Removing an LLM verifier from an offensive-security agent moves median findings from 0 to 2 per run.

A 15-run pilot, a pre-registered 20-run confirmatory ablation and a pre-registered 2x2 factorial with 40 runs across two vulnerable lab systems (arXiv 2609.15887). Removing verification raised reported findings (median 2 against 0, p = 0.00003) and cut precision (0.353 against 0.471, p = 0.0087), with the model verifier rather than the deterministic rules doing the suppression (p = 0.004) and recall unchanged (p = 0.158). The full design retained 93.8% of model-adjudicated true candidates but missed its pre-registered non-inferiority bound, and human verification is still pending. Pre-registered ablations in agent research are rare enough to call out.