Troubleshooting

Diagnose problems in pipeline order. Changing a late stage cannot restore material that disappeared earlier.

#I cannot create a goal

Check whether your email is verified and whether the account has reached a goal, source, retention, or integration capacity limit.

#My new goal is empty

1. Confirm that the goal has an active source.
2. Wait for the next scheduled crawl. A new goal does not inherit older material.
3. Inspect source health, last crawl, and 24-hour throughput.
4. In the Feed, compare the Before filter view with the After filter view.
5. Temporarily disable filtering or broaden its instruction.
6. If articles pass, inspect cluster minimum, similarity, and holdback.

#Relevant material is rejected

• add examples of material that should pass;
• remove contradictory conditions;
• enable full-content evaluation when the title and excerpt are insufficient;
• verify that the selected model and key are available;
• open the Filter tab and read the reasoning behind the decision on that article.

#Irrelevant material passes

Transient provider errors can fail open. Check the decisions in the Filter tab first. If the system is healthy, add explicit rejection cases instead of narrowing the topic with a single keyword.

#No result is created

• the cluster may be below its minimum size;
• holdback may still be running;
• the workflow may be a draft or fail validation;
• a manual cluster requires an explicit execute action;
• the result may be waiting in a queue or for research.

#A result was not delivered

1. Check whether it is waiting for approval in Post moderation, on the Needs approval tab.
2. Confirm that the Action step uses the HTTP address, Email, or Integration channel, not No delivery.
3. Check that Delivery mode is Production.
4. Verify that the Recipient URL is public and returns 2xx within 30 seconds.
5. Fix 4xx responses before retrying; most are permanent failures.
6. Allow automatic retry for 429, 5xx, and network errors.

#I cannot create a knowledge base

On the Free plan the knowledge-base catalogue is read-only. You use the default bases every account receives at registration — you can attach them to goals and query them — but you cannot create your own or run a synchronisation. Creating your own bases requires a higher plan.

Knowledge-base usage is not shown on the Plan & limits screen; that page covers other axes. The state of each base lives in Library → Knowledge bases.

#A knowledge base stopped synchronising

• Automatic sync paused after 3 failures means three consecutive runs failed and the scheduler stopped calling the source. It does not resume on its own: fix the source (address, authentication, response format), then press Resume on the base. That resets the failure counter.
• Plan entry limit reached stops only new entries. Everything already indexed stays and keeps working.
• A base in live mode has nothing to synchronise — the external API is queried only at the moment the model reaches for it.
• A base not attached to any goal is never used, even when it synchronised correctly.

#The token works in WordPress but not elsewhere

This can be expected behaviour. Tokens starting with sh_ serve selected integrations only; they do not replace ordinary sign-in to the application.

#Reporting a problem

Include the goal name, approximate event time, the stage where data disappeared, the source/result/channel status, the receiver’s HTTP status, and the result identifier. Never include secrets or tokens.