# AI evaluation project brief

An unfilled preparation template. Do not include credentials, private documents or customer records in an initial enquiry.

## Workflow
- Who uses the system?
- What task should it complete?
- Which actions or downstream systems are involved?

## Decision
- What do you need to decide after the evaluation?
- Who owns that decision?

## Observed failures
- Describe one representative failure and its consequence.
- Is the example redacted or fictional?

## Existing assets
- Documents, reference answers, traces, datasets or test suite:
- Current model/system version, if known:
- Available test environment:

## Constraints
- Languages and task types:
- Access and provider restrictions:
- Confidentiality, retention and deletion requirements:
- Domain reviewer availability:
- Target date and timezone:

## Expected handoff
- Report, comparison, reproducible cases or regression baseline:
- Questions the findings must answer:
- Ownership and support arrangements to confirm:

Scope, deliverables, cost, timing and controls require agreement before work begins.

## Acceptance criteria to discuss
| Workflow / task | Useful outcome | Unacceptable failure | Reference owner | How to check |
| --- | --- | --- | --- | --- |

Use observable outcomes, such as a correctly cited current policy or a completed action with approval. Avoid assigning universal quality targets before reviewing task complexity and consequences.

## Baseline and comparison
- Current system and candidate versions:
- What changes between them:
- Existing quality measurements and how they were calculated:
- Response-time and cost constraints, including measurement boundaries:
- Known gaps in the current test set:

## Example record (redacted or fictional)
- Input and starting state:
- Expected answer or final state:
- Actual answer or final state:
- Supporting reference and its version:
- Consequence and frequency, if known:
- Reproduction steps:

## First conversation checklist
Bring the workflow, decision, representative failure, asset inventory and constraints. Agree what can be shared and through which channel before transferring documents. Confirm deliverables, exclusions, reviewer responsibilities and the process for scope changes in the proposal.