AUTOPULSE / guide

Start with evidence you can explain.

Choose one client and workflow. Define the business item that should be delivered and the time allowed for completion. Export only the evidence needed to assess that promise.

1. Export the expectations

Use one source row per business item. Keep a stable, opaque correlation ID that the destination also records. The source event ID and correlation ID must each be unique in this export.

execution_status must be success, failed or unknown. expected_count is the positive number of incremental units promised for this item—not the workflow’s lifetime total. A source run marked success does not establish delivery.

2. Export the destination evidence

Use a unique acknowledgement ID for each event. status is completed, accepted or failed. Only completed records contribute to the observed count. HTTP acceptance, queued work and receipt of a request do not prove that a lead or order was created.

observed_count is the incremental completed count represented by this acknowledgement. Do not repeat cumulative totals or emit multiple completed rows for the same underlying delivery. Over-counts are flagged for investigation. A header-only destination file is supported when there are no records.

3. Make scope and time explicit

Both files must refer to the same client, workflow and meaning of a business item. AutoPulse does not detect a wrong-client export for you and will not infer aliases. Use UTC timestamps such as 2026-09-12T14:00:00Z. The form’s coverage fields are also UTC, regardless of your computer’s timezone.

Declare destination coverage complete only when it includes all relevant acknowledgements across the stated interval. Sampled, filtered or delayed exports are not complete. A missing outcome is supported only when the source-to-deadline window is covered and has expired. Otherwise the result stays pending or evidence incomplete.

4. Investigate, then reassess

Open a finding to see its source row, destination rows, counts and deadline. Acknowledge an exception to mark it for investigation in this session. That is not a resolution. Replace both evidence files with a fresh, consistent snapshot to see whether delivery is now evidenced; a late delivery remains late.

5. Prepare the client update

Prepare client report carries the exact assessment into a readable preview, with all six outcome counts, source references, destination evidence and next actions—even when the workbench is filtered. In the internal workbench, it uses the selected review, not automatically the latest. Historical status and freshness at the last read stay explicit; acknowledgement is not resolution.

Add optional reviewer commentary or a proposed follow-up as operator-authored context. These drafts stay in this tab for this exact selection, not in the database. Closing and reopening the report keeps them; switching review, replacing evidence, clearing or reloading discards them. Download self-contained HTML for reading and full evidence JSON for inspection. These are point-in-time drafts, not live dashboards or independently verified records. No report is sent automatically.

Format and privacy limits

Each CSV may contain up to 10,000 data rows and 5 MB. IDs use letters, digits, dots, underscores, colons or hyphens, with a maximum of 128 characters. Counts are whole numbers up to 1,000,000. Invalid dates, duplicate IDs, malformed rows or excess limits reject the entire file—there is no silent partial import.

Remove personal names, emails, addresses, message contents and credentials. Pseudonymise client labels. CSV assessments run in your browser. Stored promises and signed event metadata use your private account workspace. Native connectors, unattended monitoring, outbound alerts and AI processing are not included.

Open the outcome workbench →