[jehrr]

Fixed

• User-visible scraper_api_client.py --help: the description said these pages need JavaScript and a single fetch returns "about 5 products where a browser engine returns 60"; --url said to pass --wait-element '[data-stid="lodging-card-responsive"]' because "the grid arrives over client-side GraphQL" (VRBO/Tokopedia). This file's own docstring measures 48 of 48 vendors from one request on a server-rendered grid. Both rewritten; --category help no longer claims a URL-segment default (the client never reads args.category).
• Issue templates were an auction site's (lots, reserve prices, seller_score, page-100 cap, /en/ locales, __NEXT_DATA__, c-lot-card, --out grills, "96 products"). Rewritten from README / CONTRIBUTING / TROUBLESHOOTING.
• pyppeteer/Selenium docstrings listed --mode product (brand, EAN) and used foodpanda.pk/p/makanan-minuman/minuman/kopi-bubuk (a Tokopedia path) as the example. Now the one listing mode and /city/lahore/area/gulberg.
• Dead code: the twins' if args.mode == "product" sidecar branch (shop_facts, "Shop: %s") was unreachable (choices=["listing"]). Removed with the shop_facts field; the listing branch runs unchanged. The twins' stop_reason = "single_page_mode" if args.mode == "product" is now "completed", as in playwright_scraper (same value at runtime).
• Comments telling another site's story as this one's: /p/<slug> hub + akamaihd.net (now attributed to sibling tokopedia-scraper), "hub category" / "taxonomy hub", CATEGORY vs SEARCH grid over GraphQL, keyword=kopi + UTM tracking tail, detail-page buy box (engines); Akamai 394-byte "Access Denied" and "lots" (captcha_solver.py, replaced with the README's headless 0/4 vs headful 4/4 and PerimeterX's widget-less stub); "Oops, produk nggak ditemukan", HTTP/2 stream reset, "Orlando search", shop_rating (output_writer.py). Dangling "See 'No DataDome solver' below" removed.
• SECURITY.md said "no releases or version tags"; v0.2.0/v0.2.1 exist.
• Access badge runs without an account -> runs without an account · residential IP, matching README's "Do I need to buy anything?" (measured from a residential address; the canary notes a datacentre runner is worse).

Repo metadata

• Topic price-monitoring removed (a vendor tile carries no price: README measures 0 price nodes over 1,952 tiles), restaurants added.

Kept, deliberately

• scraper_api_client.py docstring "a sibling repo where this path reaches 3 of 50 cards": names the sibling.
• captcha_solver.py "The sibling repo ... image captcha" paragraph, farfetch-scraper notes; product_parser.py / smoke_test.py amazon-scraper review-count notes; .gitignore craigslist note: sibling provenance.
• product_parser.py "datadome" in the vendor marker list: a generic vendor name, not donor prose.
• CHANGELOG.md released sections (incl. mediamarkt mention): history.
• smoke_test.py deliveroo URL: a negative-host fixture.

Not changed, found in passing

• selenium_scraper.py and puppeteer_scraper.py accept --category but never record it as run_label, while their help says they do (playwright does). That is twin drift in code, not donor text, so it is reported here rather than fixed.
• The twins still log "This listing holds %d product(s)" behind a total_available that is always None on this site.

Suite: 535 checks, OK before and after (engine groups skipped, engines absent). ci_checks.py --secret-check --help-check --sample-check green.

🤖 Generated with Claude Code