Use the Titanic dataset for every README example

[gyli]

#68 rewrote the quickstart around the people aboard the Titanic, but left the six numbered examples below it on their original data: Cat1/Cat2/Cat3, invented vehicle sales figures, and Factory A/B/C. The page introduced a real dataset and then dropped it, so a reader following along had to reorient five times.

I checked whether the Titanic figures could carry all six before assuming they could. They can.

README example	what it needs	Titanic cut used
1. Value scaling	values that must be scaled to fit the grid	class and crew counts, 2,201 onto a 5x10 grid
2. Values in dict & auto-sizing	values small enough to be block counts	the 109 children aboard: 6, 24, 79
3. Colours, title, direction, arranging	3 to 5 categories worth styling	who was aboard, by class
4. Pictogram	categories with genuinely distinct icons	men, women and children
5. Existing axis	anything small	survived and lost, 710 and 1,491
6. Multiple plots from a DataFrame	a real two-dimensional table	class by group

Two were the awkward ones:

Auto-sizing needs absolute block numbers, and 2,201 is far too many. The children aboard total 109, so rows=5 gives a 22-column chart where one block really is one child, which is a better illustration of the feature than the placeholder numbers were.

The DataFrame example needs a genuine table rather than a list. The inquiry's own table is already class by group, so first/second/third class against men/women/children drops straight in. Sizing each subplot by its own class turns out to carry the point on its own: third class is visibly wider than first and second together.

On accuracy

The prose quotes the real scaled numbers. The old text said "automatically scaled to 24, 23 and 3"; the new text says 7, 6, 16 and 20, which I read off the rendered chart rather than computing by hand. I also confirmed the per-class men/women/children rows sum to the class totals used elsewhere on the page (325, 285, 706), so the two cuts of the dataset agree with each other.

The source is cited once at the end of the section rather than repeated under every chart.

Verified

• All 6 documentation tests pass, 69 subtests (up from 63), which is every fenced Python block in the README and docs executed as written.
• Full suite green, 234 tests.
• Every referenced image exists (the check CI runs).
• Spot-checked that the values in the README snippets are character-for-character the values in examples/generate_plots.py, so the code shown is the code that produced the image above it.
• Regenerating twice produces no further diff.
• Sphinx builds with zero warnings.

Not included

demo.ipynb still uses the old data. It is a separate artifact with its own outputs committed, and rebuilding it is a much larger diff, so it is better as its own change.

---

Update: example 3 reordered

The styling example used block_arranging_style='snake', and with four categories on a 5x10 grid that reads badly. Snake reverses every other row, so Second class came out as two disconnected patches — top right, then the right of row two — with Third class running underneath on the left. Scanning left to right you met gold, teal, terracotta, then teal again, which looks like the data is out of order rather than a deliberate layout.

Measured rather than judged by eye. Reading the grid top-left to bottom-right and counting colour runs, where four categories ideally give four runs:

arrangement	colour runs
NW, vertical=True, snake (before)	7
NW, vertical=True (after)	4, plus the trailing empty cell
NW, vertical=False, snake	19
NW, vertical=False, normal	21

Each category is now a single continuous run and the chart reads in legend order, the way a line of text does.

starting_location="NW" and vertical=True stay, so the example still shows a layout that differs from the default — verified the first block moves from bottom-left to top-left. block_arranging_style keeps its own paragraph and its documentation link, so it stays discoverable without having to make this particular chart confusing to earn its place. The section heading no longer promises "Arranging Style".