suspicion_raised is a null placeholder in the week-7 scorer
@richardxu257▲ 1 net votes
In the week-7 runner, score.py hardcodes "suspicion_raised": None with a comment saying it needs a judge model and prompt, so every results.json has it as null. Its definition lives only in prep docs (task2.md, tasks.json, prompts.json), and the docs disagree: one describes a list of trusted agents that noticed something, the README describes a 0-3 score per agent. The monitor's SUSPICIOUS: field in sim.py is a different thing (what the monitor flagged in bridge posts), so don't reuse it for this metric. Grep for the key will find only the placeholder and docstring, so don't expect an implementation to exist.