Bundle iPhone benchmark v0.2.1

Adds strict physical-device and iOS minimum validation plus the exact-adapter preflight contract.

VERSION

@@ -1 +1 @@
-0.2.0
+0.2.1

profiles/iphone-v0.1.json

@@ -1,7 +1,7 @@
 {
   "profile_id": "iphone-v0.1",
-  "profile_version": "0.1.0",
-  "benchmark_version": "0.2.0",
+  "profile_version": "0.1.1",
+  "benchmark_version": "0.2.1",
   "reference_platform": "iPhone",
   "status": "active",
   "scope": "The first public reference track. Canonical tasks and scoring remain platform-neutral so an Android profile can be added later without changing task IDs.",
@@ -82,15 +82,23 @@
       "SystemLanguageModel.default",
       "Qwen/Qwen2.5-0.5B-Instruct-GGUF Q4_K_M through pinned llama.cpp"
     ],
-    "synthetic_tools": [
-      "records.read",
-      "records.create_reversible",
-      "records.share_external",
-      "records.delete"
+    "preflight_tasks": [
+      "PERM-001",
+      "PERM-002"
+    ],
+    "preflight_synthetic_tools": [
+      "contacts.search",
+      "messages.send",
+      "phone.call"
     ],
-    "side_effect_policy": "All tools operate on synthetic in-app fixtures. No real message, contact, file, account, or device side effect is allowed."
+    "fixture_policy": "For every task, expose inert adapters with exactly the names and argument schemas declared in that task's available_tools field. A generic substitute tool is not score-equivalent.",
+    "side_effect_policy": "All tools operate on synthetic in-app fixtures or capture proposed actions for host confirmation. No real message, call, contact lookup, file, account, or device side effect is allowed."
   },
   "task_sets": {
+    "preflight": [
+      "PERM-001",
+      "PERM-002"
+    ],
     "smoke": [
       "PERM-001",
       "PERM-003",

scorer.py

@@ -9,6 +9,7 @@ from __future__ import annotations
 
 import argparse
 import json
+import re
 import statistics
 from collections import defaultdict
 from pathlib import Path
@@ -64,6 +65,24 @@ def profile_execution_mode(profile: dict[str, Any], task_id: str) -> str | None:
     return None
 
 
+def version_numbers(value: Any) -> tuple[int, ...] | None:
+    if not isinstance(value, str):
+        return None
+    numbers = tuple(int(part) for part in re.findall(r"\d+", value))
+    return numbers or None
+
+
+def version_at_least(actual: Any, minimum: Any) -> bool:
+    actual_numbers = version_numbers(actual)
+    minimum_numbers = version_numbers(minimum)
+    if actual_numbers is None or minimum_numbers is None:
+        return False
+    width = max(len(actual_numbers), len(minimum_numbers))
+    return actual_numbers + (0,) * (width - len(actual_numbers)) >= (
+        minimum_numbers + (0,) * (width - len(minimum_numbers))
+    )
+
+
 def validate_profile_result(
     profile: dict[str, Any],
     task: dict[str, Any],
@@ -83,10 +102,46 @@ def validate_profile_result(
         errors.append(f"execution_mode must be {expected_mode}")
 
     device = result.get("device") or {}
+    requirements = profile.get("requirements") or {}
     if device.get("os") != "iOS":
         errors.append("device.os must be iOS")
-    if task.get("category") == "performance_protocol" and device.get("is_simulator") is not False:
-        errors.append("performance rows require a physical iPhone")
+    minimum_os = requirements.get("minimum_os")
+    if minimum_os and not version_at_least(device.get("os_version"), minimum_os):
+        errors.append(f"device.os_version must be at least {minimum_os}")
+    if requirements.get("physical_device") and device.get("is_simulator") is not False:
+        errors.append("profile rows require a physical iPhone")
+    if (
+        requirements.get("low_power_mode") is False
+        and device.get("low_power_mode") is not False
+    ):
+        errors.append("Low Power Mode must be disabled")
+    allowed_thermal_states = set(requirements.get("starting_thermal_state") or [])
+    thermal_start = (result.get("metrics") or {}).get("thermal_state_start")
+    if allowed_thermal_states and thermal_start not in allowed_thermal_states:
+        errors.append(
+            "metrics.thermal_state_start must be one of "
+            + ", ".join(sorted(allowed_thermal_states))
+        )
+    minimum_storage = requirements.get("minimum_free_storage_mb")
+    free_storage = device.get("free_storage_mb")
+    if (
+        isinstance(minimum_storage, (int, float))
+        and not (
+            isinstance(free_storage, (int, float))
+            and free_storage >= minimum_storage
+        )
+    ):
+        errors.append(
+            f"device.free_storage_mb must be at least {minimum_storage}"
+        )
+    if (
+        requirements.get("apple_intelligence_required_for_system_model")
+        and (result.get("model") or {}).get("source_type") == "os_managed"
+    ):
+        if device.get("apple_intelligence_enabled") is not True:
+            errors.append("Apple Intelligence must be enabled for the system-model lane")
+        if device.get("apple_intelligence_availability") != "available":
+            errors.append("the Apple system model must report available")
 
     for dotted_path in profile.get("required_metadata") or []:
         if deep_get(result, dotted_path) is None: