Copyright © 2026 Saccadic LLC · Patent pending.

Browse chapters
  1. Release NotesChanges by version
  2. Chapter 1What IRIS Is and Is Not
  3. Chapter 2The IRIS Data Model
  4. Chapter 3How IRIS Measures
  5. Chapter 4Safe Interpretation and the First 15 Minutes
  6. Chapter 5Dashboard Overview
  7. Chapter 6Installation
  8. Chapter 7Configuration
  9. Chapter 8Build the Model Spec and Probe Set
  10. Chapter 9Minimal Tracker Integration
  11. Chapter 10Launch the Bench and Verify Capture
  12. Chapter 11Runtime Control, Signals, Metrics, and Logs
  13. Chapter 12Integration Patterns by Model Family
  14. Chapter 13Distributed Training
  15. Chapter 14Artifacts, Reports, Export, and Security
  16. Chapter 15Troubleshooting
  17. Chapter 16Whole Model and Layer Structure
  18. Chapter 17Domain Training and Affinity
  19. Chapter 18Attention-Head Diagnostics
  20. Chapter 19MoE Expert and Router Diagnostics
  21. Chapter 20Facets and Emergent Pathways
  22. Chapter 21Gradient Residuals, Stability, Curvature, and Update Efficiency
  23. Chapter 22Adaptive Cadence and Event Detection
  24. Chapter 23CLI Reference
  25. Chapter 24Configuration-Key Reference
  26. Chapter 25API Reference
  27. Chapter 26File and Schema Reference
  28. Chapter 27Glossary and Concept Index
  29. Chapter 28The Flow Map
  30. Appendix ADiagnostic Availability Matrix
  31. Appendix BLLM Prompt Library
On this page

Chapter 10 — Launch the Bench and Verify Capture

Bench — the live dashboard

From a run directory:

cd /path/to/your/run-dir
iris

Or point to it explicitly:

iris /path/to/your/run-dir

The launcher prints the reachable address, verifies startup, and normally backgrounds the Bench so your terminal can return. A failure stays visible instead of reporting a successful background launch. iris-bench.log in the separate Bench store records the address and server messages.

Launch options

  • --no-open or --no-browser suppresses automatic browser opening.
  • --port N selects a preferred port; check the actual address printed.
  • -f or --foreground keeps the process attached to the terminal.
  • --replay loops a recorded run over a virtual clock for demonstration.

The Bench can attach to a live run, open a completed bundle, or show a Navigator when a directory has no run artifacts. When several bundles are present without a live writer, the startup banner identifies the selected one. Check that identity before interpreting charts.

Find or stop an existing Bench

iris open
iris stop /path/to/your/run-dir

iris open opens tabs for registered Benches on the machine. iris stop stops the Bench associated with a directory; it does not stop the training job. The browser dashboard's restart control similarly restarts the Bench, not training. A directory's index.html can redirect to its running Bench.

Verify capture before interpretation

Check the run name, model identity, captured-step range, probe domains, and frame count. Read unavailable-channel messages and sampling gaps. Pending analysis is different from a channel the run never recorded.

Use Export as PDF to generate a report from the captured bundle. Read and review an export before sharing it: it can contain run labels, probe content or references, measurements, notes, and trainer-supplied signals. The Bench's AI export is intended for evidence-aware discussion, not an automatic verdict about what the model learned.

Automatic shutdown

After a Bench has been used, it normally stops about a minute after the last viewer or other keeper is gone. A browser viewer, MCP connection, enabled MCP switch, or --keep-alive keeps it running. --no-open only suppresses browser opening. A Bench nobody has opened yet does not time out. Restart with iris when needed; the captured bundle remains intact.

Remote access

The Bench binds network interfaces and has no login. A printed address is not an access-control guarantee. Use a trusted network and an appropriate encrypted access path. Tailscale names are supported; follow Artifacts, Reports, Export, and Security before exposing a Bench outside your own machine.

For control actions, open localhost on the Bench machine, or forward the port through that machine's own SSH server as your login account:

ssh -L 8765:localhost:8765 you@your-machine

Then open http://localhost:8765/, substituting the actual Bench port. Tailscale SSH forwarding is viewing-only; use the machine's own SSH server where supported. Opening the shareable machine hostname, even from that machine, also gives viewing-only access. Follow the Bench's Viewing only message for the supported reconnection method.