Recorded runs: qwen3.5:9b vs deterministic
Recorded outputs of qwen3.5:9b (operational-v6) and the deterministic arm, from Colab run holdout-9b-v6-ecd865890442, replayed as static pages. No model runs here. Synthetic rows use ATH's generated dev scenarios. Real rows are derived from the DEDALE dataset (INRIA / IRISA, CC BY 4.0, https://dedale.inria.fr/, https://doi.org/10.57745/Y5JLDG); see ATTRIBUTION.md. The Title column says what each row shows, including the failure.
12 report(s); 12 with a complete verdict. Verdicts carry no confidence score; open a report for its evidence basis.
| Case | Title | Expected (scoring only) | Verdict | Complete | Elapsed | Report |
|---|---|---|---|---|---|---|
| sample-01 · qwen3.5:9b (D1) | [synthetic] correct malicious: Probed the process tree, found reg.exe saving the SAM hive, called it malicious. | malicious | Malicious | yes | 83.7s | open |
| sample-01 · deterministic | [synthetic] correct malicious: Deterministic triage calls the same incident malicious from the detection findings. | malicious | Malicious | yes | 0.2s | open |
| sample-02 · qwen3.5:9b (D1) | [synthetic] correct benign: Found dism.exe /ScanHealth under the remote shell and cleared it as maintenance. | benign | Benign | yes | 29.4s | open |
| sample-02 · deterministic | [synthetic] FALSE ACCUSATION: Deterministic triage cannot tell maintenance from credential export: it accuses. | benign | Malicious | yes | 0.1s | open |
| sample-03 · qwen3.5:9b (D1) | [synthetic] correct abstain: The deciding child process is deliberately missing; the model abstained and named the gap. | abstain | Abstain | yes | 24.2s | open |
| sample-03 · deterministic | [synthetic] OVER-CALL: Deterministic triage calls it malicious although the telemetry cannot decide. | abstain | Malicious | yes | 0.1s | open |
| h14 · qwen3.5:9b (D1) | [DEDALE (CC BY 4.0)] correct benign: Real Windows logs: a user opening a local PDF in Firefox, correctly cleared. | benign | Benign | yes | 14.9s | open |
| h14 · deterministic | [DEDALE (CC BY 4.0)] safe abstain: The deterministic arm abstains, as it did on all 18 holdout cases. | benign | Abstain | yes | 0.1s | open |
| h07 · qwen3.5:9b (D1) | [DEDALE (CC BY 4.0)] safe abstain: APT implant (svcmon.exe): two process-tree probes, opaque child, abstained. Safe, but no answer. | malicious | Abstain | yes | 24.8s | open |
| h07 · deterministic | [DEDALE (CC BY 4.0)] safe abstain: The deterministic arm abstains. | malicious | Abstain | yes | 0.1s | open |
| h05 · qwen3.5:9b (D1) | [DEDALE (CC BY 4.0)] UNSAFE CLEAR: The failure: an APT step cleared as benign, partly because whoami.exe is signed by Microsoft. | malicious | Benign | yes | 15.1s | open |
| h05 · deterministic | [DEDALE (CC BY 4.0)] safe abstain: The deterministic arm abstains instead of clearing. | malicious | Abstain | yes | 0.1s | open |