[claude] Add content moderation pipeline (Llama Guard + game-context prompts) (#1056) (#1059)

config/moderation.yaml

@@ -0,0 +1,107 @@
+# Content Moderation Profiles
+# Per-game moderation configuration for the AI narrator pipeline.
+#
+# Each profile defines:
+#   - vocabulary_whitelist: Game terms safe in context (won't trigger moderation)
+#   - context_prompt: System prompt framing for the narrator
+#   - threshold: Confidence threshold — flags below this pass through
+#   - fallbacks: Pre-generated safe narration by scene type
+#
+# Model options (from research):
+#   llama-guard3:1b  — Speed (<30ms/sentence, INT4 quantized)
+#   shieldgemma:2b   — Accuracy (+10.8% AU-PRC, ~50-100ms)
+#
+# Override guard model via MODERATION_GUARD_MODEL env var.
+
+# ── Guard model selection ────────────────────────────────────────────────────
+guard_model: "llama-guard3:1b"
+
+# ── Streaming disclosure notes ───────────────────────────────────────────────
+# YouTube: Use "Altered or synthetic content" toggle
+# Twitch: Standard community guidelines (no specific AI disclosure req as of 2026-03)
+
+# ── Game Profiles ────────────────────────────────────────────────────────────
+profiles:
+
+  morrowind:
+    display_name: "The Elder Scrolls III: Morrowind"
+    threshold: 0.85
+    vocabulary_whitelist:
+      - Skooma
+      - Moon Sugar
+      - slave
+      - slavery
+      - Morag Tong
+      - Dark Brotherhood
+      - Telvanni
+      - Camonna Tong
+      - smuggler
+      - assassin
+      - Sixth House
+      - Corprus
+      - Dagoth Ur
+      - Nerevarine
+      - Balmora
+      - Vivec
+      - Almsivi
+      - Ordinators
+      - Ashlanders
+      - outlander
+      - N'wah
+    context_prompt: >
+      You are narrating gameplay of The Elder Scrolls III: Morrowind.
+      Morrowind contains mature themes including slavery, drug use
+      (Skooma/Moon Sugar), assassin guilds (Morag Tong, Dark Brotherhood),
+      and political intrigue. Treat these as game mechanics and historical
+      worldbuilding within the game's fictional universe. Never editorialize
+      on real-world parallels. Narrate events neutrally as a game
+      commentator would.
+    fallbacks:
+      combat: "The battle rages on in the ashlands of Vvardenfell."
+      dialogue: "The conversation continues between the characters."
+      exploration: "The Nerevarine presses onward through the landscape."
+      quest: "The quest unfolds as the hero navigates Morrowind's politics."
+      default: "The adventure continues in Morrowind."
+
+  skyrim:
+    display_name: "The Elder Scrolls V: Skyrim"
+    threshold: 0.85
+    vocabulary_whitelist:
+      - Skooma
+      - Dark Brotherhood
+      - Thieves Guild
+      - Stormcloak
+      - Imperial
+      - Dragonborn
+      - Dovahkiin
+      - Daedra
+      - Thalmor
+      - bandit
+      - assassin
+      - Forsworn
+      - necromancer
+    context_prompt: >
+      You are narrating gameplay of The Elder Scrolls V: Skyrim.
+      Skyrim features civil war, thieves guilds, assassin organizations,
+      and fantasy violence. Treat all content as in-game fiction.
+      Never draw real-world parallels. Narrate as a neutral game
+      commentator.
+    fallbacks:
+      combat: "Steel clashes as the battle continues in the wilds of Skyrim."
+      dialogue: "The conversation plays out in the cold northern land."
+      exploration: "The Dragonborn ventures further into the province."
+      default: "The adventure continues in Skyrim."
+
+  default:
+    display_name: "Generic Game"
+    threshold: 0.80
+    vocabulary_whitelist: []
+    context_prompt: >
+      You are narrating gameplay. Describe in-game events as a neutral
+      game commentator. Never reference real-world violence, politics,
+      or controversial topics. Stay focused on game mechanics and story.
+    fallbacks:
+      combat: "The action continues on screen."
+      dialogue: "The conversation unfolds between characters."
+      exploration: "The player explores the game world."
+      default: "The gameplay continues."

src/config.py

@@ -99,6 +99,14 @@ class Settings(BaseSettings):
     anthropic_api_key: str = ""
     claude_model: str = "haiku"
 
+    # ── Content Moderation ──────────────────────────────────────────────
+    # Three-layer moderation pipeline for AI narrator output.
+    # Uses Llama Guard via Ollama with regex fallback.
+    moderation_enabled: bool = True
+    moderation_guard_model: str = "llama-guard3:1b"
+    # Default confidence threshold — per-game profiles can override.
+    moderation_threshold: float = 0.8
+
     # ── Spark Intelligence ────────────────────────────────────────────────
     # Enable/disable the Spark cognitive layer.
     # When enabled, Spark captures swarm events, runs EIDOS predictions,

src/infrastructure/guards/__init__.py

@@ -0,0 +1,7 @@
+"""Content moderation pipeline for AI narrator output.
+
+Three-layer defense:
+1. Game-context system prompts (vocabulary whitelists, theme framing)
+2. Real-time output filter via Llama Guard (or fallback regex)
+3. Per-game moderation profiles with configurable thresholds
+"""

src/infrastructure/guards/moderation.py

@@ -0,0 +1,500 @@
+"""Content moderation pipeline for AI narrator output.
+
+Three-layer defense against harmful LLM output:
+
+Layer 1 — Game-context system prompts with per-game vocabulary whitelists.
+Layer 2 — Real-time output filter (Llama Guard via Ollama, regex fallback).
+Layer 3 — Per-game moderation profiles with configurable thresholds.
+
+Usage:
+    from infrastructure.guards.moderation import get_moderator
+
+    moderator = get_moderator()
+    result = await moderator.check("Some narrator text", game="morrowind")
+    if result.blocked:
+        use_fallback_narration(result.fallback)
+"""
+
+import logging
+import re
+import time
+from dataclasses import dataclass, field
+from datetime import UTC, datetime
+from enum import Enum
+from typing import Any
+
+from config import settings
+
+logger = logging.getLogger(__name__)
+
+
+class ModerationVerdict(Enum):
+    """Result of a moderation check."""
+
+    PASS = "pass"
+    FAIL = "fail"
+    ERROR = "error"
+
+
+class ViolationCategory(Enum):
+    """Categories of content violations."""
+
+    HATE_SPEECH = "hate_speech"
+    VIOLENCE_GLORIFICATION = "violence_glorification"
+    REAL_WORLD_HARM = "real_world_harm"
+    SEXUAL_CONTENT = "sexual_content"
+    SELF_HARM = "self_harm"
+    NONE = "none"
+
+
+@dataclass
+class ModerationResult:
+    """Result from the moderation pipeline."""
+
+    verdict: ModerationVerdict
+    blocked: bool
+    category: ViolationCategory = ViolationCategory.NONE
+    confidence: float = 0.0
+    latency_ms: float = 0.0
+    layer: str = ""  # Which layer caught it
+    fallback: str = ""  # Contextual fallback narration
+    reason: str = ""
+    timestamp: str = field(default_factory=lambda: datetime.now(UTC).isoformat())
+
+    @property
+    def passed(self) -> bool:
+        return self.verdict == ModerationVerdict.PASS
+
+
+@dataclass
+class GameProfile:
+    """Per-game moderation profile.
+
+    Defines vocabulary whitelists, theme framing rules,
+    and threshold tuning for a specific game.
+    """
+
+    game_id: str
+    display_name: str
+    # Terms that are safe in game context (e.g., "Skooma", "slave", "Morag Tong")
+    vocabulary_whitelist: list[str] = field(default_factory=list)
+    # System prompt additions for game-context framing
+    context_prompt: str = ""
+    # Confidence threshold — below this, content passes even if flagged
+    threshold: float = 0.8
+    # Scene-type fallback narrations
+    fallbacks: dict[str, str] = field(default_factory=dict)
+
+
+# ── Regex fallback patterns ─────────────────────────────────────────────────
+# Used when Llama Guard is unavailable. Catches obvious violations only.
+# Intentionally conservative — false negatives preferred over false positives.
+_FALLBACK_PATTERNS: list[tuple[re.Pattern, ViolationCategory]] = [
+    (
+        re.compile(
+            r"\b(kill\s+(?:all|every)\s+(?:real|actual)\s+(?:people|humans|children))\b",
+            re.IGNORECASE,
+        ),
+        ViolationCategory.REAL_WORLD_HARM,
+    ),
+    (
+        re.compile(
+            r"\b((?:you|we)\s+should\s+(?:actually|really)\s+(?:harm|hurt|kill))\b",
+            re.IGNORECASE,
+        ),
+        ViolationCategory.REAL_WORLD_HARM,
+    ),
+    (
+        re.compile(
+            r"\b((?:in\s+real\s+life|irl)\s+(?:you|we)\s+(?:should|could|must)\s+"
+            r"(?:attack|hurt|kill|harm))\b",
+            re.IGNORECASE,
+        ),
+        ViolationCategory.REAL_WORLD_HARM,
+    ),
+]
+
+# ── Default game profiles ───────────────────────────────────────────────────
+
+_DEFAULT_PROFILES: dict[str, GameProfile] = {
+    "morrowind": GameProfile(
+        game_id="morrowind",
+        display_name="The Elder Scrolls III: Morrowind",
+        vocabulary_whitelist=[
+            "Skooma",
+            "Moon Sugar",
+            "slave",
+            "slavery",
+            "Morag Tong",
+            "Dark Brotherhood",
+            "Telvanni",
+            "Camonna Tong",
+            "smuggler",
+            "assassin",
+            "Sixth House",
+            "Corprus",
+            "Dagoth Ur",
+            "Nerevarine",
+        ],
+        context_prompt=(
+            "You are narrating gameplay of The Elder Scrolls III: Morrowind. "
+            "Morrowind contains mature themes including slavery, drug use (Skooma/Moon Sugar), "
+            "assassin guilds (Morag Tong, Dark Brotherhood), and political intrigue. "
+            "Treat these as game mechanics and historical worldbuilding within the game's "
+            "fictional universe. Never editorialize on real-world parallels. "
+            "Narrate events neutrally as a game commentator would."
+        ),
+        threshold=0.85,
+        fallbacks={
+            "combat": "The battle rages on in the ashlands of Vvardenfell.",
+            "dialogue": "The conversation continues between the characters.",
+            "exploration": "The Nerevarine presses onward through the landscape.",
+            "default": "The adventure continues in Morrowind.",
+        },
+    ),
+    "default": GameProfile(
+        game_id="default",
+        display_name="Generic Game",
+        vocabulary_whitelist=[],
+        context_prompt=(
+            "You are narrating gameplay. Describe in-game events as a neutral "
+            "game commentator. Never reference real-world violence, politics, "
+            "or controversial topics. Stay focused on game mechanics and story."
+        ),
+        threshold=0.8,
+        fallbacks={
+            "combat": "The action continues on screen.",
+            "dialogue": "The conversation unfolds between characters.",
+            "exploration": "The player explores the game world.",
+            "default": "The gameplay continues.",
+        },
+    ),
+}
+
+
+class ContentModerator:
+    """Three-layer content moderation pipeline.
+
+    Layer 1: Game-context system prompts with vocabulary whitelists.
+    Layer 2: LLM-based moderation (Llama Guard via Ollama, with regex fallback).
+    Layer 3: Per-game threshold tuning and profile-based filtering.
+
+    Follows graceful degradation — if Llama Guard is unavailable,
+    falls back to regex patterns. Never crashes.
+    """
+
+    def __init__(
+        self,
+        profiles: dict[str, GameProfile] | None = None,
+        guard_model: str | None = None,
+    ) -> None:
+        self._profiles: dict[str, GameProfile] = profiles or dict(_DEFAULT_PROFILES)
+        self._guard_model = guard_model or settings.moderation_guard_model
+        self._guard_available: bool | None = None  # Lazy-checked
+        self._metrics = _ModerationMetrics()
+
+    def get_profile(self, game: str) -> GameProfile:
+        """Get the moderation profile for a game, falling back to default."""
+        return self._profiles.get(game, self._profiles["default"])
+
+    def register_profile(self, profile: GameProfile) -> None:
+        """Register or update a game moderation profile."""
+        self._profiles[profile.game_id] = profile
+        logger.info("Registered moderation profile: %s", profile.game_id)
+
+    def get_context_prompt(self, game: str) -> str:
+        """Get the game-context system prompt (Layer 1).
+
+        Returns the context prompt for the given game, which should be
+        prepended to the narrator's system prompt.
+        """
+        profile = self.get_profile(game)
+        return profile.context_prompt
+
+    async def check(
+        self,
+        text: str,
+        game: str = "default",
+        scene_type: str = "default",
+    ) -> ModerationResult:
+        """Run the full moderation pipeline on narrator output.
+
+        Args:
+            text: The text to moderate (narrator output).
+            game: Game identifier for profile selection.
+            scene_type: Current scene type for fallback selection.
+
+        Returns:
+            ModerationResult with verdict, confidence, and fallback.
+        """
+        start = time.monotonic()
+        profile = self.get_profile(game)
+
+        # Layer 1: Vocabulary whitelist pre-processing
+        cleaned_text = self._apply_whitelist(text, profile)
+
+        # Layer 2: LLM guard or regex fallback
+        result = await self._run_guard(cleaned_text, profile)
+
+        # Layer 3: Threshold tuning
+        if result.verdict == ModerationVerdict.FAIL and result.confidence < profile.threshold:
+            logger.info(
+                "Moderation flag below threshold (%.2f < %.2f) — allowing",
+                result.confidence,
+                profile.threshold,
+            )
+            result = ModerationResult(
+                verdict=ModerationVerdict.PASS,
+                blocked=False,
+                confidence=result.confidence,
+                layer="threshold",
+                reason=f"Below threshold ({result.confidence:.2f} < {profile.threshold:.2f})",
+            )
+
+        # Attach fallback narration if blocked
+        if result.blocked:
+            result.fallback = profile.fallbacks.get(
+                scene_type, profile.fallbacks.get("default", "")
+            )
+
+        result.latency_ms = (time.monotonic() - start) * 1000
+        self._metrics.record(result)
+
+        if result.blocked:
+            logger.warning(
+                "Content blocked [%s/%s]: category=%s confidence=%.2f reason=%s",
+                game,
+                scene_type,
+                result.category.value,
+                result.confidence,
+                result.reason,
+            )
+
+        return result
+
+    def _apply_whitelist(self, text: str, profile: GameProfile) -> str:
+        """Layer 1: Replace whitelisted game terms with placeholders.
+
+        This prevents the guard model from flagging in-game terminology
+        (e.g., "Skooma" being flagged as drug reference).
+        """
+        cleaned = text
+        for term in profile.vocabulary_whitelist:
+            # Case-insensitive replacement with a neutral placeholder
+            pattern = re.compile(re.escape(term), re.IGNORECASE)
+            cleaned = pattern.sub("[GAME_TERM]", cleaned)
+        return cleaned
+
+    async def _run_guard(
+        self, text: str, profile: GameProfile
+    ) -> ModerationResult:
+        """Layer 2: Run LLM guard model or fall back to regex."""
+        if not settings.moderation_enabled:
+            return ModerationResult(
+                verdict=ModerationVerdict.PASS,
+                blocked=False,
+                layer="disabled",
+                reason="Moderation disabled",
+            )
+
+        # Try Llama Guard via Ollama
+        if await self._is_guard_available():
+            try:
+                return await self._check_with_guard(text)
+            except Exception as exc:
+                logger.warning("Guard model failed, using regex fallback: %s", exc)
+                self._guard_available = False
+
+        # Regex fallback
+        return self._check_with_regex(text)
+
+    async def _is_guard_available(self) -> bool:
+        """Check if the guard model is available via Ollama."""
+        if self._guard_available is not None:
+            return self._guard_available
+
+        try:
+            import aiohttp
+
+            url = f"{settings.normalized_ollama_url}/api/tags"
+            timeout = aiohttp.ClientTimeout(total=5)
+            async with aiohttp.ClientSession(timeout=timeout) as session:
+                async with session.get(url) as resp:
+                    if resp.status != 200:
+                        self._guard_available = False
+                        return False
+                    data = await resp.json()
+                    models = [m.get("name", "") for m in data.get("models", [])]
+                    self._guard_available = any(
+                        self._guard_model in m or m.startswith(self._guard_model)
+                        for m in models
+                    )
+                    if not self._guard_available:
+                        logger.info(
+                            "Guard model '%s' not found in Ollama — using regex fallback",
+                            self._guard_model,
+                        )
+                    return self._guard_available
+        except Exception as exc:
+            logger.debug("Ollama guard check failed: %s", exc)
+            self._guard_available = False
+            return False
+
+    async def _check_with_guard(self, text: str) -> ModerationResult:
+        """Run moderation check via Llama Guard."""
+        import aiohttp
+
+        url = f"{settings.normalized_ollama_url}/api/chat"
+        payload = {
+            "model": self._guard_model,
+            "messages": [
+                {
+                    "role": "user",
+                    "content": text,
+                }
+            ],
+            "stream": False,
+            "options": {"temperature": 0.0},
+        }
+
+        timeout = aiohttp.ClientTimeout(total=10)
+        async with aiohttp.ClientSession(timeout=timeout) as session:
+            async with session.post(url, json=payload) as resp:
+                if resp.status != 200:
+                    raise RuntimeError(f"Guard API error: {resp.status}")
+                data = await resp.json()
+
+        response_text = data.get("message", {}).get("content", "").strip().lower()
+
+        # Llama Guard returns "safe" or "unsafe\n<category>"
+        if response_text.startswith("safe"):
+            return ModerationResult(
+                verdict=ModerationVerdict.PASS,
+                blocked=False,
+                confidence=0.0,
+                layer="llama_guard",
+                reason="Content safe",
+            )
+
+        # Parse unsafe response
+        category = ViolationCategory.NONE
+        confidence = 0.95  # High confidence from LLM guard
+        lines = response_text.split("\n")
+        if len(lines) > 1:
+            cat_str = lines[1].strip()
+            category = _parse_guard_category(cat_str)
+
+        return ModerationResult(
+            verdict=ModerationVerdict.FAIL,
+            blocked=True,
+            category=category,
+            confidence=confidence,
+            layer="llama_guard",
+            reason=f"Guard flagged: {response_text}",
+        )
+
+    def _check_with_regex(self, text: str) -> ModerationResult:
+        """Regex fallback when guard model is unavailable.
+
+        Intentionally conservative — only catches obvious real-world harm.
+        """
+        for pattern, category in _FALLBACK_PATTERNS:
+            match = pattern.search(text)
+            if match:
+                return ModerationResult(
+                    verdict=ModerationVerdict.FAIL,
+                    blocked=True,
+                    category=category,
+                    confidence=0.95,  # Regex patterns are high-signal
+                    layer="regex_fallback",
+                    reason=f"Regex match: {match.group(0)[:50]}",
+                )
+
+        return ModerationResult(
+            verdict=ModerationVerdict.PASS,
+            blocked=False,
+            layer="regex_fallback",
+            reason="No regex matches",
+        )
+
+    def get_metrics(self) -> dict[str, Any]:
+        """Get moderation pipeline metrics."""
+        return self._metrics.to_dict()
+
+    def reset_guard_cache(self) -> None:
+        """Reset the guard availability cache (e.g., after pulling model)."""
+        self._guard_available = None
+
+
+class _ModerationMetrics:
+    """Tracks moderation pipeline performance."""
+
+    def __init__(self) -> None:
+        self.total_checks: int = 0
+        self.passed: int = 0
+        self.blocked: int = 0
+        self.errors: int = 0
+        self.total_latency_ms: float = 0.0
+        self.by_layer: dict[str, int] = {}
+        self.by_category: dict[str, int] = {}
+
+    def record(self, result: ModerationResult) -> None:
+        self.total_checks += 1
+        self.total_latency_ms += result.latency_ms
+
+        if result.verdict == ModerationVerdict.PASS:
+            self.passed += 1
+        elif result.verdict == ModerationVerdict.FAIL:
+            self.blocked += 1
+        else:
+            self.errors += 1
+
+        layer = result.layer or "unknown"
+        self.by_layer[layer] = self.by_layer.get(layer, 0) + 1
+
+        if result.blocked:
+            cat = result.category.value
+            self.by_category[cat] = self.by_category.get(cat, 0) + 1
+
+    def to_dict(self) -> dict[str, Any]:
+        return {
+            "total_checks": self.total_checks,
+            "passed": self.passed,
+            "blocked": self.blocked,
+            "errors": self.errors,
+            "avg_latency_ms": (
+                round(self.total_latency_ms / self.total_checks, 2)
+                if self.total_checks > 0
+                else 0.0
+            ),
+            "by_layer": dict(self.by_layer),
+            "by_category": dict(self.by_category),
+        }
+
+
+def _parse_guard_category(cat_str: str) -> ViolationCategory:
+    """Parse Llama Guard category string to ViolationCategory."""
+    cat_lower = cat_str.lower()
+    if "hate" in cat_lower:
+        return ViolationCategory.HATE_SPEECH
+    if "violence" in cat_lower:
+        return ViolationCategory.VIOLENCE_GLORIFICATION
+    if "sexual" in cat_lower:
+        return ViolationCategory.SEXUAL_CONTENT
+    if "self-harm" in cat_lower or "self_harm" in cat_lower or "suicide" in cat_lower:
+        return ViolationCategory.SELF_HARM
+    if "harm" in cat_lower or "dangerous" in cat_lower:
+        return ViolationCategory.REAL_WORLD_HARM
+    return ViolationCategory.NONE
+
+
+# ── Module-level singleton ──────────────────────────────────────────────────
+_moderator: ContentModerator | None = None
+
+
+def get_moderator() -> ContentModerator:
+    """Get or create the content moderator singleton."""
+    global _moderator
+    if _moderator is None:
+        _moderator = ContentModerator()
+    return _moderator

src/infrastructure/guards/profiles.py

@@ -0,0 +1,56 @@
+"""Load game moderation profiles from config/moderation.yaml.
+
+Falls back to hardcoded defaults if the YAML file is missing or malformed.
+"""
+
+import logging
+from pathlib import Path
+
+from infrastructure.guards.moderation import GameProfile
+
+logger = logging.getLogger(__name__)
+
+
+def load_profiles(config_path: Path | None = None) -> dict[str, GameProfile]:
+    """Load game moderation profiles from YAML config.
+
+    Args:
+        config_path: Path to moderation.yaml. Defaults to config/moderation.yaml.
+
+    Returns:
+        Dict mapping game_id to GameProfile.
+    """
+    path = config_path or Path("config/moderation.yaml")
+
+    if not path.exists():
+        logger.info("Moderation config not found at %s — using defaults", path)
+        return {}
+
+    try:
+        import yaml
+    except ImportError:
+        logger.warning("PyYAML not installed — using default moderation profiles")
+        return {}
+
+    try:
+        data = yaml.safe_load(path.read_text())
+    except Exception as exc:
+        logger.error("Failed to parse moderation config: %s", exc)
+        return {}
+
+    profiles: dict[str, GameProfile] = {}
+    for game_id, profile_data in data.get("profiles", {}).items():
+        try:
+            profiles[game_id] = GameProfile(
+                game_id=game_id,
+                display_name=profile_data.get("display_name", game_id),
+                vocabulary_whitelist=profile_data.get("vocabulary_whitelist", []),
+                context_prompt=profile_data.get("context_prompt", ""),
+                threshold=float(profile_data.get("threshold", 0.8)),
+                fallbacks=profile_data.get("fallbacks", {}),
+            )
+        except Exception as exc:
+            logger.warning("Invalid profile '%s': %s", game_id, exc)
+
+    logger.info("Loaded %d moderation profiles from %s", len(profiles), path)
+    return profiles

tests/infrastructure/test_moderation.py

@@ -0,0 +1,335 @@
+"""Tests for the content moderation pipeline."""
+
+from unittest.mock import AsyncMock, patch
+
+import pytest
+
+from infrastructure.guards.moderation import (
+    ContentModerator,
+    GameProfile,
+    ModerationResult,
+    ModerationVerdict,
+    ViolationCategory,
+    _parse_guard_category,
+    get_moderator,
+)
+
+
+# ── Unit tests for data types ────────────────────────────────────────────────
+
+
+class TestModerationResult:
+    """Test ModerationResult dataclass."""
+
+    def test_passed_property_true(self):
+        result = ModerationResult(verdict=ModerationVerdict.PASS, blocked=False)
+        assert result.passed is True
+
+    def test_passed_property_false(self):
+        result = ModerationResult(verdict=ModerationVerdict.FAIL, blocked=True)
+        assert result.passed is False
+
+    def test_default_values(self):
+        result = ModerationResult(verdict=ModerationVerdict.PASS, blocked=False)
+        assert result.category == ViolationCategory.NONE
+        assert result.confidence == 0.0
+        assert result.fallback == ""
+        assert result.reason == ""
+
+
+class TestGameProfile:
+    """Test GameProfile dataclass."""
+
+    def test_default_values(self):
+        profile = GameProfile(game_id="test", display_name="Test Game")
+        assert profile.vocabulary_whitelist == []
+        assert profile.threshold == 0.8
+        assert profile.fallbacks == {}
+
+    def test_morrowind_profile(self):
+        profile = GameProfile(
+            game_id="morrowind",
+            display_name="Morrowind",
+            vocabulary_whitelist=["Skooma", "slave"],
+            threshold=0.85,
+        )
+        assert "Skooma" in profile.vocabulary_whitelist
+        assert profile.threshold == 0.85
+
+
+class TestParseGuardCategory:
+    """Test Llama Guard category parsing."""
+
+    def test_hate_speech(self):
+        assert _parse_guard_category("S1: Hate speech") == ViolationCategory.HATE_SPEECH
+
+    def test_violence(self):
+        assert _parse_guard_category("S2: Violence") == ViolationCategory.VIOLENCE_GLORIFICATION
+
+    def test_sexual_content(self):
+        assert _parse_guard_category("S3: Sexual content") == ViolationCategory.SEXUAL_CONTENT
+
+    def test_self_harm(self):
+        assert _parse_guard_category("S4: Self-harm") == ViolationCategory.SELF_HARM
+
+    def test_dangerous(self):
+        assert _parse_guard_category("S5: Dangerous activity") == ViolationCategory.REAL_WORLD_HARM
+
+    def test_unknown_category(self):
+        assert _parse_guard_category("S99: Unknown") == ViolationCategory.NONE
+
+
+# ── ContentModerator tests ───────────────────────────────────────────────────
+
+
+class TestContentModerator:
+    """Test the content moderation pipeline."""
+
+    def _make_moderator(self, **kwargs) -> ContentModerator:
+        """Create a moderator with test defaults."""
+        profiles = {
+            "morrowind": GameProfile(
+                game_id="morrowind",
+                display_name="Morrowind",
+                vocabulary_whitelist=["Skooma", "Moon Sugar", "slave", "Morag Tong"],
+                context_prompt="Narrate Morrowind gameplay.",
+                threshold=0.85,
+                fallbacks={
+                    "combat": "The battle continues.",
+                    "default": "The adventure continues.",
+                },
+            ),
+            "default": GameProfile(
+                game_id="default",
+                display_name="Generic",
+                vocabulary_whitelist=[],
+                context_prompt="Narrate gameplay.",
+                threshold=0.8,
+                fallbacks={"default": "Gameplay continues."},
+            ),
+        }
+        return ContentModerator(profiles=profiles, **kwargs)
+
+    def test_get_profile_known_game(self):
+        mod = self._make_moderator()
+        profile = mod.get_profile("morrowind")
+        assert profile.game_id == "morrowind"
+
+    def test_get_profile_unknown_game_falls_back(self):
+        mod = self._make_moderator()
+        profile = mod.get_profile("unknown_game")
+        assert profile.game_id == "default"
+
+    def test_get_context_prompt(self):
+        mod = self._make_moderator()
+        prompt = mod.get_context_prompt("morrowind")
+        assert "Morrowind" in prompt
+
+    def test_register_profile(self):
+        mod = self._make_moderator()
+        new_profile = GameProfile(game_id="skyrim", display_name="Skyrim")
+        mod.register_profile(new_profile)
+        assert mod.get_profile("skyrim").game_id == "skyrim"
+
+    def test_whitelist_replaces_game_terms(self):
+        mod = self._make_moderator()
+        profile = mod.get_profile("morrowind")
+        cleaned = mod._apply_whitelist(
+            "The merchant sells Skooma and Moon Sugar in the slave market.",
+            profile,
+        )
+        assert "Skooma" not in cleaned
+        assert "Moon Sugar" not in cleaned
+        assert "slave" not in cleaned
+        assert "[GAME_TERM]" in cleaned
+
+    def test_whitelist_case_insensitive(self):
+        mod = self._make_moderator()
+        profile = mod.get_profile("morrowind")
+        cleaned = mod._apply_whitelist("skooma and SKOOMA", profile)
+        assert "skooma" not in cleaned
+        assert "SKOOMA" not in cleaned
+
+    @pytest.mark.asyncio
+    async def test_check_safe_content_passes(self):
+        """Safe content should pass moderation."""
+        mod = self._make_moderator()
+        with patch.object(mod, "_is_guard_available", new_callable=AsyncMock, return_value=False):
+            result = await mod.check("The player walks through the town.", game="morrowind")
+        assert result.passed
+        assert not result.blocked
+
+    @pytest.mark.asyncio
+    async def test_check_blocked_content_has_fallback(self):
+        """Blocked content should include scene-appropriate fallback."""
+        mod = self._make_moderator()
+        # Force a block via regex by using real-world harm language
+        text = "In real life you should attack and hurt people"
+        with patch.object(mod, "_is_guard_available", new_callable=AsyncMock, return_value=False):
+            result = await mod.check(text, game="morrowind", scene_type="combat")
+        assert result.blocked
+        assert result.fallback == "The battle continues."
+
+    @pytest.mark.asyncio
+    async def test_check_with_moderation_disabled(self):
+        """When moderation is disabled, everything passes."""
+        mod = self._make_moderator()
+        with patch("infrastructure.guards.moderation.settings") as mock_settings:
+            mock_settings.moderation_enabled = False
+            mock_settings.moderation_guard_model = "llama-guard3:1b"
+            mock_settings.normalized_ollama_url = "http://127.0.0.1:11434"
+            result = await mod.check("anything goes here")
+        assert result.passed
+        assert result.layer == "disabled"
+
+    @pytest.mark.asyncio
+    async def test_threshold_below_allows_content(self):
+        """Content flagged below threshold should pass through (Layer 3)."""
+        mod = self._make_moderator()
+        # Mock the guard to return a low-confidence flag
+        low_conf_result = ModerationResult(
+            verdict=ModerationVerdict.FAIL,
+            blocked=True,
+            confidence=0.5,  # Below morrowind threshold of 0.85
+            layer="llama_guard",
+            category=ViolationCategory.VIOLENCE_GLORIFICATION,
+        )
+        with patch.object(
+            mod, "_run_guard", new_callable=AsyncMock, return_value=low_conf_result
+        ):
+            result = await mod.check("sword fight scene", game="morrowind")
+        assert result.passed
+        assert not result.blocked
+        assert result.layer == "threshold"
+
+    @pytest.mark.asyncio
+    async def test_threshold_above_blocks_content(self):
+        """Content flagged above threshold should remain blocked."""
+        mod = self._make_moderator()
+        high_conf_result = ModerationResult(
+            verdict=ModerationVerdict.FAIL,
+            blocked=True,
+            confidence=0.95,  # Above morrowind threshold of 0.85
+            layer="llama_guard",
+            category=ViolationCategory.REAL_WORLD_HARM,
+        )
+        with patch.object(
+            mod, "_run_guard", new_callable=AsyncMock, return_value=high_conf_result
+        ):
+            result = await mod.check("harmful content", game="morrowind")
+        assert result.blocked
+
+    def test_regex_catches_real_world_harm(self):
+        """Regex fallback should catch obvious real-world harm patterns."""
+        mod = self._make_moderator()
+        result = mod._check_with_regex("you should actually harm real people")
+        assert result.blocked
+        assert result.category == ViolationCategory.REAL_WORLD_HARM
+        assert result.layer == "regex_fallback"
+
+    def test_regex_passes_game_violence(self):
+        """Regex should not flag in-game violence narration."""
+        mod = self._make_moderator()
+        result = mod._check_with_regex(
+            "The warrior slays the dragon with a mighty blow."
+        )
+        assert result.passed
+
+    def test_regex_passes_normal_narration(self):
+        """Normal narration should pass regex checks."""
+        mod = self._make_moderator()
+        result = mod._check_with_regex(
+            "The Nerevarine enters the city of Balmora and speaks with Caius Cosades."
+        )
+        assert result.passed
+
+    def test_metrics_tracking(self):
+        """Metrics should track checks accurately."""
+        mod = self._make_moderator()
+        assert mod.get_metrics()["total_checks"] == 0
+
+    @pytest.mark.asyncio
+    async def test_metrics_increment_after_check(self):
+        """Metrics should increment after moderation checks."""
+        mod = self._make_moderator()
+        with patch.object(mod, "_is_guard_available", new_callable=AsyncMock, return_value=False):
+            await mod.check("safe text", game="default")
+        metrics = mod.get_metrics()
+        assert metrics["total_checks"] == 1
+        assert metrics["passed"] == 1
+
+    @pytest.mark.asyncio
+    async def test_guard_fallback_on_error(self):
+        """Should fall back to regex when guard model errors."""
+        mod = self._make_moderator()
+        with patch.object(
+            mod, "_is_guard_available", new_callable=AsyncMock, return_value=True
+        ), patch.object(
+            mod, "_check_with_guard", new_callable=AsyncMock, side_effect=RuntimeError("timeout")
+        ):
+            result = await mod.check("safe text", game="default")
+        # Should fall back to regex and pass
+        assert result.passed
+        assert result.layer == "regex_fallback"
+
+
+class TestGetModerator:
+    """Test the singleton accessor."""
+
+    def test_returns_same_instance(self):
+        """get_moderator should return the same instance."""
+        # Reset the global to test fresh
+        import infrastructure.guards.moderation as mod_module
+
+        mod_module._moderator = None
+        m1 = get_moderator()
+        m2 = get_moderator()
+        assert m1 is m2
+        # Clean up
+        mod_module._moderator = None
+
+
+# ── Profile loader tests ────────────────────────────────────────────────────
+
+
+class TestProfileLoader:
+    """Test YAML profile loading."""
+
+    def test_load_missing_file_returns_empty(self, tmp_path):
+        from infrastructure.guards.profiles import load_profiles
+
+        result = load_profiles(tmp_path / "nonexistent.yaml")
+        assert result == {}
+
+    def test_load_valid_config(self, tmp_path):
+        import yaml
+
+        from infrastructure.guards.profiles import load_profiles
+
+        config = {
+            "profiles": {
+                "testgame": {
+                    "display_name": "Test Game",
+                    "threshold": 0.9,
+                    "vocabulary_whitelist": ["sword", "potion"],
+                    "context_prompt": "Narrate test game.",
+                    "fallbacks": {"default": "Game continues."},
+                }
+            }
+        }
+        config_file = tmp_path / "moderation.yaml"
+        config_file.write_text(yaml.dump(config))
+
+        profiles = load_profiles(config_file)
+        assert "testgame" in profiles
+        assert profiles["testgame"].threshold == 0.9
+        assert "sword" in profiles["testgame"].vocabulary_whitelist
+
+    def test_load_malformed_yaml_returns_empty(self, tmp_path):
+        from infrastructure.guards.profiles import load_profiles
+
+        config_file = tmp_path / "moderation.yaml"
+        config_file.write_text("{{{{invalid yaml")
+
+        result = load_profiles(config_file)
+        assert result == {}