DriftWatch¶
DriftWatch is an open-source study of how long a malware detector stays useful as malware changes. It applies one time-aware evaluation protocol (TESSERACT constraints C1 to C3, reported with AUT and A-AUT) to pre-extracted feature files from LAMDA (Android) and BODMAS (Windows PE), and asks how label delay, a fixed analyst budget and conformal drift monitoring change a detector’s lifetime.
It never downloads, runs or redistributes malware. It describes how detectors decay, not how to evade them (ethics).
Start here
The study
- DriftWatch protocol: single source of truth
- DriftWatch pre-registration
- DriftWatch claims ledger
- Running the E2 grid: cells, chunks, the ledger, Kaggle and Colab
- Malware Detector Lifetime Under Delayed Labels: Small Effects and Null Monitors in a Pre-Registered Time-Aware Study on Android and Windows PE Data
- Chapter 1: Introduction (draft)
- Chapter 2: Landscape (draft)
- Chapter 3: How the results are analysed (draft)
- Chapter 4: Recent techniques and attacks (draft)
- Chapter 5: Related work (draft)
- Chapter 6: Method (draft)
- Chapter 7: Experiments and results (draft)
- Chapter 8: Discussion and limitations (partial draft)
- Chapter 9: Conclusion and future work (draft)
- Related-work addendum (Phase 7)
Architecture
Decisions and evidence
- ADR-0001: Tooling and guardrail design
- ADR-0002: Code licence
- ADR-0003: PyPI distribution name
- ADR-0004: Slot definition, minimum-size gate and gap handling
- ADR-0005: Window parameters (nT, nE, stride), size gate and ratio policy
- ADR-0006: Metrics, decision threshold, scaling policy, A-AUT windows and E0 hyperparameters
- ADR-0007: Training-pool composition, new-label weight, budget base, retrain anchor and the simulator’s queue rules
- ADR-0008: RQ1 primary comparator and the E2 grid design
- ADR-0009: Calibration rows are not training rows; the E3 initial model leaves a hold-out
- ADR-0010: Monitor inputs, reference rule, label-free p-values and bounded monitor compute
- ADR-0011: Documentation site tool
- Guardrail review (Phase 1)
- Leakage review of the E3 code (conformal layer, selectors, E3 runner)
- Leakage review: Phase 9 (E4 drift monitors)
- Leakage review: Phase 3 time slicer and constraints
- Statistics review of the E2 analysis code (Phase 6)
- Statistics review of the E3 tables and text
- Statistics review: E4 (RQ4, drift monitors)
- DriftWatch phase ledger
- Results audit
- Gate G1 evidence (Phase 4)
- Gate G2 evidence (Phase 5)
- Gate G3 evidence (Phases 6 and 7)
- Gate G4 evidence (Phases 8 and 9: Tier 2)
- Tier 3 go/no-go decision (Phase 10)
- Claims audit
- Simulated hostile review: “How Long Does a Malware Detector Last? Label Delay, Analyst Budget and Conformal Monitoring in One Time-Aware Protocol”
- Responses to the simulated review
Data, licences and ethics
Operating the project