​
🧭 Workflow Trajectory

The Workflow Trajectory graph is a core visualization in every Mass Simulation. It merges all agent behavior across multiple Sessions into a single, clustered graph — giving you a bird’s-eye view of the decision space your agent explores.

It reveals not just what happened, but what could happen — and how often.

​
🧠 What It Shows

Each node in the Workflow Trajectory represents a grouped state the agent reached (e.g., same web page, same state string, or same screenshot).

Each edge represents a transition (an action) the agent took between states.

📷

  • Click the image in the node to see it in more detail

📷

​
🔍 Interacting with the Graph

  • Orange edges are forward transitions — the agent moved forward to a new state
  • Blue edges are backward transitions — the agent moved back to a previous state
  • Pink edges are self transitions — the agent went back to the same state
  • The multiplier shows the frequency of that transition across all sessions (i.e. how many sessions took that transition)

📷

  • Click into a node to inspect all sessions that passed through that state

📷

  • Scroll down to use the Time Travel feature to see how likely it is that your agent will make a specific action with different parameters (prompt, model, temperature, etc.).

📷

  • Click into an edge to view sessions that made that specific transition

📷

  • Zoom in to see local behavior or zoom out to see the overall flow

📷

​
🔄 How It’s Built

  1. Each Session has its own step-wise graph of states and transitions
  2. As Sessions run, they’re incrementally merged into the Mass Simulation’s Workflow Trajectory
  3. Steps from different sessions are clustered if they match by state
  4. Transitions between these clustered states are recorded with frequency counts
Session 1: A → B → C
Session 2: A → B → D

Workflow Trajectory:
A
├── B
│   ├── C (50%)
│   └── D (50%)

​
🎯 Why It Matters

This graph helps you answer critical questions:

  • What are the most common trajectories the agent follows?
  • Where does the agent branch or act inconsistently?
  • Are there failure-prone paths that show up repeatedly?
  • Do sessions loop, revisit states, or get stuck?

By visualizing all possible paths, you gain a statistical view of your agent’s decision-making.

​
🔢 Probabilistic Insights

You can think of this graph as a probabilistic model of your agent’s behavior:

  • It reveals which transitions are dominant and which are rare
  • It helps you estimate failure rates in certain trajectories
  • It surfaces unexpected branches and edge cases