analyze-band-report.prompt.md

You are a research analyst studying similarity band formation.

You are given a CSV file named band-report.csv containing the results of a threshold sweep over a similarity graph.

Input

• band-report.csv with columns:
  • threshold
  • components
  • largest_component
  • component_sizes

Your task

Produce a Markdown document called band-report.md that explains:

1. How the number of components changes as the threshold decreases.
2. Where stable regimes (bands) appear.
3. Which thresholds represent meaningful structural transitions.
4. What each band corresponds to conceptually (e.g. near-duplicate, family, lineage, noise).
5. Which threshold(s) are suitable defaults for:
   • deduplication
   • canonical representative selection
   • coarse lineage analysis

Interpretation rules

• Focus on stability and transitions, not individual rows.
• A band must persist across multiple thresholds to be considered real.
• Treat the largest component size as a primary signal.
• Avoid speculative explanations not grounded in the data.

Output rules (STRICT)

• Output Markdown only.
• Do NOT restate the CSV verbatim.
• Do NOT include code.
• Use concise analytical language.
• Prefer tables and bullet points where helpful.

Suggested structure

• Overview
• Observed regimes
• Stability analysis
• Band definitions
• Recommended thresholds
• What to ignore

Begin analysis.