InspectRT v0.1.0

Added

• inspectrt_feature_memory_v1, a deterministic reduced feature-memory detector
  with complete nominal banks, exact retrieval, raw anomaly maps, and
  threshold-free metrics.
• Strict seven-file evaluation and eight-file benchmark artifacts.
• A language-neutral two-file synthetic retrieval fixture with offline
  installed validation.
• Reviewed cross-platform scientific and synchronized timing evidence.
• A static batch-1 ONNX feature-artifact workflow and experimental ONNX Runtime
  1.28.0 CPU consumer.
• Bundled baseline and fixture resources with truthful installed-distribution
  provenance.
• Documented CLI and schema support, Linux CI, and contributor, conduct, and
  security policies.

Limitations

• InspectRT is not an official or paper-compatible PatchCore reproduction.
• MVTec AD remains user-supplied; pretrained weights and generated ONNX models
  are not distributed.
• Public evaluation and benchmarking remain PyTorch-based.
• Direct Python symbols remain experimental.
• InspectRT carries no production or hard-real-time certification.