> ## Documentation Index
> Fetch the complete documentation index at: https://docs.heralded.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Verifying actions

> How Heralded judges whether finished work changed the answers.

When an [action](/concepts/actions) is done, Heralded doesn't count it as a success because a later report looks better. It looks for evidence that this piece of work moved the answers it was aimed at. How it does that depends on the kind of work.

## Site fixes

A site fix is settled by the first daily update or weekly report that checks the fix on your live site. If the check passes, the action is marked **Fix is live**. If it fails, it is marked **Not live**. Heralded waits up to seven updates or reports for a check it can use. If none comes, the action settles as **Done · not measured**.

## Work aimed at prompts

Content, rewrites and outreach are aimed at prompts, and Heralded judges them on [Mentioned](/methodology/mentioned-and-recommended) for the prompts they targeted.

1. **Baseline.** Heralded takes the 14 days that end a week before the action was filed. The week in between is left out, because that is usually the dip that prompted the work.
2. **After.** It pools the answers after the work was delivered: 28 days for new content, competitor gaps and page rewrites, and 14 days for everything else, including title and description rewrites, outreach, correction requests and prompt tracking. Work on a page also waits, up to 28 days, for Heralded to see the page change. If two checks in a row find the page removed, Heralded stops checking it and the work ends **Inconclusive** because the page was removed. A page that is missing once and comes back changed is measured from that change.
3. **Control.** It measures the same change on prompts no other work touched, and subtracts it. A market that rose for everyone doesn't get credited to your work.
4. **Test.** What is left must pass a [paired test](/methodology/change), so its 95% range has to clear zero.

Heralded needs at least three targeted prompts and three untouched prompts to judge. With fewer, the engines' normal variation would swamp any move one piece of work could make. Both groups also need enough [paired units](/methodology/change#the-paired-test) for an interval. If too few answers pair in either group, the outcome is **Inconclusive**.

## Outcomes

| Outcome | Means |
| - | - |
| Verified by a later update or report | the targeted prompts moved, net of the control. The row shows the gain, such as "+4 pts Mentioned". |
| Measured · no move in the window | measured, and nothing moved beyond the control |
| Fix is live / Not live | the check on the live site passed or failed |
| Inconclusive | there wasn't enough evidence to judge, with the reason, such as too few prompts to compare, a page that did not change or a page that was removed |
| Done · not measured | delivered work that no report could settle |

On Home, the **Paying off** panel under **Results** counts the work that verified and shows the newest.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.