What records can a generated approval authority dataset contain?

A generated authority dataset can describe both the rules and the cases that run through them, with links between the two so that every decision can be traced.

On the policy side, a release can include an **authority matrix** expressed as structured rows, a plain-language **delegation policy** document, and exception rules such as mandatory committee referral for certain risk classes. These can be generated in both forms, so a buyer can test whether an agent reads prose, tables or both.

On the organisation side, it can include a role hierarchy, people assigned to roles over time, reporting lines, leave and acting records, and per-person limit utilisation for the current period.

On the case side, it can include requests with amounts, categories and risk ratings, plus the full history of each: assignments, escalations, returns for information, approvals, rejections and reversals.

The most useful property is **referential integrity across these layers**. A decision should point to the rule that justified it, the person who held authority at that moment, and the delegation if one applied.

Buyers can also ask for deliberate irregularities: expired delegations still attached to open items, limits that overlap between roles, or a request edited after approval. Each should be labelled in a separate answer key rather than hidden in the data.

All content is generated. Part of reviewing a sample is checking how far the generated rules and cases depend on assumptions you would not want to carry into production.