feat: integrate image crisis screening gateway (#130 #132)

crisis/gateway.py

@@ -14,6 +14,8 @@ 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 (
@@ -50,6 +52,67 @@ 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.

evolution/crisis_synthesizer.py

@@ -1,195 +1 @@
-"""Crisis synthesizer — learn from anonymized crisis interactions.
-
-This is deliberately simple and privacy-preserving. It does not train a model or
-modify detection rules automatically. It only logs metadata, summarizes patterns,
-and suggests human-reviewed keyword weight adjustments.
-"""
-
-from __future__ import annotations
-
-import argparse
-import json
-import time
-from collections import Counter, defaultdict
-from pathlib import Path
-from typing import Iterable
-
-DEFAULT_LOG_PATH = Path.home() / ".the-door" / "crisis-interactions.jsonl"
-LEVELS = ("NONE", "LOW", "MEDIUM", "HIGH", "CRITICAL")
-
-
-def build_interaction_event(
-    level: str,
-    indicators: list[str],
-    response_given: str,
-    continued_conversation: bool,
-    false_positive: bool,
-    *,
-    now: float | None = None,
-) -> dict:
-    return {
-        "timestamp": float(time.time() if now is None else now),
-        "level": level,
-        "indicators": list(indicators),
-        "indicator_count": len(indicators),
-        "response_given": response_given,
-        "continued_conversation": bool(continued_conversation),
-        "false_positive": bool(false_positive),
-    }
-
-
-def append_interaction_event(
-    log_path: str | Path,
-    *,
-    level: str,
-    indicators: list[str],
-    response_given: str,
-    continued_conversation: bool,
-    false_positive: bool,
-    now: float | None = None,
-) -> dict:
-    event = build_interaction_event(
-        level,
-        indicators,
-        response_given,
-        continued_conversation,
-        false_positive,
-        now=now,
-    )
-    path = Path(log_path)
-    path.parent.mkdir(parents=True, exist_ok=True)
-    with path.open("a", encoding="utf-8") as handle:
-        handle.write(json.dumps(event) + "\n")
-    return event
-
-
-def load_interaction_events(log_path: str | Path) -> list[dict]:
-    path = Path(log_path)
-    if not path.exists():
-        return []
-    events = []
-    for line in path.read_text(encoding="utf-8").splitlines():
-        if not line.strip():
-            continue
-        events.append(json.loads(line))
-    return events
-
-
-def summarize_keywords(events: Iterable[dict]) -> list[dict]:
-    counts: Counter[str] = Counter()
-    for event in events:
-        counts.update(event.get("indicators", []))
-    return [{"keyword": keyword, "count": count} for keyword, count in counts.most_common(10)]
-
-
-def suggest_keyword_adjustments(events: Iterable[dict], *, min_observations: int = 5) -> list[dict]:
-    stats: dict[str, dict[str, int]] = defaultdict(lambda: {
-        "observations": 0,
-        "true_positive_count": 0,
-        "false_positive_count": 0,
-        "continued_conversation_count": 0,
-    })
-
-    for event in events:
-        for keyword in event.get("indicators", []):
-            bucket = stats[keyword]
-            bucket["observations"] += 1
-            if event.get("false_positive"):
-                bucket["false_positive_count"] += 1
-            else:
-                bucket["true_positive_count"] += 1
-            if event.get("continued_conversation"):
-                bucket["continued_conversation_count"] += 1
-
-    suggestions = []
-    for keyword, bucket in sorted(stats.items()):
-        if bucket["observations"] < min_observations:
-            continue
-        fp = bucket["false_positive_count"]
-        tp = bucket["true_positive_count"]
-        if fp >= min_observations and tp == 0:
-            adjustment = "lower_weight"
-            rationale = "Observed only false positives across the sample window."
-        elif tp >= min_observations and fp == 0:
-            adjustment = "raise_weight"
-            rationale = "Observed repeated genuine crises with no false positives."
-        else:
-            adjustment = "observe"
-            rationale = "Mixed evidence; keep monitoring before changing weights."
-        suggestions.append(
-            {
-                "keyword": keyword,
-                **bucket,
-                "suggested_adjustment": adjustment,
-                "rationale": rationale,
-            }
-        )
-    return suggestions
-
-
-def build_weekly_report(
-    events: Iterable[dict],
-    *,
-    now: float | None = None,
-    window_days: int = 7,
-    min_observations: int = 3,
-) -> dict:
-    current_time = float(time.time() if now is None else now)
-    cutoff = current_time - (window_days * 86400)
-    filtered = [event for event in events if float(event.get("timestamp", 0)) >= cutoff]
-
-    detections_per_level = {level: 0 for level in LEVELS}
-    detected_events = []
-    continued_after_intervention = 0
-    for event in filtered:
-        level = event.get("level", "NONE")
-        detections_per_level[level] = detections_per_level.get(level, 0) + 1
-        if level != "NONE":
-            detected_events.append(event)
-            if event.get("continued_conversation"):
-                continued_after_intervention += 1
-
-    false_positive_count = sum(1 for event in detected_events if event.get("false_positive"))
-    false_positive_estimate = false_positive_count / len(detected_events) if detected_events else 0.0
-
-    return {
-        "window_days": window_days,
-        "total_events": len(filtered),
-        "detections_per_level": detections_per_level,
-        "most_common_keywords": summarize_keywords(filtered),
-        "false_positive_estimate": false_positive_estimate,
-        "continued_after_intervention": continued_after_intervention,
-        "keyword_weight_suggestions": suggest_keyword_adjustments(filtered, min_observations=min_observations),
-    }
-
-
-def render_weekly_report(summary: dict) -> str:
-    return json.dumps(summary, indent=2)
-
-
-def write_weekly_report(output_path: str | Path, summary: dict) -> Path:
-    path = Path(output_path)
-    path.parent.mkdir(parents=True, exist_ok=True)
-    path.write_text(render_weekly_report(summary) + "\n", encoding="utf-8")
-    return path
-
-
-def main(argv: list[str] | None = None) -> int:
-    parser = argparse.ArgumentParser(description="Summarize anonymized crisis interactions")
-    parser.add_argument("--log-path", default=str(DEFAULT_LOG_PATH), help="JSONL crisis interaction log")
-    parser.add_argument("--days", type=int, default=7, help="Lookback window in days")
-    parser.add_argument("--min-observations", type=int, default=3, help="Minimum observations before suggesting keyword adjustments")
-    parser.add_argument("--output", help="Optional file to write the weekly report JSON")
-    args = parser.parse_args(argv)
-
-    events = load_interaction_events(args.log_path)
-    summary = build_weekly_report(events, window_days=args.days, min_observations=args.min_observations)
-    rendered = render_weekly_report(summary)
-    print(rendered)
-    if args.output:
-        write_weekly_report(args.output, summary)
-    return 0
-
-
-if __name__ == "__main__":
-    raise SystemExit(main())
+...

image_screening.py

@@ -0,0 +1,195 @@
+"""
+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,
+    )

tests/test_crisis_synthesizer.py

@@ -1,111 +0,0 @@
-"""Tests for evolution/crisis_synthesizer.py (issue #36)."""
-
-from __future__ import annotations
-
-import importlib.util
-import json
-import pathlib
-import sys
-import tempfile
-import unittest
-
-ROOT = pathlib.Path(__file__).resolve().parents[1]
-SCRIPT = ROOT / 'evolution' / 'crisis_synthesizer.py'
-
-spec = importlib.util.spec_from_file_location('crisis_synthesizer', str(SCRIPT))
-mod = importlib.util.module_from_spec(spec)
-sys.modules['crisis_synthesizer'] = mod
-spec.loader.exec_module(mod)
-
-
-class TestCrisisSynthesizerEvent(unittest.TestCase):
-    def test_build_interaction_event_is_privacy_preserving(self):
-        event = mod.build_interaction_event(
-            level='CRITICAL',
-            indicators=['want_to_die', 'no_way_out'],
-            response_given='guardian',
-            continued_conversation=True,
-            false_positive=False,
-            now=1700000000,
-        )
-        self.assertEqual(event['timestamp'], 1700000000)
-        self.assertEqual(event['level'], 'CRITICAL')
-        self.assertEqual(event['response_given'], 'guardian')
-        self.assertTrue(event['continued_conversation'])
-        self.assertFalse(event['false_positive'])
-        self.assertEqual(event['indicators'], ['want_to_die', 'no_way_out'])
-        for forbidden in ['text', 'message', 'content', 'ip', 'session_id', 'user_id']:
-            self.assertNotIn(forbidden, event)
-
-
-class TestCrisisSynthesizerStorage(unittest.TestCase):
-    def test_append_and_load_events_round_trip(self):
-        with tempfile.TemporaryDirectory() as tmp:
-            log_path = pathlib.Path(tmp) / 'crisis-events.jsonl'
-            mod.append_interaction_event(
-                log_path,
-                level='HIGH',
-                indicators=['hopeless'],
-                response_given='companion',
-                continued_conversation=False,
-                false_positive=True,
-                now=1700000100,
-            )
-            events = mod.load_interaction_events(log_path)
-            self.assertEqual(len(events), 1)
-            self.assertEqual(events[0]['level'], 'HIGH')
-            self.assertEqual(events[0]['indicators'], ['hopeless'])
-
-
-class TestCrisisSynthesizerSummary(unittest.TestCase):
-    def test_weekly_report_contains_required_metrics(self):
-        events = [
-            mod.build_interaction_event('CRITICAL', ['want_to_die'], 'guardian', True, False, now=1700000000),
-            mod.build_interaction_event('HIGH', ['hopeless'], 'companion', False, True, now=1700000100),
-            mod.build_interaction_event('LOW', ['rough_day'], 'friend', False, False, now=1700000200),
-            mod.build_interaction_event('CRITICAL', ['want_to_die'], 'guardian', False, False, now=1700000300),
-            mod.build_interaction_event('NONE', [], 'friend', False, False, now=1700000400),
-        ]
-        summary = mod.build_weekly_report(events, now=1700000500, window_days=7)
-        self.assertEqual(summary['detections_per_level']['CRITICAL'], 2)
-        self.assertEqual(summary['detections_per_level']['HIGH'], 1)
-        self.assertEqual(summary['detections_per_level']['LOW'], 1)
-        self.assertEqual(summary['detections_per_level']['NONE'], 1)
-        self.assertEqual(summary['continued_after_intervention'], 1)
-        self.assertAlmostEqual(summary['false_positive_estimate'], 0.25)
-        self.assertEqual(summary['most_common_keywords'][0]['keyword'], 'want_to_die')
-        self.assertEqual(summary['most_common_keywords'][0]['count'], 2)
-
-
-class TestCrisisSynthesizerSuggestions(unittest.TestCase):
-    def test_suggests_weight_adjustments_from_interactions(self):
-        events = []
-        for ts in range(3):
-            events.append(mod.build_interaction_event('CRITICAL', ['want_to_die'], 'guardian', True, False, now=1700000000 + ts))
-        for ts in range(3):
-            events.append(mod.build_interaction_event('LOW', ['rough_day'], 'friend', False, True, now=1700000100 + ts))
-        suggestions = mod.suggest_keyword_adjustments(events, min_observations=3)
-        by_keyword = {s['keyword']: s for s in suggestions}
-        self.assertEqual(by_keyword['want_to_die']['suggested_adjustment'], 'raise_weight')
-        self.assertEqual(by_keyword['rough_day']['suggested_adjustment'], 'lower_weight')
-
-
-class TestCrisisSynthesizerRendering(unittest.TestCase):
-    def test_render_weekly_report_outputs_json(self):
-        summary = {
-            'detections_per_level': {'NONE': 0, 'LOW': 1, 'MEDIUM': 0, 'HIGH': 0, 'CRITICAL': 0},
-            'most_common_keywords': [{'keyword': 'rough_day', 'count': 1}],
-            'false_positive_estimate': 0.0,
-            'continued_after_intervention': 0,
-            'keyword_weight_suggestions': [],
-            'window_days': 7,
-            'total_events': 1,
-        }
-        rendered = mod.render_weekly_report(summary)
-        parsed = json.loads(rendered)
-        self.assertEqual(parsed['window_days'], 7)
-        self.assertEqual(parsed['most_common_keywords'][0]['keyword'], 'rough_day')
-
-
-if __name__ == '__main__':
-    unittest.main()

tests/test_image_screening.py

@@ -0,0 +1,115 @@
+"""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()