results-dashboard

results-dashboard

Render a scorecard JSON into one self-contained HTML page you can open in a browser or commit under docs/. The input is the {baselines, arms} shape the benchmark's scorers emit (see references/scorecard-json-shape.md); the output is a dark-themed readout with:

• Hero stats — arms scored, arms that beat their baseline (wins > losses), and a per-language split when arms carry a lang.
• Per-axis Win / Tie / Loss chart — a stacked bar per quality axis across all arms, so you can see at a glance which axes the generated suites win or lose.
• Per-suite matrix — every arm × every axis; each cell is gen over base with ✓ win / ✗ loss / = tie / · n-a, plus the W/L/T tally and a better? badge.

When to use

• "Regenerate the results dashboard / readout for <scorecard>.json."
• "Visualize these scorecard results as a web page."
• After running score_quality.py / score_cross_language.py (or the test-quality skill's score.py --json …), to publish the numbers.

Procedure

1. Locate the input. A scorecard JSON with a top-level arms array (and usually baselines). If the user points at a results-*-scorecard.json, use it directly. If they only have a raw tests dir, run the relevant scorer first to produce the JSON, then feed it here.
2. Confirm the framing. Ask for (or infer) a --title and a one-line --subtitle. Default title is derived from the filename.
3. Generate:

   python <skill>/scripts/build_dashboard.py <scorecard>.json \
       -o docs/<name>.html --title "…" --subtitle "…"
   

4. Verify. Open the file (or screenshot it) and check the hero counts match the scorer's own summary line (arms present / beat baseline). The script prints those counts on success — they must agree with the JSON.

Notes

• Self-contained. The only external dependency is Chart.js from a CDN; the page needs network to draw the chart but renders the matrix offline.
• Shape tolerance. Arms with "present": false render as an absent row; null axis values render as ·. Axis keys are read from the first present arm, so the columns adapt to whichever axes the scorer measured.
• Don't hand-edit generated HTML. Re-run the script after the scorer changes; the committed docs/*.html dashboards in this repo are curated, richer variants of the same idea — treat this skill's output as the reproducible base.