All docs pages (49)

Investigate

Something broke in Salesforce? Pick the symptom and see the Setup and automation changes made before it, ranked by what to check first, with the evidence.

Updated 2026-11-09

Investigate answers "something broke: what changed before it?" It reads the org's saved evidence and recent Setup changes, then ranks the changes by what to check first, with the reasons. The ranking is fixed rules with points, not AI. Open Investigate in the main menu.

How to use it

  1. Under What broke?, pick a Symptom:
    • A flow is failing: type in Find the flow, click Search, then click the flow.
    • A save fails with a validation error: enter the Object API name and paste the Error message.
    • A field isn't being updated: enter the Object API name and Field API name.
  2. Optionally set When did it start? (optional).
  3. In Setup changes to read, pick 1, 3, 7, 14, 30, 90 or 180 days before.
  4. Click Investigate.
  5. In What to check first, click a row to see Why it ranked here.

What you see

What to check first. Each row shows the change, its kind (such as Validation rule, Permission or Flow), when and how long before the failure, who made it, and its points with the signals that earned them, such as "Changed the failing flow" or "Exact error message match". A signal that can't be decided shows as "can't tell" and earns nothing. The table exports to CSV or JSON.

The window ends at the earliest known failure time (the oldest saved failed run or the start you entered, whichever is earlier), or now if neither is known, and reaches back the days you chose. Changes after the failure are not read.

Per symptom:

  • A flow is failing: The failure as Salesforce saved it, a Version comparison of the failing version with the one before it, and Depends on.
  • A save fails with a validation error: the object's validation rules, matched against the message you pasted.
  • A field isn't being updated: the field's type, whether it is updateable or a formula, the automations found that write it, and the save-order verdict from the Process Graph engine.

Timeline orders changes and failures in time. What was and wasn't checked lists each source read, read in part, or refused. Export offers Copy as Markdown and Download .md. Notes holds up to 5,000 characters of your own notes.

What it reads and writes

From the org you are connected to, it reads:

  • Setup Audit Trail entries in the window.
  • Failing flow: the flow and its versions, failed runs (FlowInterview rows with status Error), the metadata of up to three versions (Tooling API), up to three other flows changed in the window, and describes of up to two objects.
  • Validation error: the object's validation rules (Tooling API), the formulas of up to 25, and the names of their last editors.
  • Field not updating: flows triggered on the object (up to 200), active Apex triggers on it (up to 50), workflow field updates on it (up to 50), and the object's describe.

It changes nothing in the org. Before a run it estimates the API calls the run can make and stops if that is more than 95% of the org's remaining daily calls. Notes are saved only in your browser, per org and per flow or object; Clear deletes them after a Delete these notes? confirmation.

Limits

  • At most 2,000 Setup changes (those closest to the failure) and 2,000 failed runs (the oldest) are read; the page says when a limit was hit.
  • Setup Audit Trail records Setup changes only, not changes to record data.
  • Salesforce does not save every failed flow run, so "no failed runs" is shown as a coverage fact, not as "nothing is wrong". If Salesforce refuses to return them, the page says they were not checked.
  • Salesforce hides the metadata of flows installed from a managed package, so their versions can't be compared.
  • For a field not updating, flows are found by the object they are triggered on, and an Apex trigger counts as a writer when its code contains the field name. A flow on another object that updates this field is not found.