OpenAI final evaluation — Verified Real-World Workflow
Lesson 2: Final Evaluation Execution — Evidence, Iteration, and Troubleshooting
Lesson Objectives
By the end of this lesson, students should be able to:
- Maintain a work log with prompts, actions, revisions, and evidence
- Troubleshoot generic, incorrect, blocked, or wasteful output
- Verify important claims and calculations
- Document where human judgment changed the result
Lesson Content
Execute in small stages. Save the original brief, important prompts, source list, drafts, diffs or versions, test results, and final output. Record failures instead of hiding them: what went wrong, likely cause, recovery step, and what changed.
Required troubleshooting evidence must include at least two of the following: restart a contaminated thread; improve weak context or constraints; replace an unreliable source; recover from a tool or upload failure; correct a calculation; reject an unsafe action; reduce unnecessary tool use; or move a task from ChatGPT to Codex or from Codex back to human control.
Verification should match the project. Research requires opened sources. Data work requires independently checked calculations. Software requires tests and diff review. Connected workflows require permission and action review. Creator work requires factual, rights, consistency, and accessibility checks.
Enroll to continue this lesson.
The preview above shows the lesson objectives and opening lesson content. Enroll to view the full lesson, complete the practice work, and take the lesson quiz.