Raw Prompt

Instead of generating a single output html file, the managers want the output to be muliple pages. The overall score and P/F summary on the first page, then links for the five criteria (detection, position accuracy, etc) linking to a different page dedicated to each. Write a prompt for an ai agent that have access to this repo to make these changes.

Prompt for AI agent (multi-page report HTML)

You have access to the repo at c:\Git\afi-evaluation-tool (Windows). Implement multi-page HTML output for the evaluation report while preserving existing split JSON support and scoring consistency.

Objective

Replace the current single-file report.html dashboard with a multi-page HTML report:

• Index page (report.html):

  • Shows Total Score gauge and Pass/Fail requirements table (same logic/data as today)
  • Shows high-level report metadata (scenario, generated_at, versions)
  • Shows links/cards to each metric category page:
    • Detection
    • Position accuracy
    • Doppler
    • RCS
    • Density & homogeneity

• Category pages (one per metric category):

  • Dedicated page containing only that category’s plots/sections
  • Example filenames (stable, deterministic):
    • report_detection.html
    • report_position_accuracy.html
    • report_doppler.html
    • report_rcs.html
    • report_density_homogeneity.html
  • Each category page should link back to the index page.

Constraints / must-haves

1. Split JSON compatibility must remain in tools/report_json_to_html.py:

   • Legacy: --input report.json
   • Split: --input-meta report_meta.json and --input-dir <report_folder>
   • Reconstruct the same in-memory report = {..., "metrics": {...}} structure and reuse plotting logic.

2. Scoring consistency:

   • Must use the shared scoring module (src/report_scoring.py) for requirements + total score (do not duplicate scoring logic).

3. Robustness:

   • If a metric category JSON is missing/empty, that category page should still render with a clear “no data” note and the index page should still work.
   • Keep the existing behavior: if Doppler scatter has no valid pairs, omit it.

4. Windows-friendly paths:

   • Use pathlib.Path everywhere.

5. Backwards behavior:

   • The report generation pipeline (src/report_generator.py) should still call the HTML tool as best-effort and should now produce:
     • report.html (index)
     • the 5 category html pages (next to it)

Implementation guidance (preferred approach)

Refactor tools/report_json_to_html.py minimally:

1. Split HTML generation into two layers:

   • generate_index_html(report, title, page_links, …) -> str
   • generate_category_html(report, category_key, title, …) -> str

2. Reuse existing plotting code:

   • Extract a helper that builds figures for a single category, e.g.
     • _build_figures_detection(report) -> list[(title, div)]
     • _build_figures_position_accuracy(report) -> …
     • etc.
   • Or modify _build_figures(report) to accept an optional only_categories=set([...]) and filter sections accordingly.

3. Plotly JS handling:

   • Today the single HTML includes Plotly JS inline once.
   • For multi-page, simplest: each page can include Plotly JS inline (works but heavier).
   • Better: write one shared plotly.min.js file in the report folder and reference it from all pages (optional).
   • Pick one approach and keep it stable/deterministic.

4. CLI changes:

   • Keep existing CLI flags working.
   • Add optional flag: --multi-page (default True if managers want it) or default to multi-page but keep --single-page for backward compatibility.

5. Output naming:

   • --output should still be supported; it refers to index HTML path.
   • Category pages should be written alongside the index output with predictable suffixes.

6. Update src/report_generator.py:

   • Ensure it still invokes tools/report_json_to_html.py and that the tool now writes multiple pages. (No need for generator to know every filename.)
   • Keep best-effort behavior: evaluation shouldn’t fail if HTML generation fails.

Acceptance criteria

• Running the tool in both modes works:
  • python tools/report_json_to_html.py --input reports/<run>/report.json
  • python tools/report_json_to_html.py --input-meta reports/<run>/report_meta.json
• Outputs written:
  • Index: report.html (score + requirements + links)
  • Category pages: one per metric category
• Links between pages work when opened from disk (no server required).
• The index page shows exactly the same score/requirement statuses as before (uses shared scoring).
• Ruff/lint stays clean.

Testing requirements

• Update/add tests in tests/test_report_outputs.py (or a new test file) to validate:
  • HTML tool produces index + 5 category pages
  • Index contains “Total Score”
  • Category pages contain <html and at least one expected section title when data exists
• Keep tests lightweight (no full evaluation run required).

Files likely to modify

• tools/report_json_to_html.py (main change)
• src/report_generator.py (invoke tool as before; no heavy changes)
• tests/test_report_outputs.py
• README.md (update usage examples and explain multi-page layout)

Deliver the changes as a clean implementation with minimal duplication and stable output filenames.