The data: a rater panel | The problem: the label tracks the study, not the raters | What grassr reports instead | Case studies | Three labels, one position | Two panels the fixed bands call identical | A divergent panel and the per-rater route | Two raters | Layered access | Building blocks | Reference resolution and the clamp guard | The machinery | Extending the calibration | References

Scope of this companion | 1. Notation glossary | 2. Recovering $\hat q$ from an observed coefficient | 3. ICC and the subject-prevalence distribution | 4. Coefficient recovery: the full diagnostic | 5. $\hat\delta$ thresholds and operating characteristics | 5b. Pairwise PABAK and pooled-reference recovery | 6. Worked code examples per Report Card branch | Aligned case (paper §4.1) | Divergent case (paper §4.2) | $k = 2$ counterexample (paper §4.3) | 7. Audit pass-rate detail | 8. Confidence-interval methods detail | References

Abstract | What's new in 0.6.0 | 1. The headline call | 2. A symmetric panel (the easy case) | 3. A divergent panel | Pairwise reliability under the divergent flag | 4. Layered access | 5. Granular building blocks | 6. Two-rater binary case (k = 2) | 7. Surface-envelope clamp (an edge case worth knowing about) | 8. Prospective design: plot_surface() | References