What is a case library?

A case library is a curated collection of real cases, stripped of personal details, each with the decision that was made and the reason for it. Teams use it to train reviewers, to run calibration and to test changes to an agent. It is the team's memory of hard decisions, kept in a form that new people can learn from.

Why build one?

Because the best training material for reviewing an agent is the work the agent actually produces. Invented examples are too tidy: the document is always clearly fake or clearly real, the amount always clearly too high. Real held actions are messier, and learning to decide them is what reviewers need.

A case library also preserves judgement. When an experienced reviewer settles a hard case, the reasoning usually disappears into one filed decision. Captured in a library, the case and its reasoning become teaching material for everyone who follows. Over time the library becomes the most direct record of how the team interprets its own rules.

What goes in a case library?

Each case should include:

• The facts, anonymised, with the provenance of each: sealed, signed, declared or external.
• What the agent proposed and its verdict, held or refused, with the bound that decided it.
• The decision a qualified person made.
• The reason, written so a new reviewer can follow it.
• Tags: case type, difficulty, which rule or red line was involved.
• Date added and last checked, so cases decided under old rules can be found and updated.

Cases come mostly from the review queue: held actions that were hard, that reviewers disagreed on, or that revealed a gap. Edge cases are especially valuable; what is an edge case explains why.

An example

Take a hypothetical bank's fraud operations team. Over six months, the team lead adds forty cases to the library: payments held because the payee was new, account changes held because the request came from an unusual device, refunds held because the customer had an open dispute. Each has its decision and reason. New analysts work through the library in their first fortnight before touching the live queue. When the agent owner changes the agent's brief, the same cases are replayed to check the agent still holds what it should.

How is it different from a policy library?

A policy library holds the rules: policies, procedures, product terms. A case library holds the decisions made under those rules. The policy says refunds above the customer's limit need approval; the case library shows twenty refunds above limit and how each was decided. Agents read the policy library as a source, a connection covered in how to connect an AI agent to your policy library. People learn judgement from the case library.

A case library is also different from a reasons library, which collects standard wording for refusals and declines. The support agent refusal reasons library template shows that format. A reasons library standardises how decisions are communicated; a case library teaches how they are made.

Who maintains it?

A team lead or senior reviewer, with subject matter experts confirming the decisions on hard cases. Review it whenever policy changes, and remove or update cases decided under rules that no longer apply. The guide to training teams for AI agents covers how it fits into a training programme.

Where the Academy fits

Pryme Intelligence Academy's module tests are set in the learner's own industry and role, which serves a similar purpose: practice on cases that look like the work. A team's own case library adds what no external course can, which is its own decisions under its own house rules.