Back to Report Blocks

FeedbackAlignmentTimeline

Trend
Report block

Plots stored feedback alignment across time buckets with rolling minimum samples so teams can see whether agreement is improving or drifting.

What It Answers

  • Is alignment improving over time?
  • Did a release or rubric change move AC1?
  • Which completed periods had no feedback?

Use When

  • Reviewing week-over-week score health.
  • Checking post-release drift.
  • Separating recent trend from long-window averages.

Avoid When

  • You only need the latest aggregate; use FeedbackAlignment.
  • You need feedback volume only; use FeedbackVolumeTimeline.

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 timeline \
  --scorecard "Customer Service QA" \
  --include-score "Resolution Accuracy" \
  --exclude-score "Experimental Variant" \
  --bucket-type calendar_week \
  --bucket-count 6 \
  --rolling-min-items 100
plexus report config create --name "FeedbackAlignmentTimeline Example" --file feedback-alignment-timeline.md
plexus report run --config "FeedbackAlignmentTimeline Example"

Minimal Configuration

class: FeedbackAlignmentTimeline
scorecard: "Customer Service QA"
include_scores:
  - "Resolution Accuracy"
exclude_scores:
  - "Experimental Variant"
bucket_type: calendar_week
bucket_count: 6
rolling_min_items: 100
show_bucket_details: false

Live Rendered Example

Rendered with the same dashboard component used by generated reports.

FeedbackAlignmentTimeline Example

Alignment metrics over complete historical buckets

Mode: Single score
Buckets: 4 x calendar_week (complete periods only)
What these charts measure
  • Each chart measures how often stored production score answers agreed with the final human correction recorded in feedback.
  • The original score answer is the value captured on the feedback record; the human answer is the final corrected value after review.
  • Buckets are based on when the feedback was edited, not when the call happened and not when the score version was released.
  • Feedback items marked invalid are excluded before bucketing and rolling-window sampling.
  • For each bucket, the point first uses feedback edited inside that bucket. If that bucket has fewer than 100 feedback records, older feedback is added until the sample reaches 100 records or no earlier feedback exists.
  • Lookback records are only used to stabilize sparse buckets; each point still ends at that bucket's end date.
  • This is not a replay or backtest of historical champion versions. It uses the answers already stored on feedback records.

Processed 1 score across 4 complete buckets.

Overall

4 buckets100 sampled itemsLookback used

Dot size indicates the number of feedback items in each rolling sample.

How To Interpret It

  • Each chart measures how often stored production score answers agreed with the final human correction recorded in feedback.
  • The bucket date is feedback edit time, not call time and not score-version release time.
  • Feedback items marked invalid are excluded before analysis, and the report includes fetched/analyzed/ignored-invalid counts.
  • Each point first uses feedback edited inside the bucket; if the bucket has fewer than rolling_min_items records, older feedback is added until the point reaches the minimum or no earlier feedback exists.
  • Lookback records only stabilize sparse buckets; each point still ends at that bucket's end date.
  • This report does not replay historical champion versions or backtest current prompts.
  • The CLI persists scorecard-level runs as top-level timeline blocks by default: one overall timeline followed by one timeline per included score.