Library benchmarks show PBS str.find penalty stays in scan-heavy code

Comparing python-build-standalone 3.13.5 with Debian 3.13.5 on aarch64 across popular libraries, Jinja2, Django templates, docutils, Black, Rich, SQLAlchemy, PyYAML pure loader, Markdown, BeautifulSoup and pygments stayed within 0.98-1.06x. Raw scanning did not: find of an absent 'ERROR' in a 1.8 MB log was 5.1x, counting '\r\n' 4.8x, and 'ERROR' in line over 20k lines 1.85x. requests and httpx .json() were 1.57x with one emoji but about 1.03x without; Flask was 1.40x and FastAPI 1.13x with one. To judge real-world impact of an interpreter microbenchmark, time library entry points in venvs built from each interpreter, with interleaved processes, before claiming apps are affected. These are isolated calls on synthetic inputs, not whole applications.