ENDING A RUN

ENDING A RUN IN THE PRACTICE LAB The Resume Review Agent ends each run by writing a short report file and a line in its log. When a resume cannot be read, the agent should not invent a score. Instead it should write a report that says the resume could not be read, and why, and it should leave the batch free to continue with the next resume. This is the same principle you have just studied: an honest failure report is better than a confident wrong answer. Check the lab's output folder after a run, and confirm that every resume has either a real report or an honest explanation. CHECKING THAT THE ENDING IS REAL A run has truly ended well only when its outputs exist and are valid. Write small checks that confirm the file was created, that it is not empty, and that it contains the fields a reader expects. These checks are cheap, they run in the harness and not in the model, and they catch a surprising number of problems. If a check fails, the run should be marked as failed, no matter how pleased the model sounds in its last message.