diff --git a/.env.example b/.env.example
index b5f12e9..4fe37a7 100644
--- a/.env.example
+++ b/.env.example
@@ -60,10 +60,11 @@ AUTH_PRELOGIN_SECRET=
 # download to reuse it rather than fetching again.
 PLENUM_DATA_DIR=
 
-# Use a parliament.yaml / prompt tree outside the repository, so a deployment's
-# own values never show up in a diff against upstream.
+# Point these outside the repository and a deployment's own values never show up
+# in a diff against upstream — config, prompts and site copy all overridable.
 # PARLIAMENT_CONFIG=/etc/plenum/parliament.yaml
 # PROMPTS_DIR=/etc/plenum/prompts
+# CONTENT_DIR=/etc/plenum/content
 
 # Re-read prompt files on every call. Development only.
 # PROMPTS_RELOAD=1
diff --git a/README.md b/README.md
index 4ae5d85..cfe0849 100644
--- a/README.md
+++ b/README.md
@@ -101,11 +101,20 @@ deploy/examples/  systemd units and an nginx site, with placeholders
 All settings are environment variables, documented in `.env.example`. Two paths let a
 deployment keep its own values outside the repository entirely:
 
-- `PARLIAMENT_CONFIG` — path to a `parliament.yaml` elsewhere on disk
-- `PROMPTS_DIR` — path to a prompt tree elsewhere on disk
+- `PARLIAMENT_CONFIG` — a `parliament.yaml` elsewhere on disk
+- `PROMPTS_DIR` — a prompt tree elsewhere on disk
+- `CONTENT_DIR` — site copy (explainer, guide) elsewhere on disk
 
 Set `PROMPTS_RELOAD=1` in development to re-read prompt files on every call.
 
+### Running a fork
+
+If you maintain a deployment as a fork, keep it differing from upstream only in
+files upstream does not have. Everything branded or private is either an env-var
+pointer (the three above), an untracked file (`.env`, `providers.yaml`), or an
+addition under `deploy/prod/`. `make check-fork-divergence` fails the build if
+anything else drifts, which turns a merge conflict into a caught mistake.
+
 ## Security
 
 The `database_query` tool executes model-authored SQL. Give it a database role with
diff --git a/backend/app.py b/backend/app.py
index 4db9a0f..4011cca 100644
--- a/backend/app.py
+++ b/backend/app.py
@@ -110,12 +110,15 @@ def _reap_abandoned_jobs() -> None:
 
 @app.get("/api/guide", response_class=PlainTextResponse)
 def get_guide() -> str:
-    """Serve user-guide.md as plain text. Single source of truth for all guide links."""
-    guide_path = os.path.join(os.path.dirname(__file__), "..", "user-guide.md")
-    if not os.path.exists(guide_path):
-        raise HTTPException(status_code=404, detail="Guide not found")
-    with open(guide_path, encoding="utf-8") as f:
-        return f.read()
+    """Serve the user guide as plain text — the single source for all guide links.
+
+    Resolved through CONTENT_DIR, so a deployment can ship its own guide without
+    editing the repository.
+    """
+    guide = PARLIAMENT.read_content("guide_file")
+    if not guide:
+        raise HTTPException(status_code=404, detail="No guide configured")
+    return guide
 
 
 @app.get("/api/meta")
diff --git a/bootstrap.py b/bootstrap.py
index c8420da..8714c24 100644
--- a/bootstrap.py
+++ b/bootstrap.py
@@ -5,8 +5,8 @@ the repository root is not on ``sys.path`` and relative paths would resolve agai
 whatever directory the caller happened to be in. Importing this module fixes both,
 deriving the root from this file's own location.
 
-The predecessor hardcoded ``/home/lasse/riksdagen`` in 22 files, which is why the
-project could only ever run on one machine.
+The predecessor hardcoded one absolute path in 22 files, which is why it could
+only ever run on a single machine.
 """
 from __future__ import annotations
 
diff --git a/content/sv/user-guide.md b/content/sv/user-guide.md
index 5640d0f..af50f8b 100644
--- a/content/sv/user-guide.md
+++ b/content/sv/user-guide.md
@@ -4,11 +4,11 @@ Den här guiden förklarar hur Riksdagen-appen är byggd, vilka val som gjorts,
 
 ---
 
-## Om rixdagen.se
+## Om den här sajten
 
-Sajten är utvecklad av [Lasse Edfast](https://lasseedfast.se), hittills utan ekonomiskt stöd. Den är gratis att använda och ingen information sparas utöver de sju dagarna som en chatt sparas i. 
+Sajten drivs utan ekonomiskt stöd. Den är gratis att använda och ingen information sparas utöver de sju dagarna som en chatt sparas i.
 
-Vill du vara med och utveckla, eller bara har frågor, kontakta gärna Lasse via [e-post](mailto:lasse@edfast.se).
+Vill du vara med och utveckla, eller bara har frågor, hör gärna av dig.
 
 ---
 
@@ -207,11 +207,11 @@ En API-nyckel är ett lösenord som identifierar dig mot en AI-leverantörs serv
 - **OpenRouter** – [openrouter.ai](https://openrouter.ai) (aggregator med tillgång till hundratals modeller)
 - **OpenAI** – [platform.openai.com](https://platform.openai.com) (samma modeller som driver ChatGPT)
 
-Användning av externa modeller kostar pengar och debiteras direkt från ditt konto hos leverantören – inte via rixdagen.se.
+Användning av externa modeller kostar pengar och debiteras direkt från ditt konto hos leverantören – inte via den här sajten.
 
 ### Hur nyckeln lagras
 
-Din nyckel sparas **enbart lokalt i din webbläsare** (`localStorage`). Den skickas till rixdagen.se:s server bara för att vidarebefordras till leverantören när ett anrop görs – den loggas inte, sparas inte i databasen och syns aldrig i någon annan användares session.
+Din nyckel sparas **enbart lokalt i din webbläsare** (`localStorage`). Den skickas till sajtens server bara för att vidarebefordras till leverantören när ett anrop görs – den loggas inte, sparas inte i databasen och syns aldrig i någon annan användares session.
 
 Konkret:
 - Nyckeln ligger kvar i webbläsaren tills du rensar den (knappen ✕ i inställningspanelen) eller tömmer webbläsarens lagringsdata.
diff --git a/docs/eval-harness.md b/docs/eval-harness.md
index e8343fc..61efec5 100644
--- a/docs/eval-harness.md
+++ b/docs/eval-harness.md
@@ -97,12 +97,12 @@ via sigmoid so 0.5 = neutral, >0.7 = likely grounded, <0.3 = likely hallucinated
 
 `NULL` means the scorer endpoint was unreachable when the judgment was recorded.
 
-Start the scorer (first run downloads the model to `/home/lasse/models`):
+Start the scorer (first run downloads the model to `$HOME/models`):
 
 ```bash
 nohup vllm serve BAAI/bge-reranker-v2-m3 \
   --port 8001 \
-  --download-dir /home/lasse/models \
+  --download-dir $HOME/models \
   --gpu-memory-utilization 0.2 \
   --max-model-len 8192 \
   --trust-remote-code \
diff --git a/parliament.py b/parliament.py
index bfae36e..cee4fd0 100644
--- a/parliament.py
+++ b/parliament.py
@@ -24,6 +24,11 @@ import yaml
 
 _ROOT = Path(__file__).resolve().parent
 
+# Site copy — the explainer, the limit warning, the user guide. Overridable so a
+# deployment's own wording lives outside the repository, the same way PROMPTS_DIR
+# works for prompts.
+CONTENT_DIR = Path(os.environ.get("CONTENT_DIR") or _ROOT / "content")
+
 # A Postgres text-search configuration name. It is interpolated into SQL rather
 # than passed as a parameter (identifiers cannot be bound), so it is validated
 # on load and never trusted from arbitrary input.
@@ -129,7 +134,7 @@ class Parliament:
         rel = self.site.get(key)
         if not rel:
             return ""
-        path = _ROOT / rel
+        path = CONTENT_DIR / rel
         return path.read_text(encoding="utf-8") if path.exists() else ""
 
     # -- serialisation ------------------------------------------------------
diff --git a/parliament.yaml b/parliament.yaml
index 2e05f7b..289e5a6 100644
--- a/parliament.yaml
+++ b/parliament.yaml
@@ -174,13 +174,18 @@ embeddings:
 
 # Served to the frontend via GET /api/meta.
 site:
-  title: rixdagen.se
+  # Deployment identity, not country data. Upstream ships neutral values; a
+  # deployment overrides them in its own PARLIAMENT_CONFIG file so its branding
+  # never appears in a diff against upstream.
+  title: plenum
   tagline: "Sök i riksdagens anföranden och motioner"
   # Markdown, so a deployment can rewrite its own copy without touching source.
-  explainer_file: content/sv/explainer.md
-  limit_warning_file: content/sv/limit_warning.md
+  # Paths are relative to CONTENT_DIR (default: ./content).
+  explainer_file: sv/explainer.md
+  limit_warning_file: sv/limit_warning.md
   # Upstream ships these empty. A deployment fills them in via its own
   # PARLIAMENT_CONFIG file, which keeps personal contact details out of the repo.
+  guide_file: sv/user-guide.md
   contact:
     email: null
     url: null