Heatmap colour ramp is invariant across `valueMode`; only labels change

Context

Three legitimate ways to answer “how hot was this cell?”:

• count — raw event count. Honest but not normalised — dense matches swamp sparse matches.
• intensity — count / maxCount. Relative within the chart.
• share — count / totalEvents. Normalised against the whole dataset (useful when the chart is cropped).

The trap: if cell fill depends on mode, then the same cell in the same dataset renders differently under share vs count, breaking pattern recognition. Consumers comparing two heatmaps side-by-side can’t tell whether a darker cell means more events or a different mode selection.

Decision

Fill is always intensity. valueMode only retitles the scale bar and tooltip rows — “3 events” vs “50% intensity” vs “12% share”. The underlying colour scale domain is [0, maxCount] computed once and shared across all modes.

Consequences

• Visual comparability is preserved across modes. A cell that reads “dark red” is the local maximum regardless of whether the user is looking at counts, shares, or intensities.
• Consumers who genuinely want a share-driven fill (e.g. for cross-chart normalisation) override with a custom colorScale keyed to share values.
• The tooltip label is the user-facing vocabulary switch; the scale bar caption follows the same rule (see metricLabel prop).
• share mode’s denominator is the filtered event total, not the raw input. Filters like teamFilter apply upstream; the denominator matches what’s actually rendered.