Scrapling HTTP on the same 30 public sites

Dates, network and method

Identical validator code and frozen thresholds: required text groups, minimum checked text length, target-specific listing patterns, soft-block and gate phrases. Head, script, style, template and noscript are excluded; CSS visibility is not resolved. Usable means HTTP 200, required content and no detected soft block; Scrapling requires a complete body capture and no request/process error; browsers also require complete DOM capture and no navigation/capture/process error. A strict failure does not prove every field is unusable.

Browser capture latency includes navigation, a two-second settle and DOM serialization. Total Camoufox and Patchright attempt time also includes startup, a viewport screenshot and teardown. Scrapling time covers HTTP fetching and selector construction; wreq time covers HTTP body transfer. These different workloads do not form a speed ranking.

PSS is sampled every 100 ms across the isolated Python attempt and descendants, including the driver and browser. Startup/teardown are included; Camoufox and Patchright also include screenshot work. Brief peaks and changing process trees can be missed; these are sampled totals, not exact peaks or pure browser memory. Scrapling PSS includes Python imports, HTTP client and selector parsing, with no browser or screenshot; it is a different workload. wreq PSS was not measured.

Scrapling RAM preflight: 13459 MiB available of 32050 MiB, with 11 MiB swap free. Full HTML, screenshots and logs stay private in ignored runner/data. Published hashes cover serialized UTF-8 DOM for browsers and HTTP body bytes for Scrapling and wreq.

All 30 targets, five dated attempts each · 150 dated records

Manifest order. Open a target to inspect all five tools; required fields and validation rules stay the same.

Reproduce the single HTTP mode

Use the requirements file to pin all installed dependencies, including Scrapling 0.4.15 and curl_cffi 0.16.3. No browser installation or launch is used by this slice. The script refuses an existing output directory; a rerun sends 30 new GET attempts with ordinary redirects.

uv run --with-requirements public/evidence/scrapling-public-30-20260929/requirements.txt python \
  public/evidence/scrapling-public-30-20260929/run.py \
  --manifest runner/corpora/public-30-v1.json \
  --output runner/data/scrapling-public-30-new-run

Executed command

/tmp/scrapingevals-scrapling-public-30-20260929/bin/python public/evidence/scrapling-public-30-20260929/run.py --manifest runner/corpora/public-30-v1.json --output runner/data/scrapling-public-30-20260929

Code SHA-256

c68cf304a92e4519eb7174ba9dc86dd1597b7bce5b75422a961833f95cc81215

Frozen manifest SHA-256

6ce519c4f7df3c32f5dfda379a8d3b280b7723bdeb48ddf64a1ae10ce6a7914c

Resolved the official Scrapling repository, asked one focused DeepWiki question, and checked official HTTP-mode documentation and v0.4.15 HTTP engine source, which matches the installed file byte for byte. Exact package versions and source hashes are in metadata. Implicit curl browser headers are not exposed by Scrapling’s explicit request_headers; no complete wire-header or network-resource trace is claimed. This pass does not test Scrapling’s browser fetchers, crawling or adaptive extraction.