Validated against live product pages; fixed three edge cases (also present in
the original JS scraper) surfaced by specialty AkzoNobel products:
- Sample image: only accept real product images on the NIC CDN
(images.nicindustries.com/prismatic/products), preferring full-size over
thumbnail. Dropped the loose "prismatic|powder|color" fallback that grabbed
the site logo on products with no image.
- SDS/TDS/app-guide links: require the href to be an actual document (NIC CDN
or a .pdf) so a generic /documents nav link isn't captured as the SDS.
- Description: also stop at PRODUCT SUPPORT / PRODUCT COLLECTIONS / CUSTOMER
SERVICE so less page footer is captured (app-side StripBoilerplate cleans the
rest).
Structural fields (sku, color, price tiers) verified correct on live data.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
spouliot
da2bb46d5a