Ticket 03: one rulebook for an interface, and taste is retired

[Tradebaas]

What changed and why

Ticket 03 of spec 011. The framework carried two anti-slop rulebooks for the same job: taste,
adapted in-house under decision 0012, and the design method begin now installs. One fact, one
place, so taste is gone.

Nothing was deleted before it was accounted for. All thirteen of its sections were read against
impeccable's craft floor, its new-work flow and the 59 rules in its detector, and where each one
ended up is a table in decision 0020. That record supersedes 0012 without deleting it: 0012
stays readable as the record of why an in-house rulebook was right while nothing better was
installed.

Two rules genuinely had no home in the method, so they moved into Groundwork files rather than
disappearing:

• The reduced-motion obligation is now DESIGN.md principle 7. Impeccable's references never
  state it (only its own tooling honors it), and the method builds on top of these principles and
  may not break them, so principle 7 is where it binds every direction.
• The live-surface contract on a redesign is now a fourth handover bullet in design step 3:
  URL structure and slugs, titles and structured data, nav labels, form field and analytics names,
  the wordmark, legal and consent copy. Preserve-or-overhaul itself is covered, by the method's own
  four-way read of what is already true; the SEO and migration half was not covered at all.

The three dials are dropped, not moved. A page configured by numeric knobs the model sets
before the owner sees anything is exactly the unwatched stretch this spec closes. The method
decides the same things in the open and further down: a committed direction contract, a color
strategy at page scale, one authored motion. DESIGN.md loses the dials row with them.

The denylist takes three entries (the dials notation, the upstream pointer, the skill pointer and
its path). Only the last needs an exclude, for the spec that records the retirement and says the
words on purpose.

design-guard's pointer is repointed and no more than that. What it should stop restating is
ticket 04's job, and folding that in here would blur two reviews into one.

Verified by

• Every gate and all ten self-test suites, on this head. check.mjs OK (enforcement line
  reports four signals), links gate OK, and 182 gate self-tests across five suites (70 runner and
  document, 35 code, 24 config, 48 trace, 5 stack) plus progress 25, links 19, cockpit-path 11,
  cockpit 21, drill 11. All pass, none skipped.
• The explainer was rendered, not reasoned about, and it caught a real defect. Removing the
  skill card left the foundation group with three cards, which tripped a .gcards.g3 rule that had
  never been exercised: three 233px columns beside 358px cards in every other group. Measured over
  CDP at a real 1400px viewport, computed grid columns and card widths per group. Dropping g3
  (the CSS rule and the JS branch that set it) puts the group back on the standard two-column grid
  at 358px, and the screenshot confirms the rhythm now matches the rest of the page.
• The two derived stats moved by themselves. explainer-stats failed on the decision count
  until the page said 20, which is the gate doing its job; skills went 21 to 20 the same way.
• The denylist entries were proven by the run, not by reading. check.mjs scans this repo's
  own text, so a pattern that over-matched a surviving file would have failed the run. It did not.
• Not verified here: nothing in this ticket runs at a user's machine, so there is no runtime
  path to exercise beyond the page. The seam this change points at (the method actually building a
  screen) was exercised end to end on ticket 02 and is unchanged by this diff.