CLOUDCRUISE UNITED

ORANGE path

ExtractDatamodel — LLM_DOM Fallback

STATIC EXTRACT_DATAMODEL reads from //table[@id='product-table']. Append ?broken=true to rename the id so the static XPath misses. If the node has enable_llm_dom_fallback: true and the backend has ENABLE_LLM_DOM_FALLBACK=true, the run retries with execution=LLM_DOM and extracts the same data straight from the page HTML.

Current table id:product-tableBreak selector

SKU	Name	Price	Status
P-001	Wireless Mouse	$29.99	In stock
P-002	Mechanical Keyboard	$149.50	In stock
P-003	USB-C Hub	$39.00	Out of stock
P-004	Monitor Stand	$79.99	In stock
P-005	Laptop Sleeve	$24.50	Out of stock

Workflow node config

{
  "action": "EXTRACT_DATAMODEL",
  "parameters": {
    "selector": "//table[@id='product-table']",
    "execution": "STATIC",
    "enable_llm_dom_fallback": true,
    "extract_data_model": {
      "type": "object",
      "properties": {
        "products": {
          "type": "array",
          "path": "//table[@id='product-table']//tbody/tr",
          "items": {
            "type": "object",
            "properties": {
              "sku":      { "type": "string", "path": ".//td[1]" },
              "name":     { "type": "string", "path": ".//td[2]" },
              "price":    { "type": "number", "path": ".//td[3]" },
              "in_stock": { "type": "string", "path": ".//td[4]" }
            }
          }
        }
      }
    }
  }
}

Test plan

Backend env: ENABLE_LLM_DOM_FALLBACK=true
Workflow flag: enable_xpath_recovery=true
Node param: enable_llm_dom_fallback: true
Run against this URL with no query — STATIC succeeds (control)
Run against ?broken=true — STATIC fails, LLM_DOM rescues
Verify: run completes, data extracted, healing-table row appears, logs show LLM_DOM fallback: STATIC extract_datamodel failed; retrying via LLM_DOM