Chart inspection: handle waterfall charts and think-cell generated charts

[tmustier]

Problem

When inspecting charts, pptx show <file> <slide> <shape> returns the raw chart data (series values, fill colors, axes). However, for waterfall charts, this data is misleading:

Example: Slide 5 of Tiber deck

Visual shows a waterfall with labels: 954, 13, 9, 17, 174, 1,166, 0, 43, 1,209

Chart data returned:

{
  "chart_type": "bar_stacked",
  "series": [
    {"index": 0, "values": [30.5, 30.5, 43.5, 52.1, 69.4, 243.2, ...], "fill": {"type": "none"}},
    {"index": 1, "values": [12.9, 8.6, 17.3, 173.8, 0.0, 43.1], "fill": {"type": "theme", "theme_color": "accent6"}}
  ]
}

Why the mismatch?

Waterfall charts in PowerPoint are faked as stacked bar charts:

• Series 0 (invisible fill) = "base offset" positioning each bar's starting point
• Series 1 (visible) = actual bar height

The displayed numbers (954, 13, etc.) are separate text boxes placed on top of the bars, not chart data.

Implications for agents

1. To "change 954 to 987", the agent must edit a text shape, not chart data
2. The chart bars are sized to match the labels visually — rarely need direct editing
3. Raw chart data is technically correct but not semantically useful

Variations to consider

• think-cell waterfall charts: May use different structure than native PowerPoint
• Excel-generated waterfall charts: Excel 2016+ has native waterfall chart type — may expose data differently
• Other chart types: Bar, line, pie charts should have straightforward data mapping

Questions

1. Should we detect waterfall patterns and warn "this is a visual waterfall, edit text labels not chart data"?
2. Should we try to correlate text labels with chart bars by position?
3. Should chart inspection include nearby text shapes that appear to be data labels?
4. How do think-cell charts differ from native PowerPoint charts in their XML structure?

Scope

This is a complex area — do not try to solve comprehensively now. Document patterns as we encounter them.