeps a sample focused on the behaviours your agent must learn or be tested on.

Audience:
Agent developers
Content type:
Checklist
Buyer question:
What can be generated

Content release: r-a220eb6e099b243b

More on Loan booking