Clamp final task scores to open interval

README.md

@@ -35,7 +35,7 @@ This makes the benchmark useful for evaluating agent planning, policy following,
 | Actions | `approve`, `flag_remove`, `escalate`, `skip` |
 | Episode End | Queue empty or 15 steps reached |
 | Policy | Configurable JSON policy mapping with ordered priority tiers |
-| Output Score | Normalized final score in `[0.0, 1.0]` |
+| Output Score | Normalized final score in `(0.0, 1.0)` |
 
 ## Tasks
 
@@ -79,6 +79,8 @@ Final normalized score is computed as:
 - 30% partial-order tier compliance
 - 20% efficiency for resolving all tier-1 items within the first 40% of the step budget
 
+To satisfy the benchmark validator, the final published task score is clamped to the open interval `(0, 1)` using a small epsilon. Exact `0.0` and `1.0` are not emitted.
+
 ## Baselines
 
 Implemented baselines in [`ethicsguard/baselines.py`](C:/Users/GODREIGN/Desktop/scalerrrr/Scaler-hack/ethicsguard/baselines.py):
@@ -101,17 +103,17 @@ Measured results from the current runtime:
 | --- | --- | ---: | ---: | ---: | ---: |
 | Easy | random | 0.2629 | 0.1327 | 0.0158 | 0.6243 |
 | Easy | greedy_by_hint | 0.5743 | 0.1804 | 0.1548 | 0.9375 |
-| Easy | rule_based | 1.0000 | 0.0000 | 1.0000 | 1.0000 |
+| Easy | rule_based | 0.9999 | 0.0000 | 0.9999 | 0.9999 |
 | Easy | always_escalate | 0.2882 | 0.1496 | 0.0000 | 0.5821 |
 | Easy | always_approve | 0.3307 | 0.1808 | 0.0375 | 0.6537 |
 | Medium | random | 0.2947 | 0.1172 | 0.1065 | 0.7667 |
 | Medium | greedy_by_hint | 0.4538 | 0.1403 | 0.2224 | 0.9000 |
-| Medium | rule_based | 0.9960 | 0.0280 | 0.8000 | 1.0000 |
+| Medium | rule_based | 0.9959 | 0.0280 | 0.8000 | 0.9999 |
 | Medium | always_escalate | 0.3104 | 0.1453 | 0.0097 | 0.6625 |
 | Medium | always_approve | 0.2894 | 0.1205 | 0.1097 | 0.6532 |
 | Hard | random | 0.2778 | 0.1113 | 0.0187 | 0.7361 |
 | Hard | greedy_by_hint | 0.3905 | 0.1245 | 0.1583 | 0.9328 |
-| Hard | rule_based | 0.9887 | 0.0462 | 0.8000 | 1.0000 |
+| Hard | rule_based | 0.9886 | 0.0462 | 0.8000 | 0.9999 |
 | Hard | always_escalate | 0.2665 | 0.1097 | 0.0805 | 0.6617 |
 | Hard | always_approve | 0.2594 | 0.1102 | 0.0477 | 0.6888 |
 
@@ -247,7 +249,7 @@ The following submission gates have been exercised successfully:
 - `docker build` succeeds
 - `inference.py` runs and produces structured logs
 - all local tests pass
-- 3 tasks are exposed and scored
+- 3 tasks are exposed and scored strictly inside `(0, 1)`
 
 ## Comparison
 

ethicsguard/grader.py

@@ -5,6 +5,9 @@ import math
 from ethicsguard.models import EpisodeHistory, QueueItem
 
 
+SCORE_EPSILON = 0.0001
+
+
 def _latest_non_skip_actions(history: EpisodeHistory) -> dict[str, tuple[str, int]]:
     actions: dict[str, tuple[str, int]] = {}
     for record in history.records:
@@ -70,4 +73,9 @@ def grade_episode(history: EpisodeHistory, queue_items: list[QueueItem], policy:
     accuracy = _compute_accuracy(history, queue_items)
     order_compliance = _compute_order_compliance(history, queue_items, policy)
     efficiency = _compute_efficiency(history, queue_items)
-    return round(min(max((0.5 * accuracy) + (0.3 * order_compliance) + (0.2 * efficiency), 0.0), 1.0), 4)
+    raw_score = (0.5 * accuracy) + (0.3 * order_compliance) + (0.2 * efficiency)
+    if raw_score <= 0.0:
+        return SCORE_EPSILON
+    if raw_score >= 1.0:
+        return round(1.0 - SCORE_EPSILON, 4)
+    return round(raw_score, 4)

tests/test_env.py

@@ -26,3 +26,13 @@ def test_unknown_item_returns_error_without_ending_episode() -> None:
     result = asyncio.run(env.step(EthicsGuardAction(item_id="missing", action_type="approve")))
     assert result.done is False
     assert result.last_action_error is not None
+
+
+def test_terminal_score_is_in_open_interval() -> None:
+    env = EthicsGuardEnv(difficulty="easy", seed=1000)
+    result = asyncio.run(env.reset())
+    while not result.done:
+        item_id = result.observation.remaining_queue[0].id
+        result = asyncio.run(env.step(EthicsGuardAction(item_id=item_id, action_type="approve")))
+    assert result.score is not None
+    assert 0.0 < result.score < 1.0

tests/test_grader.py

@@ -4,7 +4,7 @@ from ethicsguard.models import EpisodeHistory, StepRecord
 from ethicsguard.policy import DEFAULT_POLICY
 
 
-def test_perfect_episode_scores_one() -> None:
+def test_perfect_episode_scores_strictly_below_one() -> None:
     queue = generate_queue(seed=1000, difficulty="easy")
     ordered_queue = sorted(queue, key=lambda item: (item.priority_tier, item.id))
     history = EpisodeHistory(
@@ -23,10 +23,16 @@ def test_perfect_episode_scores_one() -> None:
             for index, item in enumerate(ordered_queue, start=1)
         ],
     )
-    assert grade_episode(history, queue, DEFAULT_POLICY) == 1.0
+    assert grade_episode(history, queue, DEFAULT_POLICY) == 0.9999
 
 
-def test_skipped_items_reduce_score() -> None:
+def test_empty_progress_scores_strictly_above_zero() -> None:
+    queue = generate_queue(seed=1000, difficulty="easy")
+    history = EpisodeHistory(total_items=len(queue), max_steps=15, records=[])
+    assert grade_episode(history, queue, DEFAULT_POLICY) == 0.0001
+
+
+def test_skipped_items_keep_score_in_open_interval() -> None:
     queue = generate_queue(seed=1000, difficulty="easy")
     history = EpisodeHistory(
         total_items=len(queue),
@@ -43,4 +49,5 @@ def test_skipped_items_reduce_score() -> None:
             )
         ],
     )
-    assert 0.0 <= grade_episode(history, queue, DEFAULT_POLICY) < 1.0
+    score = grade_episode(history, queue, DEFAULT_POLICY)
+    assert 0.0 < score < 1.0