Experiment: Comparative Labels for Grouped Map Output

[monadic]

Summary

Large map outputs are useful but still cognitively expensive to scan when many sibling groups share generic names.

Inspired by Browse code by meaning, we should test comparative labeling: labels are generated relative to sibling groups, not independently.

Experiment Scope (Small)

• Add an opt-in comparative label mode for grouped map output (CLI first).
• Target surfaces:
  • cub-scout map list --group-by owner
  • grouped Argo surfaces where sibling groups are common (map apps/map appsets style outputs when enabled)
• Generate short labels from existing metadata only (no external model):
  • resource name tokens
  • namespace tokens
  • owner/source tokens
  • selected labels/annotations already in graph facts
• Labeling rule: highlight distinctive tokens compared to siblings (reduce generic duplicates).
• Show evidence line per label (which fields/tokens drove it).

Why This Is Useful

• Better scanability in terminal output.
• Improves trust: labels are explainable and deterministic.
• No mutation risk; read-only UX improvement.

Success Criteria

• On maintained fixture sets (Fluxy/Banko/Arnie patterns), labels are judged more specific than baseline in review.
• Deterministic output across repeated runs on same input.
• No contract regressions in existing JSON surfaces.

Out of Scope

• LLM-generated labels.
• Changing ownership detection semantics.

Notes

• Keep this as an experiment flag first; promote only after fixture-backed confidence.