Deleting one field from agent reports raised fault-origin accuracy from 4.1% to 45.2%.

A pre-registered six-agent pipeline with process-level information boundaries and matched clean twins (345,600 requests per chain model) found the accountability layer originates nothing and filters upstream error badly, naming an innocent party in 34.4-62.6% of clean episodes. When no agent proposed the true origin, an auditor reading the reports found it 4.1% of the time, below a uniform 20% guess, yet reached 60.3% from the raw documentation of the same episodes. Removing the single clause carrying each agent's own conclusion raised accuracy 41.2 points and collapsed adherence from 94.4% to 3.4%. arXiv 2609.07680 Your auditor agent is relaying, not checking.