Chart the probes and each model's passes by conservation law on the site

[chrimerss]

Adds two charts to the site.

Probes: a pie of the 35 merged probes by law: mass 23, energy 7, momentum 5. Counts sit on the slices, and the legend gives percentages that sum to 100.

Models: three compact panels, one per law, with the models as rows in the same order as the models table.

• Each panel has its own scale: its full width is every probe under that law.
• The gray bar is how many of those probes could score the model. The fill, blue for mass, orange for energy and green for momentum, is how many it passed.
• N/A probes count in neither number, as in the table, so each row adds up to the table's N / M. A dash means no probe under that law could score the model.

Each cell has a hover and keyboard-focus tooltip (for example, summa: 1 of 4 energy probes passed, 3 N/A). Escape dismisses it. "Show the numbers" opens the per-law counts as a table.

Both charts work in en/es/zh, in light and dark mode, and at phone width. On a phone the per-cell numbers move to the tooltip and the table, and model names wrap at underscores. The law colors are checked for color-vision deficiency on both surfaces. Green #008300 failed against the orange for protanopia, so the green is #22A06B in light mode and #1BAF7A in dark.

Keeping it current

The page stays one self-contained file. The chart data is a JSON block in site/index.html, written by a new stdlib-only scripts/site_standings.py from models/result.csv.

• It uses the same standing rule as _archived_standings: the newest archived version per model, with N/A excluded. Each probe's law is read from its probe.yaml.
• After archiving rows or merging a probe, run python3 scripts/site_standings.py. --check exits 1 when the block is stale.
• tests/test_docs_in_sync.py::test_site_charts_are_drawn_from_the_archive fails when the block is stale, when a law is miscounted, when a model is missing, or when a model's per-law sums differ from the standing the tables are checked against.
• The AGENTS.md merge checklist says when to run the script.

The probe workflow now also runs on pull requests that touch site/, scripts/ or tests/, so a stale chart or table is caught before the pages workflow deploys it. Before this, such PRs skipped CI.

Checked

• pytest -q: 1279 passed.
• The new test fails on a hand-edited count.
• Rendered in the browser in all three languages, in both themes, at 375 px with no horizontal scroll, with the tooltip on pointer and focus.
• An independent review pass found six issues. All are fixed:
  • each probe's law is read from probe.yaml;
  • the CI paths are extended;
  • the test now catches a dropped model;
  • the caption element is valid HTML;
  • the tooltip stays open while focused and closes on Escape;
  • the pie percentages sum to 100.

[Copilot]

Copilot review overview

🟡 Changes recommended

The light-theme mass-slice label does not meet minimum text contrast requirements.

Review effort: Balanced
Findings: 1

Open (1)

• Choose the text color that meets WCAG AA contrast · New

What changed in this PR

Adds localized, responsive charts showing probe distribution and model performance by conservation law.

Changes:

• Adds accessible pie and bar charts with generated standings data.
• Introduces archive-driven chart generation and consistency tests.
• Expands CI paths and documents chart refresh steps.

File	Description
site/​index.html	Adds chart UI, translations, data, and rendering logic.
scripts/​site_standings.py	Generates chart data from probes and archived results.
tests/​test_docs_in_sync.py	Verifies chart data against the archive.
.github/​workflows/​probe-pr.yml	Runs validation for site, script, and test changes.
AGENTS.md	Documents chart refresh requirements.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.