Troubleshooting Prompts

Example 1: Scoring failed

User prompt

Why did this candidate fail to score?

What the MCP should do

  • inspect the scoring failure row or batch failure table
  • identify whether the issue is spectrum lookup, missing prediction, malformed candidate metadata, or another error

Expected output shape

  • concise root-cause explanation
  • relevant failure fields

Example 2: Batch is slow or silent

User prompt

Is this batch still running, and how far along is it?

What the MCP should do

  • call inspect_prediction_batch_status or inspect_scored_batch_status

Expected output shape

  • progress summary with phase, completed, total, and failed

Example 3: Field semantics are unclear

User prompt

Explain what next_beam and next_beam_margin mean in this table.

What the MCP should do

  • explain the normalized candidate schema
  • describe how the values are derived for supported adapters

Expected output shape

  • concise field-level explanation
  • no tool call needed if the schema resource is sufficient

Example 4: Unsure what to do next

User prompt

I have a candidate table and matching spectra. What is the right sequence of MCP steps if I want to inspect first and then validate a few targets?

What the MCP should do

  • recommend the normal workflow:
  • inspect table
  • choose targets
  • score/render selected targets
  • batch only if needed

Expected output shape

  • short stepwise plan
  • optionally point to specific tools by name