Back to Report Blocks

FeedbackAlignment

Analytics
Report block

Measures agreement between AI score results and human feedback using AC1, accuracy, class distribution, and confusion matrices.

What It Answers

  • How aligned is this score with human feedback?
  • Which labels are being confused?
  • Which scores need attention first?

Use When

  • Starting score optimization.
  • Comparing scorecard-level alignment across scores.
  • Checking whether recent rubric or prompt changes improved feedback agreement.

Avoid When

  • You need individual contradiction explanations; use FeedbackContradictions.
  • You only need marketing-friendly acceptance numbers; use AcceptanceRate.

Run From The CLI

Direct report commands are the simplest path for one-off usage. Saved report configurations use `plexus report config create` and `plexus report run`.

plexus feedback report alignment \
  --scorecard "Customer Service QA" \
  --score "Medication Review: Dosage" \
  --days 30 \
  --format json
plexus report config create --name "FeedbackAlignment Example" --file feedback-alignment.md
plexus report run --config "FeedbackAlignment Example"

Minimal Configuration

class: FeedbackAlignment
scorecard: "Customer Service QA"
score: "Medication Review: Dosage"
days: 30

Live Rendered Example

Rendered with the same dashboard component used by generated reports.

FeedbackAlignment Example

March 1, 2026 to March 31, 2026

Overall

-110.20.50.80
0.74
Agreement
010051.471.481.491.4
80%
Accuracy
Raw Agreement
96 / 120
96 / 120

Medication Review: Dosage

-110.20.50.80
0.74
Agreement
010051.471.481.491.4
80%
Accuracy
Confusion matrix
Yes
62
8
Yes
No
16
No
34
Predicted
Actual
Raw Agreement
96 / 120
96 / 120

How To Interpret It

  • High AC1 means the score and reviewers agree beyond chance.
  • Use the confusion matrix to choose whether to focus on false positives or false negatives.
  • Low item count should be treated as weak evidence, even if the gauge looks strong.