Datasets:

---

kimhammar

/

CSLE-IncidentResponse-V1

Like 2

Hi Kim,

Thank you for releasing the dataset and artifacts for your NDSS 2026 paper. I'm using the evaluation set (labels.json) to evaluate an LLM-based incident response agent, and I noticed a few things I wanted to check with you.

1. Incident / false-positive split:
   The paper reports 25 incidents and 5 false positives, but labels.json has 23 records labeled "Yes" and 7 labeled "No" (0-indexed).

As far as I can tell, the six CIC-IDS-2017 "No" records (#0–#5) are all dated 07/03, which is the benign Monday in CIC-IDS-2017, so they look like true false positives. The remaining "No" record is #21 (CSLE-IDS-2024), which matches your prompt's definition of minor alerts that don't warrant recovery. Even if #21 were counted as an incident, that gives 24/6 at most. Could you clarify how the 25/5 split in the paper was obtained, or whether a different version of the file was used?

2. Source dataset per record:
   The file doesn't include a source field. Based on IP ranges and log types, my mapping is: #0–#10 CIC-IDS-2017, #11–#14 CTU-Malware-2014, #15–#21 CSLE-IDS-2024, #22–#29 AIT-IDS-V2-2022. I'd appreciate it if you could confirm this.

3. File format:
   As released, the file isn't valid JSON: there are unescaped newlines inside strings, a few missing commas (e.g., before "Logs" in some records and between list items around record #24), and one record uses the key "system" instead of "System". Happy to share the exact locations if that's useful.

Thanks again for making this work available!