Spaces:
Running
Running
File size: 7,021 Bytes
c452421 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 | """Pre-submission validation script.
Validates all requirements:
- openenv.yaml exists and is valid
- All 7 tasks are defined
- reset() / step() / state() work correctly
- Graders produce scores in [0.0, 1.0]
- Baseline is reproducible
- Native OpenEnv adapter reset/step/state works when dependencies are installed
- Typed models validate
"""
import sys
import traceback
from typing import List, Tuple
from src.environment import IncidentResponseEnv
from src.models import Action, ActionType
from src.scenarios import SCENARIOS
from src.tasks import get_all_tasks
from baseline.inference import run_all_tasks
def _check(name: str, fn) -> Tuple[bool, str]:
try:
result = fn()
return True, result or "OK"
except Exception as exc:
return False, f"FAILED: {exc}\n{traceback.format_exc()}"
def validate() -> bool:
checks: List[Tuple[str, bool, str]] = []
env = IncidentResponseEnv()
# 1. openenv.yaml exists
def check_yaml():
import yaml
with open("openenv.yaml") as f:
data = yaml.safe_load(f)
assert data["name"] == "sentinel-oversight-command"
assert len(data["tasks"]) >= 3
return f"Found {len(data['tasks'])} tasks"
try:
ok, msg = _check("openenv.yaml", check_yaml)
except ImportError:
# yaml not installed, just check file exists
import os
ok = os.path.exists("openenv.yaml")
msg = "File exists (yaml not installed for full check)"
checks.append(("openenv.yaml valid", ok, msg))
# 2. Tasks defined
def check_tasks():
tasks = get_all_tasks()
assert len(tasks) >= 7
for t in tasks:
assert t.difficulty in ("easy", "medium", "hard", "expert")
return f"{len(tasks)} tasks defined"
ok, msg = _check("Tasks", check_tasks)
checks.append(("7 tasks defined", ok, msg))
# 3. reset() for all tasks
def check_reset():
for task_id in SCENARIOS:
obs = env.reset(task_id)
assert obs.step_number == 0
assert len(obs.alerts) > 0
return "All tasks reset successfully"
ok, msg = _check("reset()", check_reset)
checks.append(("reset() works", ok, msg))
# 4. step() returns correct types
def check_step():
env.reset("severity_classification")
result = env.step(Action(
action_type=ActionType.INVESTIGATE,
target="postgres-primary",
))
assert hasattr(result, "observation")
assert hasattr(result, "reward")
assert hasattr(result, "done")
assert hasattr(result, "info")
assert -1.0 <= result.reward.value <= 1.0
return "Step returns correct StepResult"
ok, msg = _check("step()", check_step)
checks.append(("step() returns StepResult", ok, msg))
# 5. state() returns correct type
def check_state():
env.reset("severity_classification")
env.step(Action(action_type=ActionType.INVESTIGATE, target="user-service"))
state = env.state()
assert state.step_number == 1
assert state.task_id == "severity_classification"
return "State snapshot correct"
ok, msg = _check("state()", check_state)
checks.append(("state() works", ok, msg))
# 6. Graders in [0.0, 1.0]
def check_graders():
for task_id in SCENARIOS:
env.reset(task_id)
svc = SCENARIOS[task_id].available_services[0]
env.step(Action(action_type=ActionType.INVESTIGATE, target=svc))
result = env.grade()
assert 0.0 <= result.score <= 1.0, f"{task_id}: {result.score}"
return "All graders in [0.0, 1.0]"
ok, msg = _check("Graders", check_graders)
checks.append(("Graders score [0.0-1.0]", ok, msg))
# 7. Baseline reproducible
def check_baseline():
r1 = run_all_tasks(env_instance=env, mode="rules")
r2 = run_all_tasks(env_instance=env, mode="rules")
for a, b in zip(r1, r2):
assert a["score"] == b["score"], f"Non-reproducible: {a['task_id']}"
scores = [r["score"] for r in r1]
return f"Baseline scores: {[f'{s:.4f}' for s in scores]}"
ok, msg = _check("Baseline", check_baseline)
checks.append(("Baseline reproducible", ok, msg))
# 8. SENTINEL environment (if available)
def check_sentinel():
try:
from sentinel.environment import SentinelEnv
sent_env = SentinelEnv()
sent_tasks = ["basic_oversight", "fleet_monitoring_conflict", "adversarial_worker", "multi_crisis_command"]
for task_id in sent_tasks:
obs = sent_env.reset(task_id, variant_seed=0)
assert hasattr(obs, "step_number")
assert hasattr(obs, "proposed_action")
grade = sent_env.grade()
assert 0.0 <= grade.score <= 1.0
return f"SENTINEL: {len(sent_tasks)} tasks validated"
except ImportError:
return "SENTINEL not installed (optional)"
ok, msg = _check("SENTINEL", check_sentinel)
checks.append(("SENTINEL environment", ok, msg))
# 9. Native OpenEnv adapter (skips only when local OpenEnv deps are absent)
def check_native_openenv():
try:
import dotenv # noqa: F401
import openenv # noqa: F401
except ImportError as exc:
return f"Skipped locally: missing OpenEnv dependency ({exc})"
from server.openenv_native import SentinelNativeAction, SentinelNativeEnvironment
native_env = SentinelNativeEnvironment()
obs = native_env.reset(task_id="basic_oversight", seed=1)
assert obs.task_id == "basic_oversight"
assert obs.proposed_action, "Native reset did not expose a proposal"
result = native_env.step(
SentinelNativeAction(
action="APPROVE",
explanation="Validation smoke test; policy correctness is checked separately.",
)
)
assert isinstance(result.reward, float)
state = native_env.state
assert state.task_id == "basic_oversight"
assert state.step_count == 1
assert isinstance(state.latest_proposal, dict)
return "Native OpenEnv reset/step/state smoke passed"
ok, msg = _check("Native OpenEnv", check_native_openenv)
checks.append(("Native OpenEnv adapter", ok, msg))
# Print results
print("\n" + "=" * 60)
print("OpenEnv Pre-Submission Validation")
print("=" * 60)
all_pass = True
for name, passed, detail in checks:
status = "PASS" if passed else "FAIL"
print(f" [{status}] {name}: {detail[:80]}")
if not passed:
all_pass = False
print("=" * 60)
if all_pass:
print("ALL CHECKS PASSED")
else:
print("SOME CHECKS FAILED - fix before submitting")
print("=" * 60)
return all_pass
if __name__ == "__main__":
success = validate()
sys.exit(0 if success else 1)
|