diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 2fbd2c325..5adbf5063 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -6,7 +6,6 @@ on:
   push:
     branches: [ main ]
   pull_request:
-    branches: [ main ]
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
diff --git a/.github/workflows/smoke-test.yml b/.github/workflows/smoke-test.yml
index ecccce0e4..ea9ddac5d 100644
--- a/.github/workflows/smoke-test.yml
+++ b/.github/workflows/smoke-test.yml
@@ -6,7 +6,6 @@ on:
   push:
     branches: [ "main" ]
   pull_request:
-    branches: [ "main" ]
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
diff --git a/docs/fork-ci-review-bases.md b/docs/fork-ci-review-bases.md
new file mode 100644
index 000000000..af27f1125
--- /dev/null
+++ b/docs/fork-ci-review-bases.md
@@ -0,0 +1,72 @@
+# Fork CI: review bases
+
+This page is fork-local (`Flopsstuff/uemcp`). It describes how to get real CI
+signal on a pull request whose base branch is **not** `main`, which is the
+normal shape of work here: a change is reviewed against a verbatim snapshot of
+`upstream/dev` before it is forwarded upstream.
+
+## Why patching `main` is not enough
+
+`ci.yml` and `smoke-test.yml` upstream declare:
+
+```yaml
+on:
+  push:
+    branches: [ main ]
+  pull_request:
+    branches: [ main ]
+```
+
+With that `pull_request` filter, a PR into `base/upstream-dev-*` runs only the
+unconditional metadata checks (Labeler, Greetings, Validate PR Title) - no lint,
+no build, no tests. A grey check list is easy to misread as "nothing to run".
+
+Removing the filter on the fork's default branch alone changes nothing, because
+GitHub Actions resolves a workflow definition for `pull_request` from **the tree
+of the pull request itself** (its head, or the generated merge commit) - never
+from the default branch. Only `schedule`, `workflow_dispatch`, `workflow_run`
+and `repository_dispatch` read their definition from the default branch.
+
+So the trigger patch has to be present in the branch under test. It is two
+deletions:
+
+```yaml
+on:
+  push:
+    branches: [ main ]     # unchanged
+  pull_request:            # no branches filter - any base
+```
+
+## Cutting a new review base
+
+1. `git fetch upstream`
+2. Push the snapshot commit as the base branch:
+   `git push origin <upstream/dev sha>:refs/heads/base/upstream-dev-floNNN`
+3. Commit the trigger patch **on top of that base branch** and push it.
+4. Cut the feature branch from the already-patched base, then open the PR into
+   it.
+
+The patch commit is then present on both sides of the comparison, so
+`base..head` stays a clean review diff, while the workflow definitions in the
+PR tree still have no `branches` filter and CI starts.
+
+If the base was pushed as a bare snapshot and the patch went onto the feature
+branch instead, CI still runs (the PR tree is what matters), but the patch shows
+up as `-2` lines in the review diff. Prefer the base-plus-patch shape for
+anything a human has to read.
+
+## Do not forward the patch upstream
+
+When a fork change is forwarded to `ChiR24/Unreal_mcp`, cherry-pick only the
+substantive commits. The trigger patch exists because we review against
+non-`main` bases; upstream does not, and the filter is deliberate there.
+
+## Dependency graph is a repository setting
+
+`dependency-review` on the fork used to fail unconditionally because the
+dependency graph was disabled, not because of anything in
+`.github/workflows/dependency-review.yml`. It was enabled via the repository's
+vulnerability-alerts / dependency-graph setting (`PUT
+/repos/Flopsstuff/uemcp/vulnerability-alerts`). There is nothing to look for in
+the workflow files - if that check goes unconditionally red again, check the
+repository settings first.