Give the year slider something to aim at

[mholzi]

Closes #2344.

The issue's own number was wrong

It said roughly four years per thumb-pixel. Measured against the real range: 1950–2026 is 76 years, and on a ~300px track that is 0.25 years per pixel — an 8px thumb movement covers about two years. Out by a factor of sixteen, and corrected in a comment on the issue.

The problem is orientation, not precision. The track carried no landmarks at all, so there was no way to see where 1985 sits. The interaction was drag, read the number, drag again — with twelve seconds on the clock, in a dim room, one-handed. A mark to aim at puts the first movement close and leaves the ±1/±5 buttons doing what they were designed for.

Derived, not pinned

#2337 made the bounds move: applyYearRange() sets min/max from the playlist in play. Marks nailed to fixed percentages would drift the moment a playlist reaches past the default — and the markup shipping an empty container is the point, not an omission.

The scale is rebuilt inside applyYearRange() itself, so the bounds and the marks cannot come from two sources and disagree.

Three details that took the extra lines

The thumb never reaches the edge. A range thumb's centre travels from thumbWidth/2 to width - thumbWidth/2, so marks are laid out inside that inset. One at a true 100% would sit past the highest year the slider can actually select.

The step widens rather than letting labels collide. Past eight marks the interval grows. Measured across the spans that actually occur:

span	step	marks	spacing on a 300px track
1950–2026 (default)	10	8	~35px
1975–2026	10	5	~53px
1980–1995	10	2	~179px
1927–2026	20	5	~54px

Two digits, until they stop being unambiguous. '80 is language-neutral — no translation needed — and narrow enough that eight fit. But a span crossing a century renders '00 twice, so the short form is used only while every label is distinct; four digits take over otherwise.

Tests

tests/unit/test_year_scale_marks_2344.py, 7 cases. Four fail against the unfixed code:

FAILED test_positions_are_inset_by_half_the_thumb
FAILED test_a_century_crossing_falls_back_to_four_digits
FAILED test_applyyearrange_rebuilds_the_scale
FAILED test_the_markup_ships_an_empty_container
4 failed, 3 passed

The three that pass are the step rule reproduced independently of the source, so a change to the constants shows up as a behavioural diff rather than a silently different picture.

Bundle and CSS rebuilt, all 18 artifacts match. 2058 unit tests green.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LYnmkDxHC2drccP1XFpJN4