Public data, code and probe weights support isolated replication, but cross-model transfer, false positives and white-box access cost remain unresolved.
Next stepWithin one day, replay benign and deceptive traces and record false positives, memory and cross-model failures.