# Developer-led AI tool evaluation — engagement brief

Blank planning template. This is not a completed client evaluation or a promise of project availability.

## Decision
- Tool and evaluated version:
- Target developer or team:
- Decision this evaluation should support:
- Task types and completion criteria:
- Regressions or security boundaries that must be protected:

## Permissions before work
- Client authorization and relevant rights holder:
- Approved provider, environment and account settings:
- Source of each task: authorized client, internal, synthetic or licensed open source:
- Permitted prompts, outputs, code, traces and metadata:
- Material that must not be captured or sent to a provider:
- Access, reviewer permissions, retention and deletion requirements:
- Separate permission for training, if requested:

Access to a repository is not permission to train on it. Redaction does not grant rights. Do not put private records or credentials into an initial enquiry.

## Run design
- Starting commits, dependencies and reproducible environment:
- Baseline and comparable task allocation, if comparing approaches:
- Agreed intervention categories: missing context, rejected suggestion, correction, retry, review or rework:
- Timing boundary, if measured: implementation, review and rework:
- Acceptance tests, regression checks and human review rubric:
- Trial count, task difficulty, selection bias and other limits:

## Handoff
- Task-by-task outcomes and permitted evidence references:
- Failure patterns and developer intervention notes:
- Recommendations and repeat-evaluation plan:
- Reviewer and business owner sign-off:

## Optional dataset work
- Approved recipient, purpose, license and reuse limits:
- Provenance and authorization references for each record:
- Annotation and sensitive-data review:
- Duplicate and leakage checks:
- Protected held-out evaluation split:
- Final export approval and deletion schedule:

Training export is disabled until separately authorized and reviewed. Evaluate quality on protected tasks that have not been included in training.