E-commerce Product Extractor (single page)

Given a single public product-page URL (or a small bounded batch of URLs), this actor deterministically extracts structured product data:

url, status, title, price, currency, availability, brand, rating,
review_count, images[], sku, description, raw_prices[], method,
parse_confidence, extracted_at, error

What it is (honest scope)

This is single-page product extraction, not bulk store crawling. You give it the URL of one product page; it fetches that page once and parses the product data out of it. It does not spider an entire shop, follow category pages, or paginate through catalogues (that would require proxies and carries ToS/legal risk).

How extraction works (deterministic, code-only)

Extraction is pure code, tried most-reliable-first:

1. JSON-LD / schema.org Product — offers (price, currency, availability), brand, aggregateRating (rating, review count), sku/mpn/gtin, image, description. This is the highest-confidence path.
2. OpenGraph / product meta tags — og:title, og:image, product:price:amount, product:price:currency, product:availability, ...
3. Plain <meta> / <title> / <h1> + visible-text heuristics — currency regex for price, keyword regex for availability.

The layer used is reported in method, and a code-owned parse_confidence (high/medium/low/none) is attached to every record.

Cost-safety ($0 idle, $0 uncovered per run)

• No proxy — direct bounded HTTP GET.
• No headless browser — static HTML fetch only.
• No AI / LLM — pure deterministic parsing.
• No paid third-party API.

The only cost is Apify platform compute for the run itself.

Always-on security (SSRF-guarded, fail-closed)

• Private / loopback / link-local / reserved IPs are blocked (SSRF guard), re-validated on every redirect hop.
• A domain blocklist rejects login-walled / litigation-magnet sites.
• Hard caps: 5s connect / 10s read timeout, 2 MB body, 3 redirects.
• The actor never raises — every URL yields a structured record (with status and error populated on failure: failed / blocked / empty).

Input

field	type	notes
url	string	single product-page URL
urls	array of string	optional bounded batch (capped at 50 per run)

Output

One dataset record per URL with the fields listed at the top.