Static previews of the interactive dashboards in ../docs/, captured
during the experiment. Open the live HTML for the per-axis charts, the worktree
selector, and hover detail.
| Preview | Live dashboard | What it shows |
|---|---|---|
cross-language-results.png |
docs/cross-language-results.html |
The cross-language result: 18/18 JS/TS + Go arms beat their human baseline, with the one-shot-vs-iterate split and the full per-suite matrix. |
python-results.png |
docs/python-results.html |
The Python decomposition: coverage-driven 2/9 → ablation 8/9 → quality 9/9, isolating the prompt effect from the model effect. |
kotlin-swift-results.png |
docs/kotlin-swift-results.html |
The Kotlin/Swift matrix: six libraries × three policies, all 18 arms beating their maintainer-written baselines. |
Regenerate the base readout from any scorecard JSON with the bundled
results-dashboard skill:
python .claude/skills/results-dashboard/scripts/build_dashboard.py \
results-cross-language-scorecard.json -o docs/cross-language-results.html