Datasets:
| # ThreatIntel fixture protocol | |
| - Seed: 5 | |
| - Generator: `src/generator.py` | |
| - Hosts use documentation IPv4 ranges and the reserved `.example` TLD | |
| - Indicators are never probed | |
| - Split key: `campaign_id` (same campaign never crosses development/test) | |
| - Claim kinds: observed / quote / author_recommendation | |
| - Brand mention is not victimhood | |
| - ATT&CK IDs are looked up in `src/data/attack_snapshot.json` | |