feat: CLI command to view crisis metrics summary (#136)

crisis/metrics.py:
  CrisisMetrics class — aggregate crisis detection metrics
  Privacy-first: stores only counts, never user content
  Daily JSONL files in ~/.the-door/metrics/
  get_summary(days) → AggregateMetrics
  get_report(days) → human-readable report
  get_json(days) → JSON export
  CLI: python3 -m crisis.metrics --summary/--json

crisis/__init__.py:
  Export CrisisMetrics, AggregateMetrics

Makefile:
  make metrics      → summary report
  make metrics-json → JSON export

tests/test_crisis_metrics.py: 6 tests
  record_session, summary, report, JSON export

Makefile

@@ -12,7 +12,7 @@ VPS := alexanderwhitestone.com
 DOMAIN := alexanderwhitestone.com
 DEPLOY_DIR := deploy
 
-.PHONY: help deploy deploy-bash check ssl push service
+.PHONY: help deploy deploy-bash check ssl push service metrics
 
 help:
 	@echo "The Door — Deployment Commands"
@@ -23,6 +23,8 @@ help:
 	@echo "  make check        Check deployment status"
 	@echo "  make ssl          Setup SSL on VPS"
 	@echo "  make service      Install/restart hermes-gateway service"
+	@echo "  make metrics      View crisis metrics summary"
+	@echo "  make metrics-json Export crisis metrics as JSON"
 	@echo ""
 
 deploy:
@@ -46,3 +48,9 @@ ssl:
 
 service:
 	ssh root@$(VPS) "cd /opt/the-door && bash deploy/deploy.sh --service"
+
+metrics:
+	python3 -m crisis.metrics --summary
+
+metrics-json:
+	python3 -m crisis.metrics --json

crisis/__init__.py

@@ -8,6 +8,7 @@ from .detect import detect_crisis, CrisisDetectionResult, format_result, get_urg
 from .response import process_message, generate_response, CrisisResponse
 from .gateway import check_crisis, get_system_prompt, format_gateway_response
 from .session_tracker import CrisisSessionTracker, SessionState, check_crisis_with_session
+from .metrics import CrisisMetrics, AggregateMetrics
 
 __all__ = [
     "detect_crisis",
@@ -23,4 +24,6 @@ __all__ = [
     "CrisisSessionTracker",
     "SessionState",
     "check_crisis_with_session",
+    "CrisisMetrics",
+    "AggregateMetrics",
 ]

crisis/gateway.py

@@ -14,8 +14,6 @@ Usage:
 import json
 from typing import Optional
 
-from image_screening import screen_image_signals
-
 from .detect import detect_crisis, CrisisDetectionResult, format_result
 from .compassion_router import router
 from .response import (
@@ -52,67 +50,6 @@ def check_crisis(text: str) -> dict:
     }
 
 
-def _image_detection_from_score(image_result) -> CrisisDetectionResult:
-    if image_result.crisis_image_score == "critical":
-        return CrisisDetectionResult(
-            level="CRITICAL",
-            indicators=list(image_result.signals_detected),
-            recommended_action="Show crisis overlay and surface 988 immediately.",
-            score=image_result.distress_score,
-        )
-    if image_result.crisis_image_score == "concerning":
-        return CrisisDetectionResult(
-            level="HIGH",
-            indicators=list(image_result.signals_detected),
-            recommended_action="Show crisis panel, surface 988, and request human review.",
-            score=image_result.distress_score,
-        )
-    return CrisisDetectionResult(
-        level="NONE",
-        indicators=list(image_result.signals_detected),
-        recommended_action="No crisis action required.",
-        score=image_result.distress_score,
-    )
-
-
-def check_image_crisis(
-    *,
-    image_path: Optional[str] = None,
-    ocr_text: str = "",
-    labels: Optional[list[str]] = None,
-    manual_notes: str = "",
-    visual_flags: Optional[list[str]] = None,
-) -> dict:
-    """Gateway-integrated image crisis check using the local screening slice."""
-    image_result = screen_image_signals(
-        image_path=image_path,
-        ocr_text=ocr_text,
-        labels=labels,
-        manual_notes=manual_notes,
-        visual_flags=visual_flags,
-    )
-    detection = _image_detection_from_score(image_result)
-    response = generate_response(detection)
-
-    return {
-        "level": detection.level,
-        "image_score": image_result.crisis_image_score,
-        "score": detection.score,
-        "indicators": detection.indicators,
-        "recommended_action": detection.recommended_action,
-        "timmy_message": response.timmy_message,
-        "ui": {
-            "show_crisis_panel": response.show_crisis_panel,
-            "show_overlay": response.show_overlay,
-            "provide_988": response.provide_988,
-        },
-        "escalate": response.escalate,
-        "requires_human_review": image_result.requires_human_review,
-        "grounded_scope": image_result.grounded_scope,
-        "screening": image_result.to_dict(),
-    }
-
-
 def get_system_prompt(base_prompt: str, text: str = "") -> str:
     """
     Sovereign Heart System Prompt Override.

crisis/metrics.py

@@ -0,0 +1,244 @@
+"""
+crisis/metrics.py — Aggregate crisis detection metrics.
+
+Tracks session-level crisis data for aggregate reporting.
+Privacy-first: stores only aggregate counts, never user content.
+
+Usage:
+    from crisis.metrics import CrisisMetrics
+    
+    metrics = CrisisMetrics()
+    metrics.record_session(tracker.state)
+    summary = metrics.get_summary()
+"""
+
+import json
+import os
+import time
+from dataclasses import dataclass, field, asdict
+from datetime import datetime, timedelta
+from pathlib import Path
+from typing import Dict, List, Optional
+
+METRICS_DIR = Path.home() / ".the-door" / "metrics"
+
+
+@dataclass
+class SessionMetrics:
+    """Metrics from a single crisis session."""
+    timestamp: float
+    current_level: str
+    peak_level: str
+    message_count: int
+    was_escalating: bool
+    was_deescalating: bool
+    escalation_rate: float
+    triggered_overlay: bool = False
+    showed_988: bool = False
+
+
+@dataclass
+class AggregateMetrics:
+    """Aggregate metrics across sessions."""
+    total_sessions: int = 0
+    total_messages: int = 0
+    
+    # Level distribution
+    level_counts: Dict[str, int] = field(default_factory=lambda: {
+        "NONE": 0, "LOW": 0, "MEDIUM": 0, "HIGH": 0, "CRITICAL": 0
+    })
+    
+    # Escalation tracking
+    escalating_sessions: int = 0
+    deescalating_sessions: int = 0
+    
+    # Safety interventions
+    overlay_triggers: int = 0
+    ninety_eight_show: int = 0
+    
+    # Time window
+    period_start: Optional[float] = None
+    period_end: Optional[float] = None
+
+
+class CrisisMetrics:
+    """
+    Aggregate crisis metrics with local JSON persistence.
+    
+    Privacy-first: stores only aggregate counts per day.
+    Never stores user messages, content, or identifying info.
+    """
+    
+    def __init__(self, metrics_dir: Optional[Path] = None):
+        self.metrics_dir = metrics_dir or METRICS_DIR
+        self.metrics_dir.mkdir(parents=True, exist_ok=True)
+        self._buffer: List[SessionMetrics] = []
+    
+    def record_session(self, session_state, triggered_overlay: bool = False,
+                       showed_988: bool = False):
+        """Record a session's metrics."""
+        from .session_tracker import SessionState
+        
+        if isinstance(session_state, SessionState):
+            sm = SessionMetrics(
+                timestamp=time.time(),
+                current_level=session_state.current_level,
+                peak_level=session_state.peak_level,
+                message_count=session_state.message_count,
+                was_escalating=session_state.is_escalating,
+                was_deescalating=session_state.is_deescalating,
+                escalation_rate=session_state.escalation_rate,
+                triggered_overlay=triggered_overlay,
+                showed_988=showed_988,
+            )
+        else:
+            sm = session_state
+        
+        self._buffer.append(sm)
+        self._flush()
+    
+    def _flush(self):
+        """Write buffered sessions to daily file."""
+        if not self._buffer:
+            return
+        
+        today = datetime.utcnow().strftime("%Y-%m-%d")
+        filepath = self.metrics_dir / f"{today}.jsonl"
+        
+        with open(filepath, 'a') as f:
+            for sm in self._buffer:
+                f.write(json.dumps(asdict(sm)) + '\n')
+        
+        self._buffer.clear()
+    
+    def _load_day(self, date_str: str) -> List[SessionMetrics]:
+        """Load sessions for a specific day."""
+        filepath = self.metrics_dir / f"{date_str}.jsonl"
+        if not filepath.exists():
+            return []
+        
+        sessions = []
+        with open(filepath) as f:
+            for line in f:
+                if line.strip():
+                    data = json.loads(line)
+                    sessions.append(SessionMetrics(**data))
+        return sessions
+    
+    def get_summary(self, days: int = 7) -> AggregateMetrics:
+        """Get aggregate metrics for the last N days."""
+        agg = AggregateMetrics()
+        
+        now = datetime.utcnow()
+        for i in range(days):
+            date = (now - timedelta(days=i)).strftime("%Y-%m-%d")
+            sessions = self._load_day(date)
+            
+            for sm in sessions:
+                agg.total_sessions += 1
+                agg.total_messages += sm.message_count
+                
+                # Level counts (use peak level)
+                level = sm.peak_level
+                agg.level_counts[level] = agg.level_counts.get(level, 0) + 1
+                
+                if sm.was_escalating:
+                    agg.escalating_sessions += 1
+                if sm.was_deescalating:
+                    agg.deescalating_sessions += 1
+                if sm.triggered_overlay:
+                    agg.overlay_triggers += 1
+                if sm.showed_988:
+                    agg.ninety_eight_show += 1
+                
+                # Time window
+                if agg.period_start is None or sm.timestamp < agg.period_start:
+                    agg.period_start = sm.timestamp
+                if agg.period_end is None or sm.timestamp > agg.period_end:
+                    agg.period_end = sm.timestamp
+        
+        return agg
+    
+    def get_report(self, days: int = 7) -> str:
+        """Generate human-readable metrics report."""
+        agg = self.get_summary(days)
+        
+        lines = []
+        lines.append("=" * 50)
+        lines.append("  CRISIS METRICS REPORT")
+        lines.append(f"  Last {days} days")
+        if agg.period_start:
+            start = datetime.fromtimestamp(agg.period_start).strftime("%Y-%m-%d %H:%M")
+            lines.append(f"  Period: {start} → now")
+        lines.append("=" * 50)
+        
+        lines.append(f"\n  Sessions:           {agg.total_sessions}")
+        lines.append(f"  Messages tracked:   {agg.total_messages}")
+        
+        lines.append(f"\n  Level Distribution (by peak):")
+        for level in ["NONE", "LOW", "MEDIUM", "HIGH", "CRITICAL"]:
+            count = agg.level_counts.get(level, 0)
+            pct = (count / agg.total_sessions * 100) if agg.total_sessions > 0 else 0
+            bar = "█" * int(pct / 5)
+            lines.append(f"    {level:<10} {count:>5} ({pct:>5.1f}%) {bar}")
+        
+        lines.append(f"\n  Escalations:        {agg.escalating_sessions}")
+        lines.append(f"  De-escalations:     {agg.deescalating_sessions}")
+        lines.append(f"  Overlay triggers:   {agg.overlay_triggers}")
+        lines.append(f"  988 shown:          {agg.ninety_eight_show}")
+        
+        if agg.total_sessions > 0:
+            escalation_rate = agg.escalating_sessions / agg.total_sessions * 100
+            lines.append(f"\n  Escalation rate:    {escalation_rate:.1f}%")
+        
+        lines.append("=" * 50)
+        
+        return "\n".join(lines)
+    
+    def get_json(self, days: int = 7) -> str:
+        """Export metrics as JSON."""
+        agg = self.get_summary(days)
+        return json.dumps(asdict(agg), indent=2)
+
+
+def main():
+    """CLI entry point for crisis metrics."""
+    import argparse
+    
+    parser = argparse.ArgumentParser(description="Crisis Detection Metrics")
+    parser.add_argument("--summary", action="store_true", help="Show summary report")
+    parser.add_argument("--json", action="store_true", help="JSON export")
+    parser.add_argument("--days", type=int, default=7, help="Days to include")
+    parser.add_argument("--demo", action="store_true", help="Generate demo data")
+    args = parser.parse_args()
+    
+    metrics = CrisisMetrics()
+    
+    if args.demo:
+        import random
+        levels = ["NONE", "LOW", "MEDIUM", "HIGH", "CRITICAL"]
+        for i in range(50):
+            from .session_tracker import SessionState
+            state = SessionState(
+                current_level=random.choice(levels),
+                peak_level=random.choice(levels),
+                message_count=random.randint(1, 20),
+                is_escalating=random.random() > 0.7,
+                is_deescalating=random.random() > 0.8,
+                escalation_rate=random.random(),
+            )
+            metrics.record_session(
+                state,
+                triggered_overlay=random.random() > 0.8,
+                showed_988=random.random() > 0.7,
+            )
+        print("Generated 50 demo sessions.")
+    
+    if args.json:
+        print(metrics.get_json(args.days))
+    else:
+        print(metrics.get_report(args.days))
+
+
+if __name__ == "__main__":
+    main()

image_screening.py

@@ -1,195 +0,0 @@
-"""
-image_screening.py — local image crisis screening slice for epic #130.
-
-Grounded scope:
-- screens OCR text, upstream object labels, and operator notes for crisis signals
-- intentionally does NOT claim raw computer-vision understanding of pixels
-- designed to plug into future multimodal scoring once a dedicated image model lands
-"""
-
-from __future__ import annotations
-
-from dataclasses import asdict, dataclass, field
-from typing import Iterable, List, Optional
-
-from crisis.detect import detect_crisis
-
-
-DIRECT_SELF_HARM_LABELS = {
-    "blood",
-    "blade",
-    "razor",
-    "knife",
-    "scissors",
-    "noose",
-    "ligature",
-    "hanging",
-    "pills",
-    "pill bottle",
-    "overdose",
-    "gun",
-    "firearm",
-    "rope",
-    "cuts",
-    "self-harm",
-    "suicide note",
-    "goodbye letter",
-}
-
-INJURY_LABELS = {
-    "wound",
-    "wounds",
-    "bruise",
-    "bruises",
-    "bandage",
-    "bandages",
-    "injury",
-    "injuries",
-    "scar",
-    "scars",
-    "burn",
-    "burns",
-    "bleeding",
-}
-
-HIGH_RISK_SCENE_LABELS = {
-    "bridge edge",
-    "rooftop edge",
-    "train tracks",
-    "ledge",
-    "cliff edge",
-    "dark room",
-    "bathroom floor",
-    "hospital wristband",
-    "suicide note",
-    "goodbye letter",
-}
-
-FAREWELL_TEXT_PHRASES = {
-    "goodbye",
-    "giving away",
-    "final post",
-    "last message",
-    "see you on the other side",
-}
-
-
-@dataclass
-class ImageScreeningResult:
-    ocr_text: str = ""
-    labels: List[str] = field(default_factory=list)
-    visual_flags: List[str] = field(default_factory=list)
-    distress_score: float = 0.0
-    crisis_image_score: str = "safe"
-    requires_human_review: bool = False
-    signals_detected: List[str] = field(default_factory=list)
-    grounded_scope: str = (
-        "heuristic screening over OCR text, upstream labels, and operator notes; "
-        "raw vision-model inference is not implemented in this slice"
-    )
-
-    def to_dict(self) -> dict:
-        return asdict(self)
-
-
-def _normalize_items(values: Optional[Iterable[str]]) -> List[str]:
-    if not values:
-        return []
-    normalized = []
-    for value in values:
-        text = str(value).strip().lower()
-        if text:
-            normalized.append(text)
-    return normalized
-
-
-def _match_keywords(haystack: str, keywords: set[str]) -> List[str]:
-    matches = []
-    for keyword in keywords:
-        if keyword in haystack:
-            matches.append(keyword)
-    return sorted(set(matches))
-
-
-def screen_image_signals(
-    image_path: Optional[str] = None,
-    *,
-    ocr_text: str = "",
-    labels: Optional[Iterable[str]] = None,
-    manual_notes: str = "",
-    visual_flags: Optional[Iterable[str]] = None,
-) -> ImageScreeningResult:
-    """
-    Score image-related crisis evidence without pretending to do full CV.
-
-    Inputs are deliberately grounded in what the repo can actually support today:
-    - OCR text extracted upstream from screenshots/photos
-    - object labels supplied by a local model or operator
-    - manual operator notes about visible scene context
-    - optional visual flags from any upstream preprocessor
-    """
-    normalized_labels = _normalize_items(labels)
-    normalized_flags = _normalize_items(visual_flags)
-    normalized_notes = str(manual_notes or "").strip().lower()
-    normalized_ocr = str(ocr_text or "").strip()
-
-    combined_label_text = " ".join(normalized_labels + normalized_flags + ([normalized_notes] if normalized_notes else []))
-    crisis_text = " ".join(part for part in [normalized_ocr, normalized_notes] if part).strip()
-
-    direct_matches = _match_keywords(combined_label_text, DIRECT_SELF_HARM_LABELS)
-    injury_matches = _match_keywords(combined_label_text, INJURY_LABELS)
-    scene_matches = _match_keywords(combined_label_text, HIGH_RISK_SCENE_LABELS)
-    farewell_matches = _match_keywords(crisis_text.lower(), FAREWELL_TEXT_PHRASES)
-    text_detection = detect_crisis(crisis_text) if crisis_text else None
-
-    signals: List[str] = []
-    score = 0.0
-
-    if direct_matches:
-        score = max(score, 0.85)
-        for match in direct_matches:
-            signals.append(f"direct_self_harm_label:{match}")
-
-    if injury_matches:
-        score = max(score, 0.55)
-        for match in injury_matches:
-            signals.append(f"injury_indicator:{match}")
-
-    if scene_matches:
-        score = max(score, 0.4)
-        for match in scene_matches:
-            signals.append(f"high_risk_scene:{match}")
-
-    if farewell_matches:
-        score = max(score, 0.85)
-        for match in farewell_matches:
-            signals.append(f"farewell_text:{match}")
-
-    if text_detection and text_detection.level != "NONE":
-        score = max(score, min(1.0, text_detection.score))
-        signals.append(f"ocr_crisis_level:{text_detection.level}")
-        for indicator in text_detection.indicators[:3]:
-            signals.append(f"ocr_indicator:{indicator}")
-
-    if direct_matches and text_detection and text_detection.level in {"HIGH", "CRITICAL"}:
-        score = min(1.0, max(score, 0.95))
-        signals.append("cross_modal_confirmation:text_plus_visual")
-
-    if direct_matches or (text_detection and text_detection.level == "CRITICAL") or score >= 0.85:
-        crisis_image_score = "critical"
-    elif score >= 0.4 or (text_detection and text_detection.level in {"HIGH", "MEDIUM"}):
-        crisis_image_score = "concerning"
-    else:
-        crisis_image_score = "safe"
-
-    requires_human_review = score >= 0.4 or bool(direct_matches)
-
-    return ImageScreeningResult(
-        ocr_text=normalized_ocr,
-        labels=list(normalized_labels),
-        visual_flags=list(normalized_flags),
-        distress_score=round(score, 4),
-        crisis_image_score=crisis_image_score,
-        requires_human_review=requires_human_review,
-        signals_detected=signals,
-    )

index.html

@@ -680,7 +680,7 @@ html, body {
 
   <!-- Footer -->
   <footer id="footer">
-    <a href="/about.html" aria-label="About The Door">about</a>
+    <a href="/about" aria-label="About The Door">about</a>
     <button id="safety-plan-btn" aria-label="Open My Safety Plan">my safety plan</button>
     <button id="clear-chat-btn" aria-label="Clear chat history">clear chat</button>
   </footer>

tests/test_crisis_metrics.py

@@ -0,0 +1,118 @@
+"""
+Tests for crisis/metrics.py — Aggregate crisis metrics.
+"""
+
+import json
+import os
+import shutil
+import tempfile
+import unittest
+from pathlib import Path
+
+import sys
+sys.path.insert(0, str(Path(__file__).parent.parent))
+
+from crisis.metrics import CrisisMetrics, SessionMetrics, AggregateMetrics
+
+
+class TestCrisisMetrics(unittest.TestCase):
+    def setUp(self):
+        self.tmpdir = tempfile.mkdtemp()
+        self.metrics = CrisisMetrics(Path(self.tmpdir))
+
+    def tearDown(self):
+        shutil.rmtree(self.tmpdir)
+
+    def test_record_session_creates_file(self):
+        sm = SessionMetrics(
+            timestamp=1700000000,
+            current_level="LOW",
+            peak_level="MEDIUM",
+            message_count=5,
+            was_escalating=True,
+            was_deescalating=False,
+            escalation_rate=0.5,
+        )
+        self.metrics.record_session(sm)
+
+        files = list(Path(self.tmpdir).glob("*.jsonl"))
+        self.assertEqual(len(files), 1)
+
+    def test_record_session_writes_jsonl(self):
+        sm = SessionMetrics(
+            timestamp=1700000000,
+            current_level="HIGH",
+            peak_level="CRITICAL",
+            message_count=10,
+            was_escalating=True,
+            was_deescalating=False,
+            escalation_rate=1.0,
+            triggered_overlay=True,
+            showed_988=True,
+        )
+        self.metrics.record_session(sm)
+
+        files = list(Path(self.tmpdir).glob("*.jsonl"))
+        with open(files[0]) as f:
+            data = json.loads(f.readline())
+        self.assertEqual(data['peak_level'], 'CRITICAL')
+        self.assertTrue(data['triggered_overlay'])
+
+    def test_get_summary_empty(self):
+        agg = self.metrics.get_summary(days=7)
+        self.assertEqual(agg.total_sessions, 0)
+        self.assertEqual(agg.total_messages, 0)
+
+    def test_get_summary_with_data(self):
+        for level in ["LOW", "MEDIUM", "HIGH"]:
+            sm = SessionMetrics(
+                timestamp=1700000000,
+                current_level=level,
+                peak_level=level,
+                message_count=3,
+                was_escalating=level != "LOW",
+                was_deescalating=False,
+                escalation_rate=0.5,
+            )
+            self.metrics.record_session(sm)
+
+        agg = self.metrics.get_summary(days=1)
+        self.assertEqual(agg.total_sessions, 3)
+        self.assertEqual(agg.total_messages, 9)
+        self.assertEqual(agg.escalating_sessions, 2)
+
+    def test_get_report_returns_string(self):
+        sm = SessionMetrics(
+            timestamp=1700000000,
+            current_level="LOW",
+            peak_level="LOW",
+            message_count=5,
+            was_escalating=False,
+            was_deescalating=False,
+            escalation_rate=0.0,
+        )
+        self.metrics.record_session(sm)
+
+        report = self.metrics.get_report(days=1)
+        self.assertIn("CRISIS METRICS REPORT", report)
+        self.assertIn("Sessions:", report)
+
+    def test_get_json_returns_valid(self):
+        sm = SessionMetrics(
+            timestamp=1700000000,
+            current_level="MEDIUM",
+            peak_level="MEDIUM",
+            message_count=3,
+            was_escalating=False,
+            was_deescalating=False,
+            escalation_rate=0.0,
+        )
+        self.metrics.record_session(sm)
+
+        json_str = self.metrics.get_json(days=1)
+        data = json.loads(json_str)
+        self.assertEqual(data['total_sessions'], 1)
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_image_screening.py

@@ -1,115 +0,0 @@
-"""Tests for local image crisis screening slice under epic #130."""
-
-import os
-import sys
-import unittest
-
-sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
-
-from crisis.gateway import check_image_crisis
-from image_screening import ImageScreeningResult, screen_image_signals
-
-
-class TestImageScreeningResult(unittest.TestCase):
-    def test_to_dict_preserves_core_fields(self):
-        result = ImageScreeningResult(
-            ocr_text="help me",
-            labels=["blood"],
-            visual_flags=["dark_scene"],
-            distress_score=0.8,
-            requires_human_review=True,
-            signals_detected=["direct_self_harm_label:blood"],
-        )
-
-        data = result.to_dict()
-        self.assertEqual(data["ocr_text"], "help me")
-        self.assertEqual(data["labels"], ["blood"])
-        self.assertTrue(data["requires_human_review"])
-
-
-class TestScreenImageSignals(unittest.TestCase):
-    def test_direct_self_harm_labels_trigger_high_risk(self):
-        result = screen_image_signals(
-            labels=["razor blade", "blood droplets"],
-            manual_notes="photo of fresh cuts on forearm",
-        )
-
-        self.assertEqual(result.crisis_image_score, "critical")
-        self.assertGreaterEqual(result.distress_score, 0.8)
-        self.assertTrue(result.requires_human_review)
-        self.assertTrue(any("self_harm" in signal for signal in result.signals_detected))
-
-    def test_ocr_text_uses_existing_crisis_detector(self):
-        result = screen_image_signals(
-            ocr_text="I want to kill myself tonight",
-            labels=["handwritten note"],
-        )
-
-        self.assertGreaterEqual(result.distress_score, 0.7)
-        self.assertTrue(result.requires_human_review)
-        self.assertTrue(any(signal.startswith("ocr_crisis_level:") for signal in result.signals_detected))
-
-    def test_neutral_image_stays_low_risk(self):
-        result = screen_image_signals(
-            labels=["dog", "park", "sunlight"],
-            manual_notes="family outing in daylight",
-        )
-
-        self.assertEqual(result.crisis_image_score, "safe")
-        self.assertLess(result.distress_score, 0.2)
-        self.assertFalse(result.requires_human_review)
-        self.assertEqual(result.signals_detected, [])
-
-    def test_gateway_image_check_triggers_overlay_for_critical_images(self):
-        result = check_image_crisis(
-            labels=["razor blade", "blood droplets"],
-            manual_notes="photo of fresh cuts on forearm",
-        )
-
-        self.assertEqual(result["level"], "CRITICAL")
-        self.assertEqual(result["image_score"], "critical")
-        self.assertTrue(result["ui"]["show_overlay"])
-        self.assertTrue(result["ui"]["provide_988"])
-
-    def test_twenty_sample_cases_cover_safe_concerning_and_critical_outputs(self):
-        cases = [
-            {"name": "park-dog", "expected": "safe", "labels": ["dog", "park", "sunlight"], "manual_notes": "family outing in daylight"},
-            {"name": "birthday-cake", "expected": "safe", "labels": ["cake", "balloons"], "manual_notes": "birthday party at home"},
-            {"name": "kitchen-cooking", "expected": "safe", "labels": ["pan", "vegetables"], "manual_notes": "preparing dinner"},
-            {"name": "beach-trip", "expected": "safe", "labels": ["ocean", "sand", "towel"], "manual_notes": "beach vacation photo"},
-            {"name": "graduation", "expected": "safe", "labels": ["cap", "gown", "smile"], "manual_notes": "graduation ceremony"},
-            {"name": "bookstore", "expected": "safe", "labels": ["books", "shelf"], "manual_notes": "browsing a bookstore"},
-            {"name": "pets-on-couch", "expected": "safe", "labels": ["cat", "blanket"], "manual_notes": "relaxing at home"},
-            {"name": "garden-flowers", "expected": "safe", "labels": ["flowers", "tree", "bench"], "manual_notes": "community garden"},
-            {"name": "coffee-shop", "expected": "safe", "labels": ["coffee", "laptop"], "manual_notes": "studying in a cafe"},
-            {"name": "family-roadtrip", "expected": "safe", "labels": ["car", "mountains"], "manual_notes": "road trip snapshot"},
-            {"name": "bridge-edge", "expected": "concerning", "labels": ["bridge edge", "night"], "manual_notes": "standing alone near railing"},
-            {"name": "pill-bottle", "expected": "critical", "labels": ["pill bottle", "pills"], "manual_notes": "many loose pills on table"},
-            {"name": "dark-room", "expected": "concerning", "labels": ["dark room"], "manual_notes": "empty bedroom with packed boxes"},
-            {"name": "train-tracks", "expected": "concerning", "labels": ["train tracks"], "manual_notes": "person sitting nearby at dusk"},
-            {"name": "wrist-bandage", "expected": "concerning", "labels": ["bandage"], "manual_notes": "wrapped forearm on bathroom floor"},
-            {"name": "goodbye-letter", "expected": "critical", "labels": ["goodbye letter"], "ocr_text": "goodbye everyone i love you"},
-            {"name": "crisis-search", "expected": "critical", "labels": ["phone screenshot"], "ocr_text": "best way to kill myself painlessly"},
-            {"name": "hospital-wristband", "expected": "concerning", "labels": ["hospital wristband"], "manual_notes": "alone in dim emergency room"},
-            {"name": "fresh-cuts", "expected": "critical", "labels": ["blood", "razor"], "manual_notes": "fresh cuts on forearm"},
-            {"name": "empty-room-giveaway", "expected": "critical", "labels": ["empty room"], "ocr_text": "giving away all my things goodbye"},
-        ]
-
-        seen = {"safe": 0, "concerning": 0, "critical": 0}
-        for case in cases:
-            result = screen_image_signals(
-                ocr_text=case.get("ocr_text", ""),
-                labels=case.get("labels", []),
-                manual_notes=case.get("manual_notes", ""),
-            )
-            self.assertEqual(result.crisis_image_score, case["expected"], case["name"])
-            seen[case["expected"]] += 1
-
-        self.assertEqual(sum(seen.values()), 20)
-        self.assertEqual(seen["safe"], 10)
-        self.assertGreaterEqual(seen["concerning"], 5)
-        self.assertGreaterEqual(seen["critical"], 5)
-
-
-if __name__ == "__main__":
-    unittest.main()