arXiv:2611.24474v1 Announce Type: new Abstract: Forensic investigators must decide which suspicious files to examine first. Machine-learning (ML) and deep-learning (DL) malware detectors usually return a bare score that neither explains why a file deserves attention nor signals when changing data make the model unreliable. We present a framework for Windows Portable Executable (PE) malware forensic triage that works only on pre-extracted static features and never executes malware. A calibrated maliciousness score sets each file's triage priority, and a per-sample feature attribution explains the prediction. A per-sample consistency score measures whether the top attributed features agree with documented behaviors from the Malware Behavior Catalog (MBC) and MITRE Adversarial Tactics, Techniques & Common Knowledge (ATT&CK), where the available features support such interpretation. Conformal evaluation and statistical distribution tests flag drift. Each file receives one investigator-oriented triage output. The evaluation protocol, specified before testing, uses a time-ordered split of EMBER2024 as the primary dataset and BODMAS for family disjoint checks, and compares every component with simple baselines on predictive performance, prioritization effectiveness, explanation quality and robustness over time.