Fetch dataset at runtime instead of build time (F4 refactor)

Dockerfile

@@ -5,23 +5,16 @@ WORKDIR /app
 # Copy space app files
 COPY . /app
 
-# Install dependencies
+# Install dependencies (huggingface-hub is in requirements.txt for runtime dataset fetch)
 RUN pip install --no-cache-dir -r requirements.txt
 
-# Set environment for dataset download
-ENV HF_HUB_DOWNLOAD_TIMEOUT=120
-
-# Fetch benchmark tasks from Hugging Face dataset (public, no token needed)
-RUN pip install --no-cache-dir huggingface-hub && \
-    python -c "from huggingface_hub import snapshot_download; snapshot_download('hummbl-hf/governance-bench', repo_type='dataset', local_dir='/bench')"
-
-# Verify download
-RUN ls /bench/tasks/ | head -5
-
 # Set environment
+ENV HF_HUB_DOWNLOAD_TIMEOUT=120
 ENV BENCH_ROOT=/bench
 ENV PYTHONPATH=/bench/scripts:/app
 
+# Dataset is fetched at runtime (app startup), not build time,
+# so dataset updates surface without a space rebuild (F4 refactor).
 EXPOSE 7860
 
 CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]

app.py

@@ -34,6 +34,23 @@ SCRIPTS_DIR = BENCH_ROOT / "scripts"
 TEMPLATES_DIR = Path(__file__).resolve().parent / "templates"
 STATIC_DIR = Path(__file__).resolve().parent / "static"
 
+
+def ensure_bench_downloaded() -> None:
+    """Download the benchmark dataset at runtime if BENCH_ROOT is empty/missing.
+
+    Fetches from hummbl-hf/governance-bench so dataset updates surface without
+    a space rebuild (F4 refactor — was previously baked in at Docker build time).
+    """
+    if TASKS_ROOT.exists() and any(TASKS_ROOT.iterdir()):
+        return  # already populated
+    from huggingface_hub import snapshot_download
+    snapshot_download(
+        "hummbl-hf/governance-bench",
+        repo_type="dataset",
+        local_dir=str(BENCH_ROOT),
+    )
+
+
 app = FastAPI(
     title="Governance-Bench Agent Reasoning Audit",
     description="Interactive benchmark for AI agent governance primitives",
@@ -44,6 +61,11 @@ app.mount("/static", StaticFiles(directory=str(STATIC_DIR)), name="static")
 templates = Jinja2Templates(directory=str(TEMPLATES_DIR))
 
 
+@app.on_event("startup")
+async def _startup_download_bench() -> None:
+    ensure_bench_downloaded()
+
+
 # ---------------------------------------------------------------------------
 # Data models
 # ---------------------------------------------------------------------------

requirements.txt

@@ -2,3 +2,4 @@ fastapi
 uvicorn[standard]
 jinja2
 hummbl-governance>=1.1.0
+huggingface-hub