scrapeless blog cover page

TL;DR

  • Define task success as a verified business outcome with evidence tied to the task.
  • Record navigation responses, browser actions and outcome checks as separate fields.
  • Use one denominator per metric. Track uncertain and partial outcomes explicitly.

An HTTP response describes a request

A browser agent navigates to a supplier portal and gets HTTP 200. Its task is to export a report for an exact account and period. The page could render the login screen, display last month's data or produce a file with the wrong header. Each case can have successful HTTP requests in the trace. The workflow still needs its own completion test.

This distinction matters in managed browser automation because a single browser task spans multiple requests and actions. Under HTTP status code semantics, a 202 response indicates that processing was accepted and has yet to complete. An accepted export job needs a separate observation when its result becomes available.

Write an acceptance rule for each task type

Start with the business object the user asked for. Give it an identity, required fields and a matching evidence source.

For a read-only export, the rule could require a verified account and period, a worker-owned copy of the file, and a CSV header plus required records that pass validation. For a booking, require a confirmation tied to the intended appointment details. If a booking may legitimately take time to finalize, define a pending state and record how the workflow can reconcile it.

Treat empty results as an explicit business state. For a report where zero records is allowed, verify that the correct filters finished loading and retain the application's empty-state marker. A blank file produced while the page is loading has different evidence.

Keep the rule version on the outcome record. When the report schema changes, this field makes a change in measured success easier to investigate across releases.

Give the verifier a record it can evaluate

An illustrative application-owned result for a report export might include four groups of fields.

  • Task identity includes the job ID, requested account and reporting period.
  • Browser diagnostics include relevant response statuses, action history and the run reference.
  • Outcome evidence includes the evidence reference for account and period, verified artifact path and validation output.
  • Decision includes the outcome, reason code and acceptance-rule version.

The worker can check identity, artifact presence and schema mechanically. Store enough evidence for a reviewer to understand why the check passed. An image of a successful-looking page is useful during review. For recurring workflows, prefer a structured confirmation ID or artifact hash when the application provides enough information for a deterministic check.

Use a free trial to test the acceptance rule

Use the Scrapeless Agent Browser free trial to run a small batch of approved test tasks. Connect your automation to Scrapeless Agent Browser and keep the verifier in your application. Include a valid export and a wrong-period case in the same pilot. The verifier should make different decisions even when both pages are reachable.

Report useful outcomes per task

For tasks that reached their reporting deadline, calculate verified success rate = verified tasks ÷ all eligible tasks. Deduplicate by job ID. Assign one final status per task for that reporting window. Use separate states such as verified, partial, failed, and unconfirmed with defined reason codes. Track cancellations and eligibility so changes in the denominator are visible. In-flight tasks whose deadlines have yet to arrive belong in an operational view.

Report attempt diagnostics alongside that task metric. A task can contain several browser sessions or multiple navigations. Track session-start failures, navigation outcomes and runtime there. A system with stable HTTP status codes and a rising wrong-period count needs investigation at the workflow or application layer.

Add browser time per verified result when comparing runs. Sum browser time spent on all eligible tasks, including failed work, then divide by the number of verified outcomes. If there are zero verified outcomes, report the ratio as unavailable and show the total browser time. Break down failures by workflow version and site or environment so a changed mix of tasks remains visible.

Run the acceptance rule across a stable case set in browser regression testing. Include wrong identities, delayed results, valid empty data, expired sessions and malformed artifacts. Sample some verified records manually. A false positive can otherwise remain hidden behind a high reported success rate.

Conclusion

Choose one browser workflow and write down which durable result proves completion. Record that check per job and keep transport data with its own diagnostic fields. The next time a dashboard shows high request success, you can trace whether the requested work also finished.

FAQ

Can a task be verified without a confirmation screen? Yes. An authorized read-back or a correctly scoped artifact can provide the necessary evidence if the acceptance rule specifies it.

How should a run with unknown final state be recorded? Use an unconfirmed status and preserve the last known step. For a state-changing action, check the application's durable record before taking another action.

Is a model's completion message enough evidence? A model can summarize the evidence or flag ambiguity. Evaluate the task using the saved facts and a versioned outcome rule.

Should a valid empty report count as success? Yes, when the task allows empty data and the verifier can establish that the correct query finished.

Scrapeless complies with applicable local laws and regulations and does not access or collect data from private, confidential or restricted sources without authorization.