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 28 — The Flow Map

What the flow map shows

The Flow Map presents measured links between neighboring layers, grouped by probe domain or subject. Supported views can use domain groups or the model’s experts as their grain. Its signal is a measured link quantity; effective pathways is an effective dimension, not a count of individually traced routes or identified circuits.

Capture and available figures

Flow Map capture is optional and off by default. Enable it in the tracker with flow_map=FlowMapConfig(enabled=True). It operates within configured time and memory budgets; a skipped or unavailable frame states its reason.

Figure View
10.1 Signal flow map by subject or domain.
10.2 Signal difference between two domains.
10.3 Pathways map.
10.4 Pathways difference.
10.5 Effective pathways strip.
10.6 What reaches the next word piece.
10.7 Effective pathways and probe loss over training.

Read each figure's own coverage and availability. Printed reports and Export for your AI also include Flow Map readings where available.

Reading the marks

Mark Meaning to retain
Link color The domain identified by the legend.
Link thickness The figure’s stated quantity and scale: signal, signal difference, effective pathways, pathways difference, or signed next-piece contribution.
Expert node area Expert traffic share under the stated coverage.
Neutral expert nodes Model experts, not assigned domains.
Uncertainty marks or intervals A restriction on how precisely the link is read.
Shared expert or dense path A separately labeled architectural component, where present.

Use the legend and hover readings rather than assigning your own meaning to color or size. A domain is a supplied label; an expert is a model component. Neither identifies a semantic circuit by itself.

Drawn, measured, and absent are different

The Links drawn control limits the displayed share of signal. A link below that drawing threshold can still have been measured. An expert not covered by the recorded subset has a different status from a measured zero. Read the caption naming the experts, grain, and coverage.

Where a figure states that a link is uncertain, preserve its interval. An unbounded or restricted reading is not an exact point value simply because the drawing has a visible midpoint.

Playback and comparisons

The view follows the Bench’s shared training-step controls. Read the measured frame labels, pause or step when inspecting values, and keep computed-frame gaps explicit. Any visual transition is a display behavior, not an additional measurement between recorded frames. Exported data must retain the actual frame and coverage it represents.

For domain differences, use the same pair of domains and the figure’s stated scale. A changed picture can reflect routing as well as internal link structure. Compare node traffic and link readings before proposing that a different computation has appeared.

Interpretation limits

A Flow Map is an instrument view, not a literal wiring photograph. It does not establish causal information flow, a circuit’s identity, or improved generalization. Use it to choose domains, experts, and training intervals for inspection, then test proposed mechanisms with independent evidence.