Batch Prediction And Scoring Prompts

Example 1: Build and score a constrained batch

User prompt

Build a prediction batch for the top 100 confident candidates from this table, then score them against the Feather spectra file.

What the MCP should do

• call build_prediction_batch
• call inspect_prediction_batch_status until complete if needed
• call score_from_prediction_batch
• call inspect_scored_batch_status until complete if needed

Expected output shape

• prediction manifest path
• scoring manifest path
• scored CSV path
• failure/status artifacts when present

Example 2: Poll a running batch

User prompt

Poll the prediction batch and tell me whether it is still running, how many items are done, and whether anything has failed.

What the MCP should do

• call inspect_prediction_batch_status

Expected output shape

• phase
• completed
• total
• succeeded
• failed
• progress_fraction

Example 3: Inspect batch failures

User prompt

Inspect the scored batch failures and group them by error type.

What the MCP should do

• call inspect_scored_batch
• summarize failure_type_counts

Expected output shape

• counts per error type
• preview rows for failures

Example 4: Render the best scored subset

User prompt

Render comparison plots for the ten best-scoring PSMs from the scored batch.

What the MCP should do

• inspect the scored batch output
• select the top-scoring rows
• prepare payloads
• call render_spectra_batch

Expected output shape

• HTML files grouped by rendering mode
• batch render status path