chore(ci): derive the source-boundary guard from the policy manifest

.github/workflows/platform-manifest.yml

@@ -130,5 +130,16 @@ jobs:
         with:
           python-version: '3.12'
 
+      # The guard reads its rules from source-policy.public.json (rendered from
+      # the canonical private manifest) and fails closed when that file is
+      # missing or incomplete. Stdlib only: no dependency install.
       - name: Check open-core source boundary
         run: python scripts/check_source_boundary.py
+
+      # Prove the guard still blocks what it blocked before the rules moved out
+      # of the script, and that it fails closed without them.
+      - name: Prove the boundary guard fails when it should
+        if: github.event_name == 'pull_request' || github.event_name == 'push'
+        run: |
+          python -m pip install --quiet 'pytest>=8.0.0'
+          python -m pytest tests/test_source_boundary.py -q

CONTRIBUTING.md

@@ -66,6 +66,11 @@ What this means for contributions to the public repos:
   proprietary system identifiers derived from real deployments (for example,
   automation content tied to a specific customer's EHR configuration). CI
   runs `scripts/check_source_boundary.py` to reject that class of content.
+  That script keeps no denylist of its own: it reads
+  [`source-policy.public.json`](source-policy.public.json), a generated file
+  rendered from OpenAdapt's canonical source-availability manifest. Do not edit
+  the generated file by hand; a rule changes at its source, and the guard fails
+  closed if the file is missing or incomplete.
 - Do not contribute deployment-derived corpora, tuned adversary parameters,
   thresholds, oracle or connector recipes, or datasets tied to real systems
   of record. Synthetic, reproducible fixtures are welcome.