For editorial testers
Simple guidance for consistent feedback

Editorial UAT & Ground Truth

Help us validate answers and build a high-quality “gold” set for ongoing checks.

What “ground truth” means

Good vs. Gold

Good

  • Accurate and helpful
  • Uses correct sources
  • Minor omissions acceptable

Gold Standard

  • Clearly correct with high confidence
  • Uses the best sources available
  • Complete enough that we’d publish as-is

When to use each feedback option

Thumbs Up → Good

  • Helpful and accurate
  • Small tweaks only
  • Optional comments

Thumbs Up → Gold Standard

  • Excellent and publication-ready
  • Comments required: note what makes it gold

Thumbs Down → Relevance/Completeness

  • Missed key details or used weaker sources
  • Off-topic or too generic
  • Comments required

Thumbs Down → Hallucination

  • Claims facts not in the sources
  • Correct with a clear example
  • Comments required

Thumbs Down → Wrong Time Period

  • Uses sources outside the time the question asked for
  • Note what period was expected
  • Comments required

Thumbs Down → Other

  • Anything else notable
  • Be specific (format, style, missing brand context)
  • Comments required

How to test consistently

  1. Pick a brand first (card on homepage).
  2. Ask a question. If you want article browsing, include words like “articles”, “coverage”, or “show me”.
  3. Scan the sources or article list. Check titles, dates, and alignment with the question.
  4. Leave feedback:
    • Thumbs up → choose Good or Gold Standard.
    • Thumbs down → select the issue and add a brief comment.
  5. Examples help. Paste a short quote or title if relevant.
Tip: Gold examples are especially valuable. Aim for 25–50 per brand.

Question types (intents)

QA (direct answer)

  • Specific question with a clear answer
  • Shows a written answer with citations
  • Example: “What is first‑party data?”

Discovery (browse articles)

  • Exploring topics or coverage
  • Shows a list of articles with links
  • Example: “Show me recent articles on retail media”

Your wording can trigger the right behavior. For browsing, include words like “show me”, “articles”, “coverage”, or “find”.

Start testing