The Error Rate Trends chart measures the difference in performance between two moments that ask the same thing in two different conversation scenes, for example a moment in Interview #1 and its matching moment in Interview #2 in the Accenture Debra module. Each pair of moments shows on the chart as one topic. Below the chart shows the 5 different topics (pairs of moments across both scenes) with the topic name at the bottom.

The chart appears for a module only after two things are done: your setup in Create, and a short engineering step.

In Create

The chart in the portal is configured via the Final Outcome Feedback Groups. There are three steps:

1. A Final Outcome Feedback Group after each scene, one for each moment of the pair

A Final Outcome Feedback Group is where a moment gets measured, and it is where you assign the moment pairs. After the first conversation scene, add a Final Outcome Feedback Group that measures the first moment of the pair. After the second conversation scene, add a second Final Outcome Feedback Group that measures the matching moment there. Each group sits right after the conditional branching that reads that moment. Five pairs of moments means ten groups.

2. A unique title on each group, never renamed

The group’s title, at the top of the group card, is the only link between the group and the chart. Engineering copies it word for word, and a mismatch or a later rename silently drops that pair. In the Debra module, for ease of use in Create, the group is labeled FOFG HC-1: FOFG for Final Outcome Feedback Group, and HC-1 because it is for the moment measuring the module topic Household Composition in the first scene. The matching group after the second scene is labeled FOFG HC-2.

3. Two outcomes per group, with “fail” in the wrong one

Each group gets exactly two outcomes, one right and one wrong. Put the word “fail” in the wrong outcome’s Analytics tag name, the box in the middle of the outcome card, as in HC-1 Failure. “Failure”, “Failed” and “fails” all count. Keep the word out of the right outcome’s tag. That tag is the field the chart reads to tell right from wrong; the outcome’s name at the top of the card is only used if the tag is left blank, so the name itself can say anything.

Hand to engineering

Send these three things. Engineering saves the pairing and adds the module to the chart’s list, and the chart shows only after both are done.

• The module’s identifier.
• For each pair of moments: the label to print on the chart, the first scene’s group title, and the second scene’s group title.
• The two column labels, for example Interview #1 and Interview #2.

These changes appear the next time the reporting tool is deployed, not the moment they’re saved, so give it until the next release before expecting the chart to show up.

If something looks wrong

• No chart at all: engineering has not added the module yet, or none of the group titles matched.
• A pair is missing from the chart: one of its group titles in Create does not match what engineering saved, or it was renamed.
• Everyone shows as passing a topic: the wrong outcome’s Analytics tag name is missing the word “fail”.