Scrape a public web page into clean markdown in one call: runs a crawler, waits, scores the result (completeness, bot-wall detection), and automatically retries with stronger settings when quality is low. Costs Apify credits per run. For arbitrary actors or fire-and-forget runs use run_actor instead.

Start an allowlisted Apify actor with an explicit input object. Returns run_id and dataset_id handles for get_run_status / fetch_dataset_items / validate_dataset. Costs Apify credits. Prefer scrape_url for simple page scrapes.

Check whether an actor run has finished and where its dataset is. Free and safe to poll.

Read items from an actor run dataset with pagination, field projection, and a hard token budget. Start with response_format "summary" to see the shape cheaply, then fetch "items" with a fields projection. Items arrive as JSON in the text content; structured content carries pagination metadata (total, next_offset, truncated).

Score a dataset before trusting it: schema pass rate (if a JSON Schema is given), field completeness, duplicate rate, and bot-wall detection. Use the score to decide between consuming the data and retry_low_quality_run.

Re-run an actor with progressively stronger anti-blocking settings (residential proxies, then a browser crawler) until the dataset quality score reaches the threshold or attempts are exhausted. Each attempt costs Apify credits.

Turn scraped items into embedding-ready documents: token-bounded chunks with overlap, source attribution, stable content-hash ids for idempotent vector upserts, and selected metadata. Paginate with offset/limit for large datasets.