pipeline

name	pipeline
description	Check analysis pipeline state for an experiment and suggest next step

Pipeline State Check

Read docs/PIPELINE.md for the full pipeline reference.

Step 1: Identify the Experiment

Ask the user which experiment to check, or list all available sessions:

GET /api/probes

Match by sentence_set_name field in session metadata (or session_name / target_word as fallback).

If multiple sessions match, present the list with dates and ask the user to pick one.

Step 2: Find the Probe Guide

From the session's sentence_set_name, find the probe guide:

glob data/sentence_sets/**/{sentence_set_name}.md

Read it — it contains classification rules and analysis focus for this experiment.

Step 3: Determine Pipeline Stage

Run these checks in order:

1. No session found → Stage 1 (design experiment with /probe)
2. Session state != 'completed' → Stage 2 (capture in progress or failed)
3. GET /api/probes/sessions/{id}/generated-outputs — check output_category field:
   • If null/empty on most probes → Stage 3 (categorize outputs with /categorize)
4. GET /api/probes/sessions/{id}/clusterings — list schemas:
   • If empty → USER GATE (user needs to explore clustering in UI)
5. Load each schema — check for reports:
   • If no reports → Stage 5 (analysis — protocol TBD)
   • If reports exist → Stage 6 (present reports) or temporal gate

Step 4: Report & Suggest

Tell the user:

• Which experiment and session you found
• Current pipeline stage
• What the next action is
• Ask if they want to proceed

If user confirms, execute the next stage following docs/PIPELINE.md instructions.