What it does: Every run is recorded with its trigger payload, its steps, their outputs and their errors. A failing step retries three times with exponential backoff; when the budget runs out the run is dead and appears on the Failed view. How to access: Flow Forge > Runs, then the Failed view. Notes:

• A failed row shows the step it stopped on, the error it stopped with, and a preview of the payload it was carrying (400 characters) with anything token-, key- or password-shaped blanked out — the same masking the dry-run report uses, from the same single regex.
• Retry from failed step (Free) writes a new run rather than rewriting the old one. It starts at the step that failed, with the original payload, so steps that already succeeded do not run twice; the original row stays exactly as it was and the new one records which run it came from (runs.retry_of).
• A retry is refused if the flow has been edited since the run failed. The cursor is a position in a step list that both editors delete and rebuild on save, so after an edit that position is no longer that step — the operator is told, rather than handed a button that silently starts in the wrong place.
• The Failed view shows fifty rows and carries a flow selector listing only the flows that actually have dead runs, with counts. There is no pagination.
• Pro adds “retry every failed run of this flow”, fifty per press, each from the step it stopped on, with an honest report of how many could not be restarted and why.
• Run history is pruned by a daily clean-up against the Settings > Log retention window (1–365 days). Runs still executing, runs parked on a wait or an approval, and the most recent run of every flow are never pruned.