Test Apify Recency Actor RAG Batches in n8n (Manual Trigger)

Test Apify “Dataset Recency & Evidence Gate” Batches in n8n (Manual Trigger)

Need to quickly verify whether a saved crawl batch passes your recency policy before ingestion? This n8n workflow manually generates (or accepts) a batch envelope, sends only the required record metadata to Apify’s Dataset Recency & Evidence Gate Actor, and then either releases the original documents for downstream processing or blocks the batch with diagnostics.

What this workflow does

• Manual trigger in n8n: Runs when you start it manually, making it ideal for on-demand QA and incident checks.
• Creates/uses a saved crawl batch: Builds a synthetic “saved crawl batch” (or uses your provided batch envelope) including document items plus capture metadata such as crawl.loadedTime.
• Validates schema & prepares Apify input: Checks the batch format and constructs an Apify Actor input containing only record IDs, observed timestamps, and source references.
• Calls the Apify Actor via API: Sends the request to Apify’s “Dataset Recency & Evidence Gate” Actor and retrieves the evaluation report.
• Validates the result: Confirms the HTTP response and verifies policy settings, record alignment, and the aggregate pass/fail decision.
• Pass vs. fail behavior:
  • If the batch passes: Outputs each original document item for ingestion along with the gate record ID.
  • If it fails: Blocks ingestion and outputs the report summary and per-record results for troubleshooting.

Use cases

• Pre-flight testing a crawl batch’s recency window (based on as_of, max_age_seconds, and item crawl.loadedTime).
• Debugging why a batch was rejected by Apify’s recency/evidence policy—without running the full pipeline.
• Operational QA for SaaS ingestion workflows where stale content must never enter downstream indexes.

Technical details

• Nodes: Manual Trigger, HTTP Request, Code, If, Sticky Note.
• Apify integration: HTTP Header Auth credential using Authorization: Bearer <APIFY_TOKEN>.
• Inputs you replace: as_of, max_age_seconds, dataset_id, offset, and batch items with crawl.loadedTime.