THREATBYTESTM research
Experiments
Experiments encode hypotheses across the four methodology phases, linking datasets, hybrid models, sandbox simulations, and red-team validation with full lifecycle documentation.
Design01
Studies progress from feature extraction and model training through integrated platform trials and comparative benchmarks against reactive IDS/IPS baselines.
What every experiment records02
- The hypothesis and the research question it tests.
- Dataset versions and the train, validation, and test split method.
- Model configuration and random seeds.
- Sandbox image digest and analysis settings.
- Metrics, outcome, and any deviation from the plan.
Planned experiment series03
- Static baselines: classical classifiers on engineered static features.
- Behaviour-augmented models: adding sandbox features and measuring the gain.
- Pre-execution prediction: predicting behaviour for held-out families using static evidence only.
- Agentic mitigation: mitigation policies evaluated in simulated networks and cyber ranges.
- Community learning: measuring how governed submissions change predictive performance.