{
  "title": "agent-probe-guard v0.1 \u2014 L43 pre_tool detection probe (epiphenomenal-softmax under steering)",
  "author": "caiovicentino",
  "type": "probe-result",
  "license": "apache-2.0",
  "model_id": "Qwen/Qwen3.6-27B-Instruct",
  "claim": "Detection-tier probe for tool-call success in SWE-bench traces. AUROC 0.83 at N=42 with random-feature baseline gap +0.27. Causality protocol verdict is epiphenomenal-softmax: probe DETECTS but cannot LEVER (paper-6 Phase 7 finding).",
  "numbers": {
    "auroc": 0.83,
    "n_samples": 42,
    "gap_over_random_baseline_pca10": 0.269,
    "control_token_normalized_delta_rel": -0.046,
    "behavioral_flips_alpha_5": 0,
    "sklearn_p95_inference_latency_ms": 0.19,
    "verdict_class": "epiphenomenal-softmax"
  },
  "artifacts": [
    "probe.joblib",
    "manifest.json",
    "README.md"
  ],
  "methodology_check": {
    "verdict": "epiphenomenal-softmax",
    "real_auroc": 0.83,
    "random_baseline_mean": 0.561,
    "auroc_gap": 0.269,
    "delta_rel_max": -0.046,
    "flip_rate_at_max_alpha": 0.0,
    "baselines_run": [
      "random_direction_random_acts"
    ],
    "caveat": "v0 protocol \u2014 predates the three-baseline upgrade in openinterp-mcp v0.0.3"
  },
  "reproduces": null,
  "schema_version": 1,
  "created_at": "2026-05-10T20:00:40Z",
  "manifest_sha256": "9f2e9c5b8e4fbb7c7eb4c9290d927e76341006c9d63277da05f4d21a0ab26c9b",
  "hf_repo_id": "caiovicentino1/agent-probe-guard-qwen36-27b",
  "hf_url": "https://huggingface.co/datasets/caiovicentino1/agent-probe-guard-qwen36-27b",
  "doi": null,
  "paper_url": "https://openinterp.org/research/papers/two-forms-epiphenomenal-probes"
}