ECharts adapter correctness and honest visualization tests (#509, #512)

[kgdunn]

Summary

Fixes #509 and Fixes #512.

• Visualization tests assert tautologies, and test_themes leaks a theme into every later test in the worker #512 (tests only, no adapter behaviour changed): the four TestPlotlyAnnotationRendering tests asserted only "data" in result, which every Plotly figure dict satisfies. They now assert on the rendered layout["shapes"] (count, x0/x1 vs y0/y1, xref/yref to pin the axis, fillcolor) and layout["annotations"], in the style of TestPlotlyContourStyling; the incomplete-band test compares against a band-free baseline so the skip is actually pinned. test_set_theme_changes_default now restores the plotly.io.templates.default value captured before the test (via monkeypatch) instead of forcing the package default, so the theme no longer leaks into later tests in the same xdist worker. The background-colour test no longer claims all four themes are distinct; it pins all four values (pi_tufte and pi_journal deliberately share white).
• ECharts adapter: annotations land on the wrong panel, missing data fields silently plot as 0, and short colour lists silently drop marks #509 (ECharts adapter):
  • Annotations attach to a series of their own panel: single-panel and multi-panel paths share one _attach_annotations helper (attaching to the panel's first series), and a panel with annotations but no layers gets an empty, silent carrier series bound to its own axes instead of painting the previous panel via all_series[-1].
  • _paired_data and _wireframe_series use direct row[field] lookups, so a missing or typo'd field raises KeyError, matching the Plotly adapter, instead of silently plotting 0.
  • A ragged or wrongly shaped z_matrix in _heatmap_series raises ValueError naming the layer and expected (len(y_grid), len(x_grid)) shape instead of filling cells with 0.
  • The three non-strict zips are now strict=True, preceded by an explicit length check that raises a clear ValueError naming the layer and style key. Decision: types.py documents style as "extra visual properties" with nothing implying a shorter colour list should cycle, and every in-repo producer (experiments/visualization/plots/significance.py) builds colors/error_y one-per-point from the same data, so a length mismatch indicates a bug and raising (not cycling) is the behaviour that matches the spec and visualization/tools.py.
  • MarkType.area and AnnotationType.label are declared in types.py but implemented in neither adapter; both adapters now raise NotImplementedError with a message naming the member, instead of Plotly silently falling back to scatter and ECharts dropping the label.
• Callers checked: no in-repo caller uses MarkType.area or AnnotationType.label, and all producers of colors/error_y/grid-style layers (src/process_improve/visualization/charts/, src/process_improve/experiments/visualization/plots/, experiments/_tools/visualize_doe.py) build matched-length lists and rectangular matrices, so no caller adjustments were needed; tests/test_experiments_tools.py confirms.
• Housekeeping: version 1.73.1 in pyproject.toml and CITATION.cff (date-released 2026-08-29), new ## [1.73.1] - 2026-08-29 changelog section with footer links. The version is set assuming the earlier queued PRs (carrying 1.72.x/1.73.0) merge first; if they land differently this needs only a trivial rebase.

Test plan

• uv run pytest tests/test_visualization_spec.py tests/test_themes.py tests/test_echarts_multi_panel.py tests/test_visualization.py tests/test_viz_boxplot.py --no-cov -q : 240 passed
• uv run pytest tests/test_experiments_tools.py --no-cov -q : 39 passed (DOE callers of the adapters unaffected)
• New regression tests: annotation-only middle panel lands on its own panel; single vs multi panel attach identically; missing field raises KeyError; ragged heatmap raises ValueError; short colors/error_y raise ValueError; MarkType.area / AnnotationType.label raise NotImplementedError in both adapters
• uv run ruff check . and uv run ruff format --check . both pass
• uv run mypy src/process_improve: only the pre-existing mcp_server.py:39 FastMCP error

Checklist

• Version bumped in pyproject.toml (PATCH for fixes/docs/config, MINOR for new features)
• Tests added or updated where relevant
• ruff check . passes
• CHANGELOG.md updated

---

Generated by Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 95.83333% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
..._improve/visualization/adapters/echarts_adapter.py	97.61%	0 Missing and 1 partial ⚠️
...s_improve/visualization/adapters/plotly_adapter.py	83.33%	0 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!

[kgdunn]

Two notes on the red checks:

1. The typecheck failure is not this PR's: it is red on main's own latest runs. The unpinned mcp extra now resolves to mcp 2.x, where FastMCP was renamed MCPServer, so mypy fails on src/process_improve/mcp_server.py everywhere. The fix is PR MCP server: publish real per-tool JSON Schemas (#506) #528; its migrated mcp_server.py and mcp>=2.0 floor are ported verbatim into this branch so CI can go green in any merge order. The ported files disappear from this diff once MCP server: publish real per-tool JSON Schemas (#506) #528 merges.
2. The test (3.12, ubuntu-latest, false) failure was tests/test_designs_screening_optimal.py::TestDOptimalDispatch::test_budget_clamped_to_minimum_model_size (assert 3 >= 4), in DOE dispatch code this visualization diff does not touch. It passed 6 consecutive local runs, so it looks like a stochastic search flake; the push carrying the port above re-runs it. If it fails again on this head I will treat it as real and root-cause it.

---

Generated by Claude Code