Skip to content
AutoResearch
StaleKeptMedium bandPlatform

Feedback reason taxonomy for denied and corrected cards

Baseline
49%
Final
81%
Delta
+32 pts
Variants
4
Objective

What we set out to improve

Map free-text denial and correction feedback into a closed reason taxonomy the loop can learn from, without surfacing the raw feedback text.

KeptPromoted to a templateWrote to a KB

Kept. An eight-class taxonomy with a confidence gate classified denial and correction reasons at 0.81 accuracy, turning free-text feedback into structured, learnable signal. Only the closed-enum reason classes are stored; raw feedback text is never retained.

Iterations

Variants we tried

Each variant and its coarse objective metric. The kept variant is marked; bars are relative to the best run.

  • 1Baseline — uncategorized free-text reasonsLow49%
  • 2Variant A — six-class taxonomyMedium68%
  • 3Variant B — eight-class taxonomyMedium76%
  • 4Variant C — eight-class + confidence gateWinnerMedium81%
Run

Stages

  1. baseline

    Succeeded · 2.4s

  2. variant run

    Succeeded · 8.8s

  3. eval

    Succeeded · 1.2s

  4. promote

    Succeeded · 300ms

Output

Artifacts and what shipped

Redaction-safe artifact previews, diffs, metric tables, and prompt variants with sensitive text removed.

  • Metric table

    Reason-classification accuracy by variant (0.49 → 0.81)

  • Report

    Taxonomy coverage report (closed-enum labels only)

  • KB write

    Stored the reason taxonomy in the platform learning KB

What you can see, and what is hidden

Every projection on this page is redaction-safe by construction. Redaction level: Sample content, curated, public-safe excerpts only.

Shown

  • Identifiers & counts
  • Closed-enum statuses
  • Coarse quality / resource bands
  • Timestamps & freshness

Intentionally hidden

  • Raw prompts
  • Raw documents
  • raw tool log
  • Raw trace spans
  • Embedding vectors
  • Free-text feedback
  • Auth internals & secrets
  • Secrets

Related in the Lab

All experimentsThe full gallery, grouped by product area.Family Manager observatoryThe capture → approve → route loop.News DigestRanked items, source health, and feedback.Knowledge Base observatoryWhere promoted learnings are written.