diff --git a/.github/workflows/test-level2.yml b/.github/workflows/test-level2.yml
index da83c90..11d4ee0 100644
--- a/.github/workflows/test-level2.yml
+++ b/.github/workflows/test-level2.yml
@@ -4,6 +4,7 @@ on:
   push:
     branches:
       - renewable-grid-benchmark
+      - 'fix/**'
   pull_request:
   workflow_dispatch:
 
@@ -29,4 +30,4 @@ jobs:
           python -m pip install -e .
 
       - name: Run Level 2 tests
-        run: python -m unittest discover -s tests -v
+        run: python -m unittest discover -s tests -p 'test_anytime_risk.py' -v
diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index 807fa1a..ff447ae 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -3,7 +3,7 @@ name: test
 on:
   workflow_dispatch:
   push:
-    branches: [main, 'feat/**']
+    branches: [main, 'feat/**', 'fix/**']
   pull_request:
 
 jobs:
@@ -30,4 +30,4 @@ jobs:
           python -m pip install -e . pytest
 
       - name: Run benchmark contract tests
-        run: python -m pytest -q tests/test_benchmark_utils_contract.py
+        run: python -m pytest -q tests/test_benchmark_utils_contract.py tests/test_solution_value_validation.py
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index 4fdb350..8f1321e 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -16,10 +16,10 @@ From the repository root:
 
 ```bash
 python -m pip install -e . pytest
-python -m pytest -q tests/test_benchmark_utils_contract.py
+python -m pytest -q tests/test_benchmark_utils_contract.py tests/test_solution_value_validation.py
 ```
 
-The root contract suite is the shared Level 1/library contract suite. Repository CI runs this same suite on Python 3.10, 3.11, and 3.12 for pull requests and for pushes to `main` and branches matching `feat/**`. These commands do not run the self-contained Level 3 test suite.
+The shared Level 1/library contract suite and solution-value regression tests run on Python 3.10, 3.11, and 3.12 for pull requests and for pushes to `main` and branches matching `feat/**` or `fix/**`. These commands do not run the self-contained Level 3 test suite.
 
 ## Benchmark changes
 
diff --git a/README.md b/README.md
index 6303660..d5d3422 100644
--- a/README.md
+++ b/README.md
@@ -315,7 +315,7 @@ If an OpenAI run stops early after a retry failure, partial outputs are preserve
 
 ## Development Notes
 
-- Run the top-level Level 2/shared-library regression tests from the repository root with `python -m unittest discover -s tests -v`. The self-contained Level 3 suite remains separate.
+- Run the top-level Level 2/shared-library regression tests from the repository root with `python -m unittest discover -s tests -p 'test_anytime_risk.py' -v`. The self-contained Level 3 suite remains separate.
 - Use Level 1 to test basic steady-state action submission and physical validation.
 - Use Level 2 to test agentic behavior, tool use, validation-budget allocation, evidence-backed reporting, and LLM workflows.
 - Keep hidden oracle quantities, private endpoint URLs, and API keys outside the public repository.
diff --git a/benchmarks/steady/level_1/README.md b/benchmarks/steady/level_1/README.md
index 09b5d71..7f129af 100644
--- a/benchmarks/steady/level_1/README.md
+++ b/benchmarks/steady/level_1/README.md
@@ -211,6 +211,7 @@ Expected format:
 Rules:
 
 - Omitted actions default to zero.
+- Values must be finite numeric values; booleans are not numeric action values. NaN, positive or negative infinity, and integers that cannot be represented as finite floats are invalid.
 - Values must match the step size in `actionspace.json`.
 - Values outside the published min/max range are invalid.
 - Duplicate action ids are invalid.
diff --git a/poweragentbench/benchmark_utils.py b/poweragentbench/benchmark_utils.py
index bfefe10..d0b9294 100644
--- a/poweragentbench/benchmark_utils.py
+++ b/poweragentbench/benchmark_utils.py
@@ -119,11 +119,18 @@ def load_solution(path: Path) -> dict[str, float]:
         value = item.get("value")
         if not isinstance(action_id, str):
             raise ValueError("Each action entry must include a string 'id'.")
-        if not isinstance(value, (int, float)):
+        # bool subclasses int in Python, but JSON booleans are not numeric actions.
+        if isinstance(value, bool) or not isinstance(value, (int, float)):
             raise ValueError(f"Action '{action_id}' must include a numeric 'value'.")
+        try:
+            numeric_value = float(value)
+        except (OverflowError, ValueError) as exc:
+            raise ValueError(f"Action '{action_id}' must include a finite numeric 'value'.") from exc
+        if not math.isfinite(numeric_value):
+            raise ValueError(f"Action '{action_id}' must include a finite numeric 'value'.")
         if action_id in values:
             raise ValueError(f"Duplicate action '{action_id}' in solution file.")
-        values[action_id] = float(value)
+        values[action_id] = numeric_value
 
     return values
 
@@ -144,16 +151,24 @@ def validate_action_values(
     for action_id, value in action_values.items():
         if action_id not in actionspace["action_index"]:
             raise ValueError(f"Unknown action '{action_id}'.")
+        if isinstance(value, bool) or not isinstance(value, (int, float)):
+            raise ValueError(f"Action '{action_id}' must be a numeric value.")
+        try:
+            numeric_value = float(value)
+        except (OverflowError, ValueError) as exc:
+            raise ValueError(f"Action '{action_id}' must be a finite numeric value.") from exc
+        if not math.isfinite(numeric_value):
+            raise ValueError(f"Action '{action_id}' must be a finite numeric value.")
         action = actionspace["action_index"][action_id]
         min_value = float(action["min_value"])
         max_value = float(action["max_value"])
         step = float(action["step"])
-        if value < min_value - 1e-9 or value > max_value + 1e-9:
+        if numeric_value < min_value - 1e-9 or numeric_value > max_value + 1e-9:
             raise ValueError(
-                f"Action '{action_id}' value {value} outside [{min_value}, {max_value}]."
+                f"Action '{action_id}' value {numeric_value} outside [{min_value}, {max_value}]."
             )
-        _assert_step(value, min_value, step, action_id)
-        validated[action_id] = value
+        _assert_step(numeric_value, min_value, step, action_id)
+        validated[action_id] = numeric_value
 
     return validated
 
diff --git a/tests/test_solution_value_validation.py b/tests/test_solution_value_validation.py
new file mode 100644
index 0000000..2d3ad09
--- /dev/null
+++ b/tests/test_solution_value_validation.py
@@ -0,0 +1,105 @@
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+import pytest
+
+from poweragentbench.benchmark_utils import load_solution, validate_action_values
+
+
+ACTIONSPACE = {
+    "action_index": {
+        "tap_T1_up": {
+            "min_value": 0.0,
+            "max_value": 2.0,
+            "step": 1.0,
+        },
+        "phase_shift_T9": {
+            "min_value": -2.0,
+            "max_value": 2.0,
+            "step": 2.0,
+        },
+        "shed_load_L20": {
+            "min_value": 0.0,
+            "max_value": 5.0,
+            "step": 2.5,
+        },
+    }
+}
+
+
+@pytest.mark.parametrize(
+    ("value", "message"),
+    [
+        (True, "numeric"),
+        (False, "numeric"),
+        (float("nan"), "finite"),
+        (float("inf"), "finite"),
+        (float("-inf"), "finite"),
+        (10**400, "finite"),
+    ],
+    ids=["true", "false", "nan", "positive-infinity", "negative-infinity", "huge-integer"],
+)
+def test_load_solution_rejects_invalid_numeric_values(
+    tmp_path: Path, value: object, message: str
+) -> None:
+    path = tmp_path / "solution.json"
+    path.write_text(
+        json.dumps({"actions": [{"id": "tap_T1_up", "value": value}]}),
+        encoding="utf-8",
+    )
+
+    with pytest.raises(ValueError, match=message):
+        load_solution(path)
+
+
+@pytest.mark.parametrize(
+    ("value", "message"),
+    [
+        (True, "numeric"),
+        (False, "numeric"),
+        (float("nan"), "finite"),
+        (float("inf"), "finite"),
+        (float("-inf"), "finite"),
+        (10**400, "finite"),
+    ],
+    ids=["true", "false", "nan", "positive-infinity", "negative-infinity", "huge-integer"],
+)
+def test_validate_action_values_rejects_invalid_numeric_values(
+    value: object, message: str
+) -> None:
+    with pytest.raises(ValueError, match=message):
+        validate_action_values({"tap_T1_up": value}, ACTIONSPACE)
+
+
+@pytest.mark.parametrize(
+    ("action_id", "value", "expected"),
+    [
+        ("tap_T1_up", 0, 0.0),
+        ("tap_T1_up", 1, 1.0),
+        ("tap_T1_up", 1.0, 1.0),
+        ("phase_shift_T9", -2, -2.0),
+        ("shed_load_L20", 2.5, 2.5),
+    ],
+    ids=["zero", "integer", "float", "permitted-negative", "permitted-fraction"],
+)
+def test_load_solution_accepts_numeric_values_and_validates_defaults(
+    tmp_path: Path, action_id: str, value: int | float, expected: float
+) -> None:
+    path = tmp_path / "solution.json"
+    path.write_text(
+        json.dumps({"actions": [{"id": action_id, "value": value}]}),
+        encoding="utf-8",
+    )
+
+    loaded = load_solution(path)
+    assert loaded == {action_id: expected}
+
+    validated = validate_action_values(loaded, ACTIONSPACE)
+    assert validated[action_id] == expected
+    assert all(
+        validated[other_id] == 0.0
+        for other_id in ACTIONSPACE["action_index"]
+        if other_id != action_id
+    )