The markup check says something, or says nothing

[sepehr-safari]

Closes #382. Follow-up to #378, which made the check run in CI for the first time.

It then reported 130 warnings, which is a wall nobody reads and a real finding nobody can see.

Five were real

Declared and referenced nowhere at all, in src/main.zig or src/tests.zig:

active_npub · identity_kind_label · logout_confirming · previews_state · previews_action

Plain accessors, left behind when settings controls moved into the Zig view and took different accessors with them. The warning offers bind it, remove it, or declare it. For these the answer was remove.

The other 125 are expected

This app draws its own view in Zig and src/onboarding.native is 24 lines of join screen, so most of the model genuinely is read only by Zig. They are declared in the two view_unbound lists with a note saying why, which is what those lists are for.

Why clear them at all

Not tidiness. At 130 lines a genuine finding is indistinguishable from the noise around it, which is exactly how five dead functions sat there unnoticed. At zero, a newly added field or Msg tag is the only thing the check says.

So the thing worth checking is that it can still fail:

$ (remove one declared tag)
warning: Msg tag "toggle_bookmark" is never dispatched from markup - wire it to
an on-* event, or add it to pub const view_unbound if only Zig code sends it
  (warnings: 1)
$ (put it back)
  warnings: 0

A check silenced into permanent silence is worse than the noise it replaced.

The guard from #378 earned its keep immediately

Editing this file after emitting the contract leaves the contract stale, and a stale contract makes the typed half skip and report zero warnings for the wrong reason. I hit that mid-change. The zero above was verified against a freshly built contract, and the CI step now fails rather than believing it.