How to use this viewer
This is a fully local static viewer generated from the uploaded Eval Reports.zip. It does not call any external APIs and it does not require a server.
Prompt flow: choose a GEPA session, then use Final lineage, Promoted only, or All candidates. Every card shows the full prompt text inline so you can compare prompt-to-prompt changes side by side while horizontally scrolling the flow chart. Solid arrows are parent-child optimization edges; dashed arrows mean the saved optimized prompt has the same SHA as a candidate prompt.
Sample variation matrix: rows are HumanEval tasks and columns are the full 5x eval conditions. Cells show pass count over evaluated attempts. Click a row to inspect all repeats, generated code, prompts/messages, prompt fingerprints, and failure summaries.
This repo copy commits the static viewer, browser-loadable data bundle, and CSV exports. The one-off preprocessing script from the original Desktop bundle is intentionally not included here.