1) Fetch raw HTML with a Googlebot-like user agent.
2) Decompress response (`gzip`/`br`/`deflate`) and measure raw bytes + header bytes.
3) Extract inline scripts/styles and estimate app-level inline byte contribution.
4) Locate JSON-LD/microdata blocks and compute byte positions relative to 2MB cutoff.
5) Compute score from size, structured-data reachability, metadata presence, and inline efficiency.