A successful Run Now result proves only that a filter acted during that manual run. It does not prove that the filter is attached to the event and folder where new mail arrives. If manual and automatic behaviour differ, keep the matching criteria stable while you trace the execution path.

Record the two situations separately

For the manual success, note the account, selected folder and filter used. Then record where a newly received message first appears and which event you expected to activate the rule. Use an existing harmless example; do not repeatedly submit messages just to manufacture another arrival.

Inspect Message Filters for that same account. Thunderbird keeps account filter lists separately. Under the application conditions, confirm whether the rule includes Getting New Mail, a manual run, After Sending or Archiving. A manual option alone cannot establish an incoming-mail trigger.

New-mail filtering normally acts on the inbox. Running a rule manually against a project subfolder can therefore succeed even though ordinary automatic new-mail filtering does not sweep that subfolder. Archive and after-send rules apply to the messages involved in those events. Decide which actual event the workflow requires before moving mail around to fit a test.

Look at timing and action order

For a rule that examines the message body, compare its before/after junk-classification setting with Mozilla's guidance to use after classification when body filtering causes difficulty. Preserve the current value and the observed failure so the adjustment has a specific reason.

Inspect See execution order if Thunderbird reports that actions will run differently from their listed order. A move can affect where a later action finds the message. Do not add a reply or forward as a diagnostic action, because that introduces an external message instead of clarifying execution.

Finish with a concise explanation of the discrepancy: for example, manual processing targeted a subfolder, while automatic processing was tied to inbox arrivals. After an authorised setting correction, observe an ordinary matching arrival and record its path. That evidence identifies the intended trigger without broadening the rule's criteria unnecessarily.

Source: Thunderbird documentation.