Campaign sources here are fixtures now, but nothing in the tree says so

[The-Running-Dev]

Campaign authoring left this repository in W74. The published campaigns Adventures fetches now
come from
SubZeroDev.Adventures.Content,
which authors them in its own src/campaigns/ and imports only this engine's /authoring
toolkit to export them.

What did not happen in that move is the other half: this repository still carries its own copies
of the same campaigns under src/engine/src/campaigns/, plus its own exported
site/public/campaigns/*.json. Nothing marks them as no longer the live content. Anyone
improving an arc here — as PR #296 did — is editing a copy nobody publishes, and finds out only
after the work is done.

The copies cannot simply be deleted. They are load-bearing for this repository's own suite:
replay-corpus.ts and the determinism, replay, observability, and locale tests all build against
them. They are fixtures now, not content, and the tree should say so.

Done when

• src/engine/src/campaigns/ states in-file that its campaigns are test fixtures, not the
  published content, and names Content as the authoring home
• site/public/campaigns/*.json is either removed, or documented as a fixture snapshot that
  is not what Adventures fetches
• A contributor opening any campaign source in this repository can tell within one screen
  that editing it does not change what ships
• The engine's suite still passes with the fixtures in place

Agent detail

• Origin: PR #296, closed
  as superseded — its five expanded Bulgaria arcs were already published from Content
  (PR Docs site information architecture: sections, anchors, and site links #8, d99a34c), verified byte-identical on normalized JSON across all five campaigns
• Divergence measured: this repo's main shares only ~45–49 of ~98 strings per arc with what
  Content publishes; the engine copies are pre-expansion
• Not in scope: moving the fixtures' content back into sync with Content. They are fixtures;
  they pin the behaviour the tests were written against, and churning them to match published
  content would defeat the point
• Also from Expand the five Bulgaria arcs: sharper content, five endings per route #296: two test edits (clients/text/client.test.ts, mcp/server.test.ts) that
  have no counterpart in Content and were not evaluated on their own merits