[jehrr]

Leftovers from the sibling repos kose-scraper was bootstrapped from, which described those sites as if they were this one. Replacements come from this repo's own README, CHANGELOG, code and live_results; nothing new was measured.

Fixed

• --pages help (all three engines): named --mode category / --mode search (this repo has listing / product) and said a category past its end is "a served, empty grid". The README measured the opposite: it serves the last page again.
• Playwright blocked-page error: said the edge "refuses curl, python-requests and friends outright" and cited a Hetzner measurement. The README measured all of those served. It now says what pyppeteer and Selenium already said.
• Parse-failure error and image-coverage log (all three engines): pointed at a JSON-LD ItemList. A listing here has no JSON-LD (0 blocks in 8 captures), so both now point at the tile markup.
• --mode product wording: the log line said "variant(s)", --concurrency said "however many variants it holds", and scraper_api_client --mode help said "one row per variant out of the page's ProductGroup". A Kosé product page is one Product block, so one row.
• --retries help spoke of "a hub page". --dump-html help pointed at a TROUBLESHOOTING.md this repo does not have. It now points at the README's "Traps that look like bugs".
• blocked_by: pyppeteer and Selenium recorded cloudflare (hard block) (no Cloudflare has been observed here), while Playwright recorded edge refusal, so the engines disagreed on stop_reason. All three now use edge refusal.
• diff_runs.py tracked Montblanc columns (collection, sub_collection, color, size, special_edition, base_sku) that this row model does not have, plus Montblanc stock counts in the comment. It now tracks tax_rate and subcategory, and subcategory / variant_of are product-mode-only.
• Issue templates were Etsy's (DataDome, t=bv, shop_rating, an etsy.com URL, a missing TROUBLESHOOTING.md). They were rewritten from this README.
• captcha_solver.py: the BBB Turnstile docstring and the "what BBB does instead" block were replaced with this site's measurement (no challenge in 13 captures).
• Engine comments: Montblanc /en-fi/... URLs and 280/sz/start= arithmetic became this site's /c/c15/ 13 pages / 304. BBB's find_country was in two locale comments. An uncredited /p/<slug> hub anecdote is now credited to tokopedia-scraper. "business-at-a-location" became "product". "Montblanc" in output_writer/page_flow now names montblanc-scraper as a sibling.
• smoke_test.py: the sidecar fixtures came from a montblanc.com run (en-fi, sort_requested, 280 results). They now use the measured /c/c15/ run and assert the extra fields this engine actually writes (pages_available, page_size, canonical_url, route), plus that there is no locale. There is one more check than before. The prose naming BBB/Montblanc as this site was fixed.
• .dockerignore: bbb_businesses.* became kose_products.* / kose_products_*.*, the argparse default, mirroring .gitignore. The Dockerfile example --out /out/writing-instruments became kose_products.
• The CHANGELOG has a new [Unreleased] section. There is no version bump.

Kept deliberately

• Sibling comparisons that name the sibling repo and state a lesson: "the opposite of bbb-scraper" (sort, page cap), "Unlike montblanc-scraper" (refused, image_url floor), "mediamarkt-scraper needs" (Omnibus), "quora-scraper made the same call", the tokopedia binding-check note, and the farfetch notes in captcha_solver.
• The .gitignore craigslist-scraper note, which is family-wide.
• The akamai / errors.edgesuite.net marker in product_parser, which is family detection coverage, and CLAUDE.md's assets.mmsrg.com reference.
• Released CHANGELOG sections were left untouched.
• category still differs between modes (c15 on a listing row, スキンケア on a product row) and is still tracked by diff_runs as before. That is pre-existing behaviour, not a donor leftover.

Verification

• smoke_test.py: 463 passed before, 464 passed after, 0 failed. The same 4 engine-absent skips appear both times.
• ci_checks.py --help-check --sample-check passes, and --secret-check finds nothing.
• The repo metadata (description, topics, homepage) is clean, so nothing was changed there.

🤖 Generated with Claude Code