Give the report
a truthful label.
By Alvin
Our first walkthrough passed eight of nine restart/export checks. With Ollama running, the original scripted report named the model selected in its dropdown. The application had used prepared replies. That label could make a reader mistake a demonstration for a real model result.
Fix the stored choice and the exported record
The walkthrough copy stores an empty model name for scripted runs and derives the report label from the run’s actual mode. It also labels an older scripted checkpoint correctly. The same nine checks then passed, including a real report download with installed models visible.
We corrected reference code in the walkthrough edition. The prompt above is a reader exercise for reproducing the repair; it was not a separate fresh agent-generation trial.
The original comparison stays available
The unchanged baseline remains under programs/003-research-relay. These original commands run from the extracted pack. Its scripted report has the labeling problem described above; use the walkthrough launch route for your first run.
cd programs/003-research-relay python3 relay.py # Open http://127.0.0.1:8802