flake-update-20260505

Report Workflow

Generate an interactive HTML report from the LLM analysis.

Steps

1. Load Analysis

Read srvkp-backlog-llm-analysis.json and the original backlog data for metadata enrichment.

2. Generate Report

Run the report generator:

python3 gen-report-llm.py

Or, if the script doesn’t exist, generate the HTML directly. The report should include:

  • Total issues analyzed
  • Model used and confidence distribution
  • Recommendation summary cards (CLOSE, REVIEW_TO_CLOSE, NEEDS_TRIAGE, KEEP, HIGH_PRIORITY)

Filters

  • Filter by recommendation
  • Search by text (key, summary, reason)
  • Filter by component, type, priority

Component Sections

Grouped by Jira component, sorted by cleanup potential (most closable first).

Each section is collapsible and shows:

  • Component name with recommendation badge counts
  • Auto-expanded if it contains CLOSE or HIGH_PRIORITY items

Per-Issue Cards

Each issue displays:

  • Key (linked to Jira)
  • Type, Priority, Score badges
  • Confidence indicator (high/medium/low)
  • Recommendation badge
  • LLM Reason — the full explanation from the analysis (this is the key differentiator from heuristic reports)
  • Upstream Evidence — linked to GitHub PR/issue/commit
  • Suggested Comment — what to post before closing
  • Tags — semantic tags from analysis
  • Age and dates

Styling

  • Dark theme (GitHub dark style)
  • Color-coded issue borders: red=CLOSE, orange=REVIEW, yellow=TRIAGE, green=KEEP, purple=HIGH
  • Responsive layout

3. Open Report

xdg-open srvkp-backlog-llm-report.html

4. Summary

Print stats:

  • Report file path and size
  • Recommendation distribution
  • Top components by cleanup potential