Skills Quality assurance

Bot knowledge grounding audit

A fluent wrong answer is usually two failures stacked: retrieval did not surface the right source, and generation asserted anyway. Teams tune the model when the KB is empty, or expand the KB when search returns the wrong chunk. Measure retrieval and generation separately or you will fix the wrong layer.

Quality assuranceAI oversightAnalysisAny helpdeskRead-only
Installnpx rulebase-skills install cx-bot-knowledge-grounding

When to use it

Reach for this when someone says any of these — they are the phrases the skill itself triggers on:

  • why does the bot make things up
  • RAG audit
  • wrong article cited
  • retrieval quality
  • bot answers not in the KB
  • grounding failures

How it works

The method, in the order the skill runs it. The full procedure — tables, worked examples and the edge cases — is in the skill itself.

  1. Failure taxonomy

    For each bad or unsupported answer, classify root cause.

  2. Measure retrieval without generation

    Run retrieval-only evaluation on a labelled query set: Query, Expected document(s), Metrics.

  3. Measure generation given retrieval

    Fix retrieval logs, then re-run or replay: Supported, Unsupported, Contradicted.

  4. Missing KB vs wrong chunk

    Signals for missing KB: Repeated questions on the same driver with fabrication or hand-waving.

  5. Stratified sampling for grounding audit

    Do not sample uniformly. Over-weight: Regulated, fees, eligibility, timelines, Drivers with high repeat contact after bot resolution, Languages with known thin content, Cases where the bot cited something.

Related skills

Free and open source, and vendor-neutral — it reads the conversations from whichever helpdesk you already run. Browse all 149 skills · connect your helpdesk over MCP · source on GitHub

Review every conversation. Act on what it finds.

AI for customer operations, built for financial services. Specialist agents chase every issue to resolution and every stalled customer to activation.

Rulebase dashboard