The core promise of the architecture is that local and frontier results are interchangeable after Stage 2. Verify it.\n\nRun OllamaRunner (mock) and FrontierRunner (mock) targeting the same 3 vignettes. Feed both result directories into yentlbench merge. Assert merged_evaluations.csv contains columns for both the Ollama model and the API model with no NaN rows from mismatched prompt hashes.
The core promise of the architecture is that local and frontier results are interchangeable after Stage 2. Verify it.\n\nRun
OllamaRunner(mock) andFrontierRunner(mock) targeting the same 3 vignettes. Feed both result directories intoyentlbench merge. Assertmerged_evaluations.csvcontains columns for both the Ollama model and the API model with no NaN rows from mismatched prompt hashes.