Correct processed items¶

Compare a flow's output with the source story and correct an article-specific mistake. This tutorial removes a duplicate place from the Duluth cooling-program story.

You'll learn¶

How to verify a result against highlighted source text.
How to remove an unsupported or duplicate extraction.
Which corrections belong on the processed item and which belong in Stylebook.

Before you begin¶

Complete Extract people and places. Open the successful Tutorial Flow run for Duluth officials announce summer cooling program, then select View on its processed item.

1. Review the item¶

Start with Info. Confirm that the item succeeded and that the headline preview identifies the expected story.

Then open Places. The story appears beside the map and place table. Highlighted passages show the evidence that produced the place results.

Duluth story with five geocoded place results

The verified run contains five rows, but the article supports only four distinct places. Lake Superior appears twice:

Lake Superior, US is correctly identified as a natural feature.
Lake Superior is also listed as Region (City), which is an incorrect duplicate.

Do not judge a result from its confidence or map alone. Compare its name, type, address, geometry, and highlighted source passage.

2. Remove the duplicate¶

In the Lake Superior row with type Region (City), select the red Remove from story button.

Agate explains the scope before making the change: mentions for this article will be removed, and a saved place used by no other stories may be unlinked from the catalog.

Confirmation before removing the duplicate place

Select Remove from story to confirm.

3. Confirm the corrected result¶

The incorrect row disappears. The remaining table contains:

St. Louis County, Minnesota;
Duluth, Minnesota;
Lake Superior as a natural feature;
Duluth City Hall.

Corrected place table with the duplicate removed

The correction applies to this processed story. Agate preserves the flow's machine output as part of the run record and applies editorial changes as a review layer.

4. Know when to use Stylebook¶

Removing a result from a story is different from editing shared canonical knowledge.

Correct the processed item when a person, place, mention, tag, or field is wrong for this article.
Open Stylebook when the shared identity itself is wrong—for example, a canonical name, geography, metadata value, or connection used across stories.

The Open in Stylebook action beside a matched result takes you to that canonical record. Do not rename a shared place merely to fix one article's extraction.

5. Review the other tabs¶

Use the same source-first check throughout the item:

People — verify names, titles, affiliations, mentions, and quotations.
Meta — review each category with its rationale and confidence.
Info — add or correct article-level source, URL, headline, author, or publication date when available.
JSON — inspect or download the structured output.

Use Add person, Add place, or Add tag only when the source supports something the flow missed. Remove results that are unsupported rather than trying to make them fit.

Next step¶

Continue with Rerun processed items safely. Rerunning regenerates the item from its original run's saved flow settings and clears its run-level review changes.