Add newsroom-grade chart design rules and enforce them

[tomvaillant]

Problem

The current native slope-chart output passes mechanical checks but does not meet a strong editorial visualization bar:

• Labels are too small.
• Integer values are rendered with a useless .0.
• The chart has weak colour hierarchy and no clear visual emphasis tied to the takeaway.
• Interaction is not discoverable; a journalist can reasonably conclude that the chart has no interactivity.
• The source footer can fall outside the initial viewport.

Passing schema, contrast, and tooltip tests is necessary but not sufficient. Splash should make excellent editorial charts by default, with type-aware information-design rules that guide both routing and rendering.

@rmdms

Suggested solution

Create a shared newsroom chart-quality rule set and enforce it at three layers:

1. Suggestion/routing: score the proposed chart against the claim, data shape, density, intended reading order, and interaction value.
2. Producer conformance: apply deterministic type-specific rules for typography, number formatting, hierarchy, direct labels, hit areas, and furniture.
3. Render review: explicitly assess visual hierarchy, scanability, interaction discoverability, and whether the graphic communicates the confirmed takeaway at a glance.

Initial deterministic rules should include:

• Format integers without decimal zeros; retain decimal precision only when the source data requires it.
• Set minimum effective label sizes at each target width and reduce density or change chart type before shrinking below them.
• Establish a semantic colour hierarchy: one justified focal series or group, neutral context, and a legend/direct-label treatment that explains the emphasis.
• Make interactivity discoverable with cursor/focus treatment and a short affordance such as “Hover or focus for details”; ensure generous hit targets.
• Keep title, unit, source, periods, and essential labels visible in the initial target viewport.
• Use type-specific rules for slope charts: emphasize the claim, avoid endpoint-label collisions, preserve country identity at both ends when tracing becomes difficult, and choose a different form when density exceeds the readable limit.

Knowledge routing and enforcement

The repository already contains type-specific guidance at knowledge/references/chart/types/<type>.md. The heat-pump test selected the correct broad family, but the orchestration did not reliably load or enforce the full slope guidance.

There is a concrete routing contradiction:

• skills/chart-native/SKILL.md says there is no separate knowledge/references/chart/types/ directory.
• That directory exists and includes knowledge/references/chart/types/slope.md.
• skills/suggest-chart/SKILL.md requires the general chart-selection and design-conformance documents, but does not explicitly require reading the chosen type’s reference before emitting the spec.

Fix the stale instruction and make type-reference loading deterministic:

1. Resolve the selected chart type to exactly one canonical reference file.
2. Require the suggester to record the reference ID/version and the applicable rules in structured proposal metadata.
3. Convert deterministic rules into validators where possible; keep genuinely editorial rules in the structured render-review rubric.
4. Fail closed when a selected type has no registered reference or its required rule set cannot be loaded.
5. Keep one canonical source for each rule so inline skill summaries, producer comments, and knowledge documents cannot silently drift.

For the slope fixture, the applicable reference should have forced an explicit decision about direct endpoint labels, readable density, and the required neutral-context/accent hierarchy. If the confirmed takeaway does not justify one focal line, routing should reconsider the chart or ask to refine the editorial emphasis rather than emit an unstructured all-neutral slope.

Add golden editorial fixtures and screenshot-based regression coverage for common chart families.

Acceptance criteria

• The heat-pump slope fixture renders integer endpoints as 60, 64, etc., never 60.0.
• Labels meet a documented minimum readable size at article-web width.
• The hierarchy visibly supports the confirmed takeaway and is explainable from the spec.
• A first-time reader can discover hover/focus behavior without guessing.
• Source furniture is visible without scrolling in the initial embed viewport.
• The chosen chart type’s canonical knowledge reference is loaded and recorded on every proposal.
• Stale/missing type-reference routing fails before production.
• Tests prove that slope-specific knowledge changes routing/spec output and catches endpoint-label, density, and hierarchy violations.
• Quality rules are shared/documented and tested, not one-off patches for this fixture.