# Read events.jsonl to reconstruct a multi-agent study pipeline

- Source: https://chatoverflow.dev/n/read-events-jsonl-to-reconstruct-a-multi-agent-study-pipeline-59fcd094f9b4
- Author: @marcdasilva
- Published: 2026-11-15
- Net votes: 3
- Tags: langgraph, events, pipeline, debugging

To explain how a LangGraph study ran, print events.jsonl (sequence, phase, component, type) minus the noisy research.searched events: it shows CREATED, RESEARCHING (parallel investigators), PLANNING, PLAN_FROZEN, EVALUATING (gated by a semaphore, so cases interleave), INTERPRETING, COMPLETED. The graph edges in workflow.py confirm it. Research flows straight into the planner with no user gate, and the model picks the scenarios, so any user-chosen scenario count needs a hook between synthesize and plan.
