Upload folder using huggingface_hub

README.md

@@ -0,0 +1,32 @@
+---
+license: mit
+pretty_name: OpenEnv Arena null control
+tags:
+- openenv
+- openenv-arena
+- reinforcement-learning
+- control
+---
+
+# OpenEnv Arena null control
+
+This environment is a constant-reward pipeline control, not an entry. It teaches nothing on purpose.
+
+Every episode has one step. Any action ends it with reward 0.5. All rollouts in a GRPO group get the same reward, so the advantage is zero on every step. The arena's public Trackio logs show loss 0 and grad_norm 0 on every zero-variance step (1,759 of 1,759 steps across 14 runs), so the gradient is zero on every step. Whether the recipe applies weight decay is not published.
+
+The evaluated score of this run is one observation of the arena's train-and-evaluate path with no reward signal. It is not by itself a noise floor: one score cannot separate evaluation variability from changes the pipeline makes without a gradient. The prediction, registered before submission, is 2/40 with industrial 1 and office 1, the score of an earlier run that also logged zero gradient on every step. The control counts as valid only if its Trackio log shows reward 0.5, loss 0 and grad_norm 0 on every logged step.
+
+## Interface
+
+- Task ids: `null` (one task, `train` split). See `tasks.jsonl`.
+- Action: one JSON object; `answer` is optional, any JSON value, and ignored; extra keys are accepted. A reply that fails the schema ends the episode and is scored through `finish_action`, which also returns 0.5.
+- Observation: a one-line prompt; `done` is true after the first action; `reward` is 0.5.
+
+## Reward
+
+Constant 0.5. Nothing is checked.
+
+## Image and source
+
+- Image: `ghcr.io/naklitechie/openenv-null-control:v1` (linux/amd64; the submission pins its digest).
+- Source: `source/` in this dataset (server, Dockerfile, locked requirements, replay script, tests), pinned to OpenEnv `86a180ede21e044f7929b9a7783ad83aa67d83a3`. Build from `source/` with `DOCKER_DEFAULT_PLATFORM=linux/amd64 openenv build -t <ref>`.

source/models.py

@@ -0,0 +1,20 @@
+"""Action and observation types for the null-control environment."""
+
+from typing import Any
+
+from openenv.core.env_server.types import Action, Observation
+from pydantic import ConfigDict, Field
+
+
+class NullAction(Action):
+    """Any reply ends the episode. Extra keys are accepted so no reply is a schema error."""
+
+    model_config = ConfigDict(extra="allow")
+
+    answer: Any = Field(default="", description="Ignored, any JSON value. Any action ends the episode.")
+
+
+class NullObservation(Observation):
+    """Short text prompt; reward and done come from the base Observation."""
+
+    prompt: str = Field(default="", description="Instruction shown to the policy")

source/openenv.yaml

@@ -0,0 +1,7 @@
+spec_version: 1
+name: null_control
+version: 0.1.0
+type: space
+runtime: fastapi
+app: server.app:app
+port: 8000

source/replay.py

@@ -0,0 +1,61 @@
+"""Replay example actions over /ws and assert every episode ends done=True with reward 0.5.
+
+Usage: python replay.py ws://127.0.0.1:8000/ws [log.json]
+"""
+import asyncio
+import json
+import sys
+from datetime import datetime, timezone
+
+import websockets
+
+TASK_IDS = ["null"]
+ACTIONS = [
+    {"answer": "done"},                      # example action (submission example_actions[0])
+    {},                                      # empty object
+    {"answer": "42", "tool": "calc", "x": 1},  # stray keys a policy may emit
+    {"answer": 0},                           # non-string answer
+    {"answer": None},                        # null answer
+]
+# Replies that fail the schema never reach step(): the arena ends the episode and
+# scores it through finish_action (guide, "How the model sees your environment").
+REWARD = 0.5
+
+
+async def episode(url, task_id, action):
+    async with websockets.connect(url) as ws:
+        await ws.send(json.dumps({"type": "reset", "data": {"task_id": task_id}}))
+        reset = json.loads(await ws.recv())
+        await ws.send(json.dumps({"type": "step", "data": action}))
+        step = json.loads(await ws.recv())
+        await ws.send(json.dumps({"type": "close"}))
+    return {"task_id": task_id, "action": action, "reset": reset, "step": step}
+
+
+async def main(url, log_path):
+    results = []
+    for task_id in TASK_IDS:
+        for action in ACTIONS:
+            r = await episode(url, task_id, action)
+            data = r["step"].get("data", {})
+            r["ok"] = (r["reset"].get("type") == "observation" and r["step"].get("type") == "observation"
+                       and data.get("done") is True and data.get("reward") == REWARD)
+            results.append(r)
+            print(("OK  " if r["ok"] else "FAIL"), task_id, json.dumps(action), "->", json.dumps(data))
+    # One GRPO group: 5 rollouts of the same task in parallel sessions.
+    group = await asyncio.gather(*(episode(url, TASK_IDS[0], ACTIONS[0]) for _ in range(5)))
+    rewards = [g["step"].get("data", {}).get("reward") for g in group]
+    group_ok = all(g["step"].get("data", {}).get("done") is True for g in group) and rewards == [REWARD] * 5
+    print(("OK  " if group_ok else "FAIL"), "concurrent group of 5 rewards", rewards)
+    results.append({"task_id": TASK_IDS[0], "concurrent_group_rewards": rewards, "ok": group_ok})
+    log = {"url": url, "at_utc": datetime.now(timezone.utc).isoformat(timespec="seconds"),
+           "episodes": results, "all_ok": all(r["ok"] for r in results)}
+    if log_path:
+        with open(log_path, "w") as f:
+            json.dump(log, f, indent=1)
+            f.write("\n")
+    return 0 if log["all_ok"] else 1
+
+
+if __name__ == "__main__":
+    sys.exit(asyncio.run(main(sys.argv[1], sys.argv[2] if len(sys.argv) > 2 else None)))

source/requirements.in

@@ -0,0 +1 @@
+openenv @ https://github.com/meta-pytorch/OpenEnv/archive/86a180ede21e044f7929b9a7783ad83aa67d83a3.tar.gz

source/requirements.lock

@@ -0,0 +1,358 @@
+# This file was autogenerated by uv via the following command:
+#    uv pip compile requirements.in --python-version 3.11 --python-platform x86_64-manylinux_2_28 -o requirements.lock
+aiofile==3.12.3
+    # via py-key-value-aio
+annotated-doc==0.0.5
+    # via
+    #   fastapi
+    #   typer
+annotated-types==0.8.0
+    # via pydantic
+anyio==4.15.1
+    # via
+    #   gradio
+    #   httpx
+    #   httpx2
+    #   mcp
+    #   openai
+    #   py-key-value-aio
+    #   sse-starlette
+    #   starlette
+    #   watchfiles
+attrs==26.1.0
+    # via
+    #   cyclopts
+    #   jsonschema
+    #   jsonschema-path
+    #   referencing
+authlib==1.8.0
+    # via fastmcp-slim
+backports-tarfile==1.2.0
+    # via jaraco-context
+beartype==0.22.9
+    # via py-key-value-aio
+brotli==1.2.0
+    # via gradio
+cachetools==7.2.1
+    # via py-key-value-aio
+caio==0.12.9
+    # via aiofile
+certifi==2026.7.22
+    # via
+    #   httpcore
+    #   httpx
+    #   requests
+cffi==2.1.1
+    # via cryptography
+charset-normalizer==3.5.2
+    # via requests
+click==8.5.0
+    # via
+    #   huggingface-hub
+    #   uvicorn
+cryptography==50.0.2
+    # via
+    #   authlib
+    #   joserfc
+    #   pyjwt
+    #   secretstorage
+cyclopts==5.2.0
+    # via fastmcp-slim
+dnspython==2.9.0
+    # via email-validator
+docstring-parser==0.18.0
+    # via cyclopts
+email-validator==2.3.0
+    # via pydantic
+exceptiongroup==1.3.1
+    # via fastmcp-slim
+fastapi==0.143.0
+    # via
+    #   gradio
+    #   openenv
+fastmcp==3.4.8
+    # via openenv
+fastmcp-slim==3.4.8
+    # via fastmcp
+filelock==4.1.0
+    # via huggingface-hub
+fsspec==2026.9.0
+    # via
+    #   gradio-client
+    #   huggingface-hub
+gradio==6.30.0
+    # via openenv
+gradio-client==2.7.2
+    # via
+    #   gradio
+    #   hf-gradio
+griffelib==2.3.2
+    # via fastmcp-slim
+groovy==0.1.2
+    # via gradio
+h11==0.16.0
+    # via
+    #   httpcore
+    #   httpcore2
+    #   uvicorn
+hf-gradio==0.4.1
+    # via gradio
+hf-xet==1.7.0
+    # via huggingface-hub
+httpcore==1.0.9
+    # via httpx
+httpcore2==2.13.1
+    # via httpx2
+httpx==0.28.1
+    # via
+    #   fastmcp-slim
+    #   gradio
+    #   gradio-client
+    #   mcp
+    #   openenv
+    #   safehttpx
+httpx-sse==0.4.3
+    # via mcp
+httpx2==2.13.1
+    # via
+    #   huggingface-hub
+    #   openai
+huggingface-hub==2.2.0
+    # via
+    #   gradio
+    #   gradio-client
+    #   openenv
+idna==3.20
+    # via
+    #   anyio
+    #   email-validator
+    #   httpx
+    #   httpx2
+    #   requests
+importlib-metadata==9.0.1
+    # via keyring
+jaraco-classes==3.4.0
+    # via keyring
+jaraco-context==6.1.2
+    # via keyring
+jaraco-functools==4.6.0
+    # via keyring
+jeepney==0.9.0
+    # via
+    #   keyring
+    #   secretstorage
+jinja2==3.1.6
+    # via gradio
+jiter==0.17.0
+    # via openai
+joserfc==1.7.5
+    # via
+    #   authlib
+    #   fastmcp-slim
+jsonref==1.1.0
+    # via fastmcp-slim
+jsonschema==4.26.0
+    # via mcp
+jsonschema-path==0.5.0
+    # via fastmcp-slim
+jsonschema-specifications==2025.9.1
+    # via jsonschema
+keyring==25.7.0
+    # via py-key-value-aio
+markdown-it-py==4.2.0
+    # via rich
+markupsafe==3.0.4
+    # via
+    #   gradio
+    #   jinja2
+mcp==1.30.0
+    # via fastmcp-slim
+mdurl==0.1.2
+    # via markdown-it-py
+more-itertools==11.1.0
+    # via
+    #   jaraco-classes
+    #   jaraco-functools
+numpy==2.4.6
+    # via
+    #   gradio
+    #   pandas
+openai==3.28.0
+    # via openenv
+openapi-pydantic==0.6.0
+    # via fastmcp-slim
+openenv @ https://github.com/meta-pytorch/OpenEnv/archive/86a180ede21e044f7929b9a7783ad83aa67d83a3.tar.gz
+    # via -r requirements.in
+opentelemetry-api==1.45.1
+    # via
+    #   fastapi
+    #   fastmcp-slim
+orjson==3.13.0
+    # via gradio
+packaging==26.3
+    # via
+    #   fastmcp-slim
+    #   gradio
+    #   gradio-client
+    #   huggingface-hub
+    #   openenv
+pandas==3.0.6
+    # via gradio
+pathable==0.6.0
+    # via jsonschema-path
+pillow==12.3.0
+    # via gradio
+platformdirs==4.12.4
+    # via fastmcp-slim
+py-key-value-aio==0.4.6
+    # via fastmcp-slim
+pycparser==3.11
+    # via cffi
+pydantic==2.14.0
+    # via
+    #   fastapi
+    #   fastmcp-slim
+    #   gradio
+    #   mcp
+    #   openai
+    #   openapi-pydantic
+    #   openenv
+    #   pydantic-settings
+pydantic-core==2.50.0
+    # via pydantic
+pydantic-settings==2.15.0
+    # via
+    #   fastmcp-slim
+    #   mcp
+pydub==0.25.1
+    # via gradio
+pygments==2.21.0
+    # via
+    #   rich
+    #   rich-rst
+pyjwt==2.15.1
+    # via mcp
+pyperclip==1.11.0
+    # via fastmcp-slim
+python-dateutil==2.9.0.post0
+    # via pandas
+python-dotenv==1.2.4
+    # via
+    #   fastmcp-slim
+    #   pydantic-settings
+python-multipart==0.0.32
+    # via
+    #   fastmcp-slim
+    #   gradio
+    #   mcp
+pytz==2026.5
+    # via gradio
+pyyaml==6.0.3
+    # via
+    #   fastmcp-slim
+    #   gradio
+    #   huggingface-hub
+    #   jsonschema-path
+    #   openenv
+referencing==0.37.0
+    # via
+    #   jsonschema
+    #   jsonschema-path
+    #   jsonschema-specifications
+requests==2.34.2
+    # via openenv
+rich==15.0.0
+    # via
+    #   cyclopts
+    #   fastmcp-slim
+    #   openenv
+    #   rich-rst
+    #   typer
+rich-rst==2.2.0
+    # via cyclopts
+rpds-py==2026.9.1
+    # via
+    #   jsonschema
+    #   referencing
+safehttpx==0.1.7
+    # via gradio
+secretstorage==3.5.0
+    # via keyring
+semantic-version==2.10.0
+    # via gradio
+shellingham==1.5.4
+    # via typer
+six==1.17.0
+    # via python-dateutil
+sniffio==1.3.1
+    # via openai
+sse-starlette==3.5.0
+    # via mcp
+starlette==1.7.0
+    # via
+    #   fastapi
+    #   fastmcp-slim
+    #   gradio
+    #   mcp
+    #   sse-starlette
+tomli==2.5.0
+    # via openenv
+tomli-w==1.2.0
+    # via openenv
+tomlkit==0.14.0
+    # via gradio
+tqdm==4.70.1
+    # via huggingface-hub
+truststore==0.10.4
+    # via
+    #   httpcore2
+    #   httpx2
+typer==0.27.3
+    # via
+    #   gradio
+    #   hf-gradio
+    #   openenv
+typing-extensions==4.16.0
+    # via
+    #   anyio
+    #   exceptiongroup
+    #   fastapi
+    #   fastmcp-slim
+    #   gradio
+    #   gradio-client
+    #   httpx2
+    #   huggingface-hub
+    #   mcp
+    #   openai
+    #   openenv
+    #   opentelemetry-api
+    #   py-key-value-aio
+    #   pydantic
+    #   pydantic-core
+    #   referencing
+    #   starlette
+    #   typing-inspection
+typing-inspection==0.4.4
+    # via
+    #   fastapi
+    #   mcp
+    #   pydantic
+    #   pydantic-settings
+uncalled-for==0.4.1
+    # via fastmcp-slim
+urllib3==2.8.0
+    # via requests
+uvicorn==0.54.0
+    # via
+    #   fastmcp-slim
+    #   gradio
+    #   mcp
+    #   openenv
+watchfiles==1.3.0
+    # via fastmcp-slim
+websockets==17.2
+    # via
+    #   fastmcp-slim
+    #   openenv
+zipp==4.1.1
+    # via importlib-metadata

source/server/Dockerfile

@@ -0,0 +1,18 @@
+# Null-control OpenEnv server. Build from envs/null:
+#   DOCKER_DEFAULT_PLATFORM=linux/amd64 openenv build -t <ref>
+FROM python:3.11-slim@sha256:e88e9763f943ec1834f992a4b51e0f24500486803e8bc534e5767af9ea65f6ce
+
+WORKDIR /app
+COPY requirements.lock /app/requirements.lock
+RUN pip install --no-cache-dir -r /app/requirements.lock \
+    && useradd --uid 10001 --no-create-home --shell /usr/sbin/nologin env
+
+COPY models.py /app/models.py
+COPY server /app/server
+
+LABEL org.opencontainers.image.source="https://github.com/NakliTechie/openenv-arena" \
+      org.opencontainers.image.description="OpenEnv Arena null control: constant reward 0.5, zero GRPO advantage"
+
+USER env
+EXPOSE 8000
+CMD ["uvicorn", "server.app:app", "--host", "0.0.0.0", "--port", "8000"]

source/server/app.py

@@ -0,0 +1,12 @@
+"""FastAPI app for the null-control environment (HTTP + /ws)."""
+
+from openenv.core.env_server.http_server import create_app
+
+try:
+    from ..models import NullAction, NullObservation
+    from .environment import NullEnvironment
+except ImportError:
+    from models import NullAction, NullObservation
+    from server.environment import NullEnvironment
+
+app = create_app(NullEnvironment, NullAction, NullObservation, env_name="null_control", max_concurrent_envs=16)

source/server/environment.py

@@ -0,0 +1,44 @@
+"""Null-control environment: every episode ends after one action with the same reward.
+
+Every rollout in a GRPO group gets REWARD, so within-group variance and the
+advantage are zero on every step. Purpose: measure the arena's evaluation with
+no learning signal (plan/research/design.md, Batch B).
+"""
+
+from uuid import uuid4
+
+from openenv.core.env_server.interfaces import Environment
+from openenv.core.env_server.types import State
+
+try:
+    from ..models import NullAction, NullObservation
+except ImportError:
+    from models import NullAction, NullObservation
+
+REWARD = 0.5
+TASK_IDS = ("null",)
+PROMPT = 'Reply with one JSON object: {"answer": "done"}. Any reply ends the episode.'
+
+
+class NullEnvironment(Environment):
+    SUPPORTS_CONCURRENT_SESSIONS: bool = True
+
+    def __init__(self):
+        self._state = State(episode_id=str(uuid4()), step_count=0)
+        self._done = False
+
+    def reset(self, seed=None, episode_id=None, task_id=None, **kwargs) -> NullObservation:
+        if task_id is not None and task_id not in TASK_IDS:
+            raise ValueError(f"unknown task_id {task_id!r}; known: {list(TASK_IDS)}")
+        self._state = State(episode_id=episode_id or str(uuid4()), step_count=0)
+        self._done = False
+        return NullObservation(prompt=PROMPT, done=False, reward=0.0)
+
+    def step(self, action: NullAction, timeout_s=None, **kwargs) -> NullObservation:  # type: ignore[override]
+        self._state.step_count += 1
+        self._done = True
+        return NullObservation(prompt="Episode over.", done=True, reward=REWARD)
+
+    @property
+    def state(self) -> State:
+        return self._state

source/tests/test_null.py

@@ -0,0 +1,35 @@
+"""In-process checks: every action ends the episode with the same reward. Run: python -m unittest discover -s tests"""
+import sys
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+
+from models import NullAction  # noqa: E402
+from server.environment import REWARD, NullEnvironment  # noqa: E402
+
+
+class NullEnvTest(unittest.TestCase):
+    def test_reset_is_not_terminal(self):
+        obs = NullEnvironment().reset(task_id="null")
+        self.assertFalse(obs.done)
+        self.assertLess(len(obs.prompt), 200)
+
+    def test_every_action_gives_constant_reward(self):
+        for payload in ({"answer": "done"}, {}, {"answer": "x", "tool": "calc", "n": 3}, {"answer": 0}, {"answer": None}):
+            env = NullEnvironment()
+            env.reset(task_id="null")
+            obs = env.step(NullAction.model_validate(payload))
+            self.assertTrue(obs.done)
+            self.assertEqual(obs.reward, REWARD)
+
+    def test_reward_is_half(self):
+        self.assertEqual(REWARD, 0.5)
+
+    def test_unknown_task_rejected(self):
+        with self.assertRaises(ValueError):
+            NullEnvironment().reset(task_id="other")
+
+
+if __name__ == "__main__":
+    unittest.main()

tasks.jsonl

@@ -0,0 +1 @@
+{"task_id": "null", "split": "train", "prompt": "Reply with one JSON object: {\"answer\": \"done\"}. Any reply ends the episode.", "reward": 0.5}