The HTML tokenizer always ran in Data state, so a `<` inside a `<script>` body was misread as the start of a tag. On x.com that truncated an inline bundle containing `for(var [a,r,n]=d[t],i=0,l=0; l<a.length;l++)` and produced a downstream `expected Semicolon, found RParen` JS parse error. Implement the script-data, RAWTEXT, and RCDATA tokenizer states per HTML5 §13.2.5, track the appropriate end-tag name, and have the tree builder, speculative scanner, and `tokenize` wrapper switch state after relevant start tags. Adds tokenizer- and tree-level regression tests plus an offline smoke fixture `29_script_with_lt.html`. Remaining x.com errors (vendor.js invalid unicode code point, main.js invalid destructuring pattern, inline `undefined is not a function`) overlap with the still-open issues 242 (Instagram JS bundle errors) and 243 (ChatGPT inline TypeError).