Record data.retina.fm, and the theme system three surfaces now share

[Babissimo]

The guide described five surfaces and one themed one. Both are out of date.
data.retina.fm was never captured, and offworldlabs/retina-server#382 and
offworldlabs/retina-server#383 gave dash, admin and data a dark theme built on
the map's palette. Read cold today, the guide would have someone invent a sixth
palette for a surface that already has one.

What changes

• data joins the tables. Where it and dash hold identical values they share
  a column, so the tables stay the width they were.
• The pair of palettes is documented as a pair rather than as the map's
  private business — including the part that is easy to get wrong: the three
  themed surfaces do not answer the same question the same way. dash and data
  have three states and default to system, which stamps no attribute and
  lets a prefers-color-scheme block answer, so the OS preference keeps working
  when it changes mid-session. The map has two states and defaults to dark,
  never consulting the OS. What they share is the cascade rule — the half a
  surface must paint before JavaScript runs goes on the bare selector,
  data-theme buys the other — which is a separate claim from which half a
  viewer actually meets. §10 records the split rather than resolving it.
• §3 gains --bg-sunk and --accent-ink, the two tokens the sharing
  introduced.
• The component matrix keeps dash and data apart, since that is where they
  differ. data has the shared control rule, the focus treatment and the styled
  overlay §10 has been asking dash for, so those are recorded as the versions to
  copy.
• tokens.css gains a .data class (one declaration with .dash) and a
  shared [data-theme="dark"] block, and says plainly that it cannot reproduce
  the system state the live surfaces default to.
• preview.html stops presenting the pair as two surfaces. Its dash and map
  panels were tagged "the console in light" and "the console in dark", which
  reads as two different looks. They now carry a shared accent rail, a kicker
  naming which half they are, and a note in each saying which surfaces arrive
  there and how.

The top bar is recorded but not proposed for dash, which has two navigation
trees the bar cannot hold.

Corrections found while checking claims against the stylesheets

• §2 said the console sets its micro-label in mono. dash never has: its stat
  labels, table headers and nav-section titles are all the system stack, and
  data copied that. Only the two marketing sites use mono here. (The marketing
  half of the claim I did verify — both set it from their --font-mono.)
• tokens.css had the map at radius 6/3/10, against the 8/4/12 the surface
  uses and §5 already records. §1's row said 3–10px and is now 2–12px.
• The two light themes disagree on the sunk tier. dash and data set
  --bg-sunk to the canvas value #f1f5f9; the map's light theme sets #f8fafc,
  lighter than its canvas. Everything else in the pair matches token for token.
  Recorded in §10 rather than resolved, per this file's policy.

Verification

Every value in tokens.css was read back out of a browser against the live
stylesheets rather than eyeballed: all six class/attribute combinations resolve
to exactly what dashboard/src/App.css, data-explorer/app.css and
frontend/src/map-surface.css hold, preview.html renders with the pair marked
on exactly two panels, and the map is confirmed to define no --accent-ink at
all. The theming defaults were read from ThemeContext.tsx, app.js and
useMapTheme.tsx rather than inferred from the CSS cascade, which is what
separates "light is on the bare selector" from "light is the default" — they are
not the same claim, and an earlier draft of this branch conflated them.

One dependency

The buttons table now shows dash and data sharing 8px/16px padding. That is true
once offworldlabs/retina-server#386 lands; on main today data is at 7/14. If
that PR changes shape, this row needs revisiting.

🤖 Generated with Claude Code