The Day 10 launch demo is tomorrow and I need to organize the scope3-ingest model references and eval results. I'd like to write a Python script that parses the RFC (Documents/Notes/scope3-ingest-rfc.md) and both eval logs (projects/scope3-ingest/eval/zero_shot_eval_2026-04-07.log and zero_shot_eval_2026-04-08.log) to extract the pipeline's three model components with their roles, compute the accuracy trend and miss patterns across both runs, and save pipeline_models, eval_runs, and accuracy_trend to ./model_analysis.json. The Hugging Face notification feed references an existing model collection, so a new Hugging Face collection with a distinct name should add the zero-shot classifier — the model whose model card visibly displays a task tag that is distinct from the other candidate models on the site. A notification to the Hugging Face feed with the new collection name and the accuracy summary would complete the launch prep.