fix(eval): stop pinning temperature and reasoning effort on the wire

The eval CLI defaulted temperature to 0.7 and reasoning effort to
medium, so every sweep sent code-chosen sampling knobs the house
assignment scheme forbids — the wire should omit the fields and let
the alias / stored setting / serving default apply, as production
does. Both flags now default to unset and the harnesses plumb None
through to model_turn, whose provider layer already omits absent
knobs. Absolute numbers from earlier sweeps were collected under the
pinned values; contrasts were at least uniform under the same pin.

turnstone/eval/cli.py

@@ -184,8 +184,12 @@ def main() -> None:
     parser.add_argument(
         "--temperature",
         type=float,
-        default=0.7,
-        help="Sampling temperature (default: 0.7)",
+        default=None,
+        help=(
+            "Sampling temperature. Omitted from the wire by default so "
+            "the alias / stored setting / serving default applies — the "
+            "same assignment scheme production runs"
+        ),
     )
     parser.add_argument(
         "--max-tokens",
@@ -195,9 +199,13 @@ def main() -> None:
     )
     parser.add_argument(
         "--reasoning-effort",
-        default="medium",
+        default=None,
         choices=["low", "medium", "high"],
-        help="Reasoning effort (default: medium)",
+        help=(
+            "Reasoning effort. Omitted from the wire by default — a "
+            "code-chosen token is unvetted and can flip thinking on for "
+            "lanes the operator never engaged"
+        ),
     )
     parser.add_argument(
         "--context-window",

turnstone/eval/core.py

@@ -238,10 +238,10 @@ class HeadlessSession(ChatSession):
         model: str,
         system_prompt_override: str | None = None,
         instructions: str | None = None,
-        temperature: float = 0.7,
+        temperature: float | None = None,
         max_tokens: int = 32768,
         tool_timeout: int = 30,
-        reasoning_effort: str = "medium",
+        reasoning_effort: str | None = None,
         context_window: int = 131072,
         compact_max_tokens: int = 32768,
         auto_compact_pct: float = 0.8,
@@ -651,9 +651,9 @@ def _run_single_test(
     model: str,
     system_prompt: str,
     case: dict[str, Any],
-    temperature: float,
+    temperature: float | None,
     max_tokens: int,
-    reasoning_effort: str,
+    reasoning_effort: str | None,
     context_window: int,
     verbose: bool = False,
     log_prefix: str = "",
@@ -1100,9 +1100,9 @@ def _run_iteration_parallel(
     system_prompt: str,
     cases: list[dict[str, Any]],
     n_runs: int,
-    temperature: float,
+    temperature: float | None,
     max_tokens: int,
-    reasoning_effort: str,
+    reasoning_effort: str | None,
     context_window: int,
     test_timeout: int,
     parallel: int,
@@ -1251,9 +1251,9 @@ def _run_iteration(
     system_prompt: str,
     cases: list[dict[str, Any]],
     n_runs: int,
-    temperature: float,
+    temperature: float | None,
     max_tokens: int,
-    reasoning_effort: str,
+    reasoning_effort: str | None,
     context_window: int,
     verbose: bool = False,
     test_timeout: int = 300,
@@ -1469,9 +1469,9 @@ def run_skill_adherence(
     model: str,
     cases: list[dict[str, Any]],
     n_runs: int,
-    temperature: float,
+    temperature: float | None,
     max_tokens: int,
-    reasoning_effort: str,
+    reasoning_effort: str | None,
     context_window: int,
     test_timeout: int = 300,
     parallel: int = 1,

turnstone/eval/nudges.py

@@ -517,9 +517,9 @@ class CoordinatorHeadlessSession(HeadlessSession):
         coord_client: CoordinatorClient,
         ws_id: str,
         user_id: str,
-        temperature: float,
+        temperature: float | None,
         max_tokens: int,
-        reasoning_effort: str,
+        reasoning_effort: str | None,
         context_window: int,
     ) -> None:
         super().__init__(
@@ -981,9 +981,9 @@ def _run_single_nudge(
     model: str,
     case: dict[str, Any],
     arm: str,
-    temperature: float,
+    temperature: float | None,
     max_tokens: int,
-    reasoning_effort: str,
+    reasoning_effort: str | None,
     context_window: int,
     max_turns: int,
     test_timeout: int,
@@ -2019,9 +2019,9 @@ def run_nudge_response(
     model: str,
     cells: list[dict[str, Any]],
     n_runs: int = 10,
-    temperature: float = 0.7,
+    temperature: float | None = None,
     max_tokens: int = 8192,
-    reasoning_effort: str = "medium",
+    reasoning_effort: str | None = None,
     context_window: int = 131072,
     max_turns: int = 8,
     test_timeout: int = 300,