Recon find (was internal #907). parser.go does an upfront findHeaderEnd whole-block SIMD completeness scan, but parseHeaders already detects incompleteness line-by-line — so header bytes are CRLF-scanned twice on every fully-arrived request (the entire bench workload). Removing the upfront scan: ~-9% parse CPU on get-json, ~-13% on auth-chain (Mac micro). Lands as rps on CPU-bound chain-api/security/fullstack; ~0 on NIC-bound get-json. consumed==0 contract must be preserved; full h1 suite + fuzz + smuggling must pass. Validate via interleaved cluster A/B on chain cells.
Recon find (was internal #907). parser.go does an upfront findHeaderEnd whole-block SIMD completeness scan, but parseHeaders already detects incompleteness line-by-line — so header bytes are CRLF-scanned twice on every fully-arrived request (the entire bench workload). Removing the upfront scan: ~-9% parse CPU on get-json, ~-13% on auth-chain (Mac micro). Lands as rps on CPU-bound chain-api/security/fullstack; ~0 on NIC-bound get-json. consumed==0 contract must be preserved; full h1 suite + fuzz + smuggling must pass. Validate via interleaved cluster A/B on chain cells.