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 26 — File and Schema Reference

Run artifacts

Artifact Purpose
*.iris Primary captured diagnostic bundle; opened by the Bench.
iris.log Tracker-side log, capture outcomes, and applied commands.
scratch/ Temporary runtime support; a clean tracker close removes its own temporary files.
iris.cmd Supported live commands staged for the tracker when requested.
Bench-derived analysis and Bench log Kept in a separate machine store; derived analysis is rebuildable from the bundle.
Run notes User-authored observations; not included in the MCP measurement stream.
Exported PDF or AI report A selected presentation of captured data, with scope and coverage to review.
Recorded playback A display recording; its frames are not additional model measurements.

Do not treat a derived cache as the primary run record. Preserve the bundle and tracker log, and keep exported or user-authored artifacts according to your team's recordkeeping policy.

Data sources and missing values

Reported training/holdout metrics originate in your training code. Probe outcomes and internal channels originate in the instrument's captures. User-defined functions derive values from existing channels. Keep those sources separate when reading an export.

A missing key, an explicit null, a zero reading, a not-applicable label, and a pending result are different states. Read the channel's definition and stated reason. Do not fill a missing measurement with zero or fabricate intermediate samples to smooth an exported series.

Provenance for comparisons

Keep run identity, step, metric name, unit, probe/domain definitions, capture mode, and any uncertainty or reduced-coverage labels with the data. A same-named channel can have a different definition in another version; preserve a definitions-differ warning instead of silently combining it.

The bundle is not a full model checkpoint, but it and its derived artifacts can contain confidential information about the model and inputs. Apply the sharing and access guidance in Chapter 14.