Bread Crumbs: A Passive Compression Probe for Structural Mapping in Large Language Model Weight Files

Bread Crumbs is a low-cost, CPU-only method for probing structural regularity in large language model weight files using lossless compression. The method scans model binaries in chunks and measures compression ratios as a proxy for local statistical regularity, producing a coarse structural map without loading the model into memory or executing inference. Across multiple models (DeepSeek R1 Distill 8B, Gemma 3 4B/12B, Gemma 2 9B, GPT-OSS 20B), we observe consistent separation between tensor types under per-tensor chunking. Dense weight matrices exhibit high compressibility, while normalization layers compress significantly less. Fixed-window and Fibonacci chunking produce similar global statistics, while per-tensor alignment reveals structure that blind chunking obscures. A second-stage coordinate-sensitivity analysis evaluates whether these signals depend on traversal order. Compressibility remains stable under native and reversed traversals, changes substantially under transposition, and degrades under random permutation. These results indicate that compression-based signals reflect real local adjacency structure rather than aggregate file statistics alone. Bread Crumbs is intended as a lightweight structural screening tool that can guide more computationally expensive interpretability methods.