ProofCodec compresses deterministic policy tables (WAF rules, DNS filters, routing tables) by compressing the decision function, not the data. Every compressed artifact carries a proof that every input produces the correct output.

Results: - 2,074,698 DNS filter domains → 253 bytes (1,764x smaller than Huffman) - Not a Bloom filter — exact, zero false positives/negatives - 12-leaf decision tree + 51 residual corrections

Other benchmarks: - IP routing: 672x vs Huffman (16.7M states) - Rate-limiting: 39,262x (33.5M states) - Chess endgames: 21/21 tablebases beat Huffman (worst-case stress test)

How it works: 1. Train a decision tree predictor on the input space 2. Exhaustively verify against ground truth 3. Encode corrections via MDL-selected residual encoding (DELTA_GAPS / ENUM_RANK / BITMAP - whichever uses fewer bits)

Verifier + decoder: MIT-licensed. Encoder: proprietary. You can verify every claim without our software.

Source: https://github.com/ProofCodec/proofcodec-verify

Why does it speak about chess positions, IP addresses and add-with-carry - but not DNS domains? Is the title confusing DNS domains with domains of a mathematical function?