# 4now evaluation starter pack

Original synthetic material, published November 15, 2026. All organizations, people and documents in these cases are fictional.

1. Choose a case that matches the job you need done. These cases are not interchangeable across products.
2. Give the candidate tool only that case from `synthetic-inputs.json`. Keep `answer-key.json` separate.
3. Use the same task, input and configuration for each candidate. Record the exact product, plan, version and date.
4. Compare the result with the answer key. Record each missed or invented fact, time spent correcting it, and success or failure of export into the next system.
5. Apply your required pass/fail conditions before comparing scores. Repeat with a larger set reflecting your actual workload, including failures and exceptions.

The contract case checks a supplied fictional playbook, not legal validity. The invoice case uses stated test values, not tax advice. The encounter case checks factual documentation, not diagnosis or clinical safety; a qualified clinician must design and approve any clinical evaluation.

Passing a few synthetic cases does not establish accuracy, compliance, production reliability or suitability for confidential data. No vendor outputs or benchmark results are included. Use approved test data and verify vendor terms before any real deployment.

Use the interactive worksheet: https://4now.ai/guides/how-to-choose-ai-tools#evaluation-worksheet
