Skip to main content
Thread Outcome Feedback is opt-in. It is not on by default — it must be enabled for your organization or workspace. Contact your Gradial team to get access.
A thread can finish successfully and still miss the mark. The work completed, the steps ran, nothing errored — but the tone was wrong, the CTA requirements were ignored, or it simply isn’t what you asked for. Today that judgment usually lives outside the product, in Slack messages, calls, and personal notes. It never makes it back to the thread that produced the work, so there’s no way to connect a good or bad outcome to the specific thread, version, and agent behavior behind it. Thread Outcome Feedback closes that loop. Once a thread is complete, you can mark the outcome good or bad and add optional context. That turns “technically finished” into “actually approved” — and gives Gradial a concrete signal to learn from.

​
Giving feedback

When a thread reaches a terminal state and has no active workflow running, a good/bad feedback control appears on the thread.
1

Open the completed thread

Feedback becomes available once the thread has finished. It does not appear while a workflow is still in flight.
2

Mark the outcome

Select the positive or negative rating depending on whether the work landed.
3

Add context (optional, but useful)

Describe what worked or what missed. Specifics are what make the signal actionable:
“This missed the CTA requirements and used the wrong tone for the audience.”
“This matched the brief and is a good pattern for future launch-page QA.”
Giving feedback is always optional — a completed thread never requires a rating before you can move on. Anyone who can access the thread can see the feedback on it. Submitting feedback follows the same edit boundary as the thread or project itself.

​
What happens with your feedback

Submitting feedback posts a follow-up turn to the agent behind the scenes, so the thread can respond to your judgment rather than simply recording it.

​
Negative feedback opens a recovery loop

When you mark an outcome bad, the agent can re-engage with a concrete plan to address the issue — so a missed result becomes the start of a fix rather than a dead end. Your written context is what that plan is built from, which is why specifics matter more than a bare thumbs down. Negative feedback also becomes a high-signal example for evaluation, prompt tuning, and workflow debugging.

​
Positive feedback identifies reusable patterns

Marking an outcome good flags the agent behavior behind it as something that worked, and the agent can confirm back what it understood to be successful. Those patterns become candidates for reuse — saved as examples, or promoted into a Skill.
If a thread produced an approach you want your whole team to follow, feedback is the signal — but a Skill is the durable artifact. Promotion is deliberate: marking a thread good does not save a Skill on its own.

​
Feedback history on a thread

Feedback is recorded as append-only checkpoints in the thread timeline. Nothing is overwritten, so the full arc of an outcome stays visible — a thread can show negative feedback, the agent’s recovery, and later positive feedback, in order. Each checkpoint is tied to the thread, organization, user, and the specific thread version being judged. Feedback stays attached to the state of the work it actually assessed, even after the thread moves on.

​
Availability

Until the feature is turned on, threads behave exactly as they do today, with no feedback control.

​
Current limitations

  • Binary rating only. Good or bad — there is no multi-point scale or category breakdown.
  • Completed threads only. Feedback appears only once a thread is finished with no workflow actively running.
  • No analytics dashboard in this release. The release creates the feedback signal; aggregate reporting can follow.
  • Not automatic training. Feedback creates labeled examples for evaluation and improvement. It does not retrain the agent.
  • Not automatic Skill saving. A good thread becomes a candidate for a Skill or saved pattern, but promoting it stays a deliberate step.
  • Learning System → — how Gradial encodes and applies your organization’s knowledge
  • Memory → — carrying knowledge from completed threads into later work
  • Managing Skills → — turning a successful pattern into a reusable capability
  • Evals → — measuring agent output quality