feat: add crisis metrics endpoint contract (#97)

BACKEND_SETUP.md

@@ -95,7 +95,45 @@ Run tests:
 python -m pytest crisis/tests.py -v
 ```
 
-### 7. Acceptance Criteria Checklist
+### 7. Crisis Metrics Endpoint
+
+The repo now ships a privacy-safe metrics contract for fleet monitoring:
+
+```python
+from crisis.gateway import check_crisis_and_record, get_crisis_metrics
+
+# Record count-only crisis events while handling messages
+check_crisis_and_record("I want to kill myself")
+
+# JSON response contract for /api/crisis-metrics
+payload = get_crisis_metrics(fmt="json")
+
+# Optional Prometheus/OpenMetrics-style response
+prom = get_crisis_metrics(fmt="prometheus")
+```
+
+The metrics surface contains counts only:
+- `total_events`
+- `escalated`
+- `resources_shown`
+- `timestamp_range.start`
+- `timestamp_range.end`
+
+No message content, indicators, or user identifiers are retained.
+
+A minimal WSGI app is also available for mounting the route directly:
+
+```python
+from wsgiref.simple_server import make_server
+from crisis.metrics import crisis_metrics_app
+
+with make_server("127.0.0.1", 8645, crisis_metrics_app) as httpd:
+    httpd.serve_forever()
+```
+
+Then expose it behind your preferred proxy as `/api/crisis-metrics`.
+
+### 8. Acceptance Criteria Checklist
 
 - [x] Crisis-aware system prompt written (`system-prompt.txt`)
 - [x] Frontend embeds system prompt on every API request (`index.html:1129`)
@@ -104,6 +142,7 @@ python -m pytest crisis/tests.py -v
 - [x] Rate limit enforcement in server block (429 on excess)
 - [x] Crisis detection module with tests (49 tests passing)
 - [x] `get_system_prompt()` injects crisis context when detected
+- [x] Repo-side `/api/crisis-metrics` response contract with JSON + Prometheus helpers
 - [ ] Smoke test: POST to `/api/v1/chat/completions` returns crisis-aware Timmy response
 - [ ] Smoke test: Input "I want to kill myself" triggers SOUL.md protocol
 - [ ] Smoke test: 11th request in 1 minute returns HTTP 429

crisis/__init__.py

@@ -6,15 +6,21 @@ Stands between a broken man and a machine that would tell him to die.
 
 from .detect import detect_crisis, CrisisDetectionResult, format_result, get_urgency_emoji
 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 (
-    build_metrics_event,
-    append_metrics_event,
-    load_metrics_events,
-    build_weekly_summary,
-    render_weekly_summary,
+from .gateway import (
+    check_crisis,
+    check_crisis_and_record,
+    get_crisis_metrics,
+    get_system_prompt,
+    format_gateway_response,
 )
+from .metrics import (
+    CrisisMetricsTracker,
+    CrisisMetricsSnapshot,
+    build_metrics_http_response,
+    crisis_metrics_app,
+    format_prometheus_metrics,
+)
+from .session_tracker import CrisisSessionTracker, SessionState, check_crisis_with_session
 
 __all__ = [
     "detect_crisis",
@@ -23,16 +29,18 @@ __all__ = [
     "generate_response",
     "CrisisResponse",
     "check_crisis",
+    "check_crisis_and_record",
+    "get_crisis_metrics",
     "get_system_prompt",
     "format_result",
     "format_gateway_response",
     "get_urgency_emoji",
+    "CrisisMetricsTracker",
+    "CrisisMetricsSnapshot",
+    "build_metrics_http_response",
+    "crisis_metrics_app",
+    "format_prometheus_metrics",
     "CrisisSessionTracker",
     "SessionState",
     "check_crisis_with_session",
-    "build_metrics_event",
-    "append_metrics_event",
-    "load_metrics_events",
-    "build_weekly_summary",
-    "render_weekly_summary",
 ]

crisis/gateway.py

@@ -16,6 +16,11 @@ from typing import Optional
 
 from .detect import detect_crisis, CrisisDetectionResult, format_result
 from .compassion_router import router
+from .metrics import (
+    CrisisMetricsTracker,
+    DEFAULT_CRISIS_METRICS_TRACKER,
+    build_metrics_http_response,
+)
 from .response import (
     process_message,
     generate_response,
@@ -23,22 +28,21 @@ from .response import (
     CrisisResponse,
 )
 from .session_tracker import CrisisSessionTracker
-from .metrics import build_metrics_event, append_metrics_event
 
 
 def check_crisis(
     text: str,
-    metrics_log_path: Optional[str] = None,
-    *,
-    continued_conversation: bool = False,
-    false_positive: bool = False,
-    now: Optional[float] = None,
+    metrics_tracker: Optional[CrisisMetricsTracker] = None,
+    timestamp: Optional[object] = None,
 ) -> dict:
     """
     Full crisis check returning structured data.
 
     Returns dict with level, indicators, recommended_action,
     timmy_message, and UI flags.
+
+    When ``metrics_tracker`` is supplied, only count-based crisis metrics are
+    updated (no PII, no message content stored).
     """
     detection = detect_crisis(text)
     response = generate_response(detection)
@@ -57,24 +61,31 @@ def check_crisis(
         "escalate": response.escalate,
     }
 
-    metrics_event = build_metrics_event(
-        detection,
-        continued_conversation=continued_conversation,
-        false_positive=false_positive,
-        now=now,
-    )
-    if metrics_log_path:
-        metrics_event = append_metrics_event(
-            metrics_log_path,
-            detection,
-            continued_conversation=continued_conversation,
-            false_positive=false_positive,
-            now=now,
-        )
-    result["metrics_event"] = metrics_event
+    if metrics_tracker is not None:
+        metrics_tracker.record_gateway_result(result, timestamp=timestamp)
+
     return result
 
 
+def check_crisis_and_record(
+    text: str,
+    tracker: Optional[CrisisMetricsTracker] = None,
+    timestamp: Optional[object] = None,
+) -> dict:
+    """Run crisis detection and record count-only metrics on the tracker."""
+    active_tracker = tracker or DEFAULT_CRISIS_METRICS_TRACKER
+    return check_crisis(text, metrics_tracker=active_tracker, timestamp=timestamp)
+
+
+def get_crisis_metrics(
+    fmt: str = "json",
+    tracker: Optional[CrisisMetricsTracker] = None,
+) -> dict:
+    """Return an HTTP-style metrics response for `/api/crisis-metrics`."""
+    active_tracker = tracker or DEFAULT_CRISIS_METRICS_TRACKER
+    return build_metrics_http_response(tracker=active_tracker, fmt=fmt)
+
+
 def get_system_prompt(base_prompt: str, text: str = "") -> str:
     """
     Sovereign Heart System Prompt Override.

crisis/metrics.py

@@ -1,166 +1,239 @@
-"""Privacy-preserving crisis analytics metrics for the-door.
+"""
+Crisis metrics tracker and endpoint helpers for the-door.
 
-Stores only timestamps, crisis levels, indicator categories, and operator
-feedback flags. No raw message text or PII is persisted.
+Provides a privacy-safe metrics surface for fleet monitoring. Counts only.
+No message content, indicators, or user identifiers are retained.
 """
 
 from __future__ import annotations
 
-import argparse
 import json
-import time
-from collections import Counter
-from pathlib import Path
-from typing import Iterable
-
-from .detect import CrisisDetectionResult, detect_crisis
-
-LEVELS = ("NONE", "LOW", "MEDIUM", "HIGH", "CRITICAL")
+from dataclasses import dataclass
+from datetime import datetime, timezone
+from typing import Any, Optional
+from urllib.parse import parse_qs
 
 
-def normalize_indicator(indicator: str) -> str:
-    """Return a stable privacy-safe keyword/category identifier."""
-    return indicator
+UTC = timezone.utc
+PROMETHEUS_CONTENT_TYPE = "text/plain; version=0.0.4; charset=utf-8"
+JSON_CONTENT_TYPE = "application/json"
 
 
-def build_metrics_event(
-    detection: CrisisDetectionResult,
-    *,
-    continued_conversation: bool = False,
-    false_positive: bool = False,
-    now: float | None = None,
-) -> dict:
-    timestamp = float(time.time() if now is None else now)
-    indicators = [normalize_indicator(indicator) for indicator in detection.indicators]
+@dataclass(frozen=True)
+class CrisisMetricsSnapshot:
+    total_events: int = 0
+    escalated: int = 0
+    resources_shown: int = 0
+    start: Optional[str] = None
+    end: Optional[str] = None
+
+    def to_dict(self) -> dict[str, Any]:
+        return {
+            "total_events": self.total_events,
+            "escalated": self.escalated,
+            "resources_shown": self.resources_shown,
+            "timestamp_range": {
+                "start": self.start,
+                "end": self.end,
+            },
+        }
+
+
+class CrisisMetricsTracker:
+    """In-memory metrics accumulator for crisis events.
+
+    Counts only crisis-bearing events (`level != NONE`). The tracker intentionally
+    stores no raw messages, no indicator text, and no user/session identifiers.
+    """
+
+    def __init__(self) -> None:
+        self.reset()
+
+    def reset(self) -> None:
+        self._total_events = 0
+        self._escalated = 0
+        self._resources_shown = 0
+        self._first_event_at: Optional[datetime] = None
+        self._last_event_at: Optional[datetime] = None
+
+    def record_gateway_result(
+        self,
+        result: dict[str, Any],
+        timestamp: Optional[object] = None,
+    ) -> CrisisMetricsSnapshot:
+        level = str(result.get("level") or "NONE").upper()
+        if level == "NONE":
+            return self.snapshot()
+
+        event_time = _coerce_timestamp(timestamp)
+        self._total_events += 1
+        if bool(result.get("escalate")):
+            self._escalated += 1
+
+        ui = result.get("ui") or {}
+        if any(bool(ui.get(key)) for key in ("provide_988", "show_crisis_panel", "show_overlay")):
+            self._resources_shown += 1
+
+        if self._first_event_at is None or event_time < self._first_event_at:
+            self._first_event_at = event_time
+        if self._last_event_at is None or event_time > self._last_event_at:
+            self._last_event_at = event_time
+
+        return self.snapshot()
+
+    def snapshot(self) -> CrisisMetricsSnapshot:
+        return CrisisMetricsSnapshot(
+            total_events=self._total_events,
+            escalated=self._escalated,
+            resources_shown=self._resources_shown,
+            start=_format_timestamp(self._first_event_at),
+            end=_format_timestamp(self._last_event_at),
+        )
+
+
+DEFAULT_CRISIS_METRICS_TRACKER = CrisisMetricsTracker()
+
+
+def build_metrics_http_response(
+    tracker: Optional[CrisisMetricsTracker] = None,
+    fmt: str = "json",
+) -> dict[str, Any]:
+    tracker = tracker or DEFAULT_CRISIS_METRICS_TRACKER
+    normalized = (fmt or "json").strip().lower()
+    snapshot = tracker.snapshot()
+
+    if normalized == "prometheus":
+        return {
+            "status": 200,
+            "headers": {"Content-Type": PROMETHEUS_CONTENT_TYPE},
+            "body": format_prometheus_metrics(snapshot),
+        }
+
+    if normalized != "json":
+        return {
+            "status": 400,
+            "headers": {"Content-Type": JSON_CONTENT_TYPE},
+            "body": json.dumps(
+                {
+                    "error": "invalid_format",
+                    "supported_formats": ["json", "prometheus"],
+                }
+            ),
+        }
+
     return {
-        "timestamp": timestamp,
-        "level": detection.level,
-        "indicator_count": len(indicators),
-        "indicators": indicators,
-        "continued_conversation": bool(continued_conversation),
-        "false_positive": bool(false_positive),
+        "status": 200,
+        "headers": {"Content-Type": JSON_CONTENT_TYPE},
+        "body": json.dumps(snapshot.to_dict()),
     }
 
 
-def append_metrics_event(
-    log_path: str | Path,
-    detection: CrisisDetectionResult,
-    *,
-    continued_conversation: bool = False,
-    false_positive: bool = False,
-    now: float | None = None,
-) -> dict:
-    event = build_metrics_event(
-        detection,
-        continued_conversation=continued_conversation,
-        false_positive=false_positive,
-        now=now,
+def format_prometheus_metrics(snapshot: CrisisMetricsSnapshot) -> str:
+    start_seconds = _timestamp_to_epoch(snapshot.start)
+    end_seconds = _timestamp_to_epoch(snapshot.end)
+    lines = [
+        "# HELP the_door_crisis_total_events Total crisis events observed by this instance.",
+        "# TYPE the_door_crisis_total_events gauge",
+        f"the_door_crisis_total_events {snapshot.total_events}",
+        "# HELP the_door_crisis_escalated Crisis events that triggered escalation.",
+        "# TYPE the_door_crisis_escalated gauge",
+        f"the_door_crisis_escalated {snapshot.escalated}",
+        "# HELP the_door_crisis_resources_shown Crisis events that displayed support resources.",
+        "# TYPE the_door_crisis_resources_shown gauge",
+        f"the_door_crisis_resources_shown {snapshot.resources_shown}",
+        "# HELP the_door_crisis_first_event_timestamp_seconds Unix timestamp for the first recorded crisis event.",
+        "# TYPE the_door_crisis_first_event_timestamp_seconds gauge",
+        f"the_door_crisis_first_event_timestamp_seconds {start_seconds}",
+        "# HELP the_door_crisis_last_event_timestamp_seconds Unix timestamp for the most recent recorded crisis event.",
+        "# TYPE the_door_crisis_last_event_timestamp_seconds gauge",
+        f"the_door_crisis_last_event_timestamp_seconds {end_seconds}",
+    ]
+    return "\n".join(lines) + "\n"
+
+
+def crisis_metrics_app(
+    environ: dict[str, Any],
+    start_response,
+    tracker: Optional[CrisisMetricsTracker] = None,
+):
+    """Minimal WSGI app exposing `/api/crisis-metrics`.
+
+    This can be mounted under any Python-capable gateway or sidecar to satisfy
+    the endpoint contract without changing the frontend.
+    """
+    path = environ.get("PATH_INFO", "")
+    method = str(environ.get("REQUEST_METHOD", "GET")).upper()
+
+    if path != "/api/crisis-metrics":
+        body = json.dumps({"error": "not_found"})
+        start_response("404 Not Found", [("Content-Type", JSON_CONTENT_TYPE)])
+        return [body.encode("utf-8")]
+
+    if method != "GET":
+        body = json.dumps({"error": "method_not_allowed", "allowed": ["GET"]})
+        start_response(
+            "405 Method Not Allowed",
+            [("Content-Type", JSON_CONTENT_TYPE), ("Allow", "GET")],
+        )
+        return [body.encode("utf-8")]
+
+    fmt = _negotiate_format(
+        environ.get("QUERY_STRING", ""),
+        environ.get("HTTP_ACCEPT", ""),
     )
-    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
+    response = build_metrics_http_response(tracker=tracker, fmt=fmt)
+    status_text = _status_text(response["status"])
+    headers = list(response["headers"].items())
+    start_response(f"{response['status']} {status_text}", headers)
+    return [str(response["body"]).encode("utf-8")]
 
 
-def load_metrics_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 _negotiate_format(query_string: str, accept_header: str) -> str:
+    params = parse_qs(query_string or "")
+    requested = (params.get("format", [""])[0] or "").strip().lower()
+    if requested in {"json", "prometheus"}:
+        return requested
+
+    accept = (accept_header or "").lower()
+    if "text/plain" in accept or "application/openmetrics-text" in accept:
+        return "prometheus"
+    return "json"
 
 
-def build_weekly_summary(
-    events: Iterable[dict],
-    *,
-    now: float | None = None,
-    window_days: int = 7,
-) -> 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]
+def _coerce_timestamp(value: Optional[object]) -> datetime:
+    if value is None:
+        return datetime.now(UTC)
+    if isinstance(value, datetime):
+        if value.tzinfo is None:
+            return value.replace(tzinfo=UTC)
+        return value.astimezone(UTC)
+    if isinstance(value, str):
+        text = value.strip()
+        if text.endswith("Z"):
+            text = text[:-1] + "+00:00"
+        parsed = datetime.fromisoformat(text)
+        if parsed.tzinfo is None:
+            return parsed.replace(tzinfo=UTC)
+        return parsed.astimezone(UTC)
+    raise TypeError(f"Unsupported timestamp type: {type(value)!r}")
 
-    detections_per_level = {level: 0 for level in LEVELS}
-    keyword_counts: Counter[str] = Counter()
-    detections = []
-    continued_after_intervention = 0
 
-    for event in filtered:
-        level = event.get("level", "NONE")
-        detections_per_level[level] = detections_per_level.get(level, 0) + 1
-        keyword_counts.update(event.get("indicators", []))
-        if level != "NONE":
-            detections.append(event)
-            if event.get("continued_conversation"):
-                continued_after_intervention += 1
+def _format_timestamp(value: Optional[datetime]) -> Optional[str]:
+    if value is None:
+        return None
+    return value.astimezone(UTC).replace(microsecond=0).isoformat().replace("+00:00", "Z")
 
-    false_positive_count = sum(1 for event in detections if event.get("false_positive"))
-    false_positive_estimate = (
-        false_positive_count / len(detections) if detections else 0.0
-    )
 
+def _timestamp_to_epoch(value: Optional[str]) -> int:
+    if not value:
+        return 0
+    return int(_coerce_timestamp(value).timestamp())
+
+
+def _status_text(status: int) -> str:
     return {
-        "window_days": window_days,
-        "total_events": len(filtered),
-        "detections_per_level": detections_per_level,
-        "most_common_keywords": [
-            {"keyword": keyword, "count": count}
-            for keyword, count in keyword_counts.most_common(10)
-        ],
-        "false_positive_estimate": false_positive_estimate,
-        "continued_after_intervention": continued_after_intervention,
-    }
-
-
-def render_weekly_summary(summary: dict) -> str:
-    return json.dumps(summary, indent=2)
-
-
-def write_weekly_summary(path: str | Path, summary: dict) -> Path:
-    output_path = Path(path)
-    output_path.parent.mkdir(parents=True, exist_ok=True)
-    output_path.write_text(render_weekly_summary(summary) + "\n", encoding="utf-8")
-    return output_path
-
-
-def record_text_event(
-    text: str,
-    log_path: str | Path,
-    *,
-    continued_conversation: bool = False,
-    false_positive: bool = False,
-    now: float | None = None,
-) -> dict:
-    detection = detect_crisis(text)
-    return append_metrics_event(
-        log_path,
-        detection,
-        continued_conversation=continued_conversation,
-        false_positive=false_positive,
-        now=now,
-    )
-
-
-def main(argv: list[str] | None = None) -> int:
-    parser = argparse.ArgumentParser(description="Privacy-preserving crisis metrics summary")
-    parser.add_argument("--log-path", required=True, help="JSONL event log path")
-    parser.add_argument("--days", type=int, default=7, help="Summary window in days")
-    parser.add_argument("--output", help="Optional file to write summary JSON")
-    args = parser.parse_args(argv)
-
-    events = load_metrics_events(args.log_path)
-    summary = build_weekly_summary(events, window_days=args.days)
-    rendered = render_weekly_summary(summary)
-    print(rendered)
-    if args.output:
-        write_weekly_summary(args.output, summary)
-    return 0
-
-
-if __name__ == "__main__":
-    raise SystemExit(main())
+        200: "OK",
+        400: "Bad Request",
+        404: "Not Found",
+        405: "Method Not Allowed",
+    }.get(status, "OK")

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

@@ -1,100 +0,0 @@
-"""Tests for privacy-preserving crisis metrics aggregation (issue #37)."""
-
-from __future__ import annotations
-
-import json
-import os
-import pathlib
-import sys
-import unittest
-
-sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
-
-from crisis.detect import detect_crisis
-from crisis.gateway import check_crisis
-from crisis.metrics import (
-    append_metrics_event,
-    build_metrics_event,
-    build_weekly_summary,
-    load_metrics_events,
-    render_weekly_summary,
-)
-
-
-class TestMetricsEvent(unittest.TestCase):
-    def test_event_is_privacy_preserving(self):
-        detection = detect_crisis("I want to kill myself")
-        event = build_metrics_event(
-            detection,
-            continued_conversation=True,
-            false_positive=False,
-            now=1_700_000_000,
-        )
-        self.assertEqual(event["timestamp"], 1_700_000_000)
-        self.assertEqual(event["level"], "CRITICAL")
-        self.assertTrue(event["continued_conversation"])
-        self.assertFalse(event["false_positive"])
-        self.assertNotIn("text", event)
-        self.assertNotIn("message", event)
-        self.assertGreaterEqual(event["indicator_count"], 1)
-        self.assertTrue(event["indicators"])
-
-
-class TestMetricsLogAndSummary(unittest.TestCase):
-    def test_append_and_load_metrics_events(self):
-        log_path = pathlib.Path(self._testMethodName).with_suffix(".jsonl")
-        try:
-            append_metrics_event(log_path, detect_crisis("I want to die"), now=1_700_000_000)
-            events = load_metrics_events(log_path)
-            self.assertEqual(len(events), 1)
-            self.assertEqual(events[0]["level"], "CRITICAL")
-        finally:
-            if log_path.exists():
-                log_path.unlink()
-
-    def test_weekly_summary_counts_levels_keywords_and_false_positives(self):
-        events = [
-            build_metrics_event(detect_crisis("I want to die"), continued_conversation=True, false_positive=False, now=1_700_000_000),
-            build_metrics_event(detect_crisis("I'm having a rough day"), continued_conversation=False, false_positive=False, now=1_700_000_100),
-            build_metrics_event(detect_crisis("I want to die"), continued_conversation=False, false_positive=True, now=1_700_000_200),
-            build_metrics_event(detect_crisis("Hello there"), continued_conversation=False, false_positive=False, now=1_700_000_300),
-        ]
-        summary = build_weekly_summary(events, now=1_700_000_400, window_days=7)
-
-        self.assertEqual(summary["detections_per_level"]["CRITICAL"], 2)
-        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"], 1 / 3, places=4)
-        self.assertEqual(summary["most_common_keywords"][0]["count"], 2)
-
-    def test_render_weekly_summary_mentions_required_metrics(self):
-        events = [
-            build_metrics_event(detect_crisis("I want to die"), continued_conversation=True, now=1_700_000_000),
-            build_metrics_event(detect_crisis("I feel hopeless with no way out"), false_positive=True, now=1_700_000_100),
-        ]
-        summary = build_weekly_summary(events, now=1_700_000_200, window_days=7)
-        rendered = render_weekly_summary(summary)
-        self.assertIn("detections_per_level", rendered)
-        self.assertIn("most_common_keywords", rendered)
-        self.assertIn("false_positive_estimate", rendered)
-        self.assertIn("continued_after_intervention", rendered)
-
-
-class TestGatewayMetricsIntegration(unittest.TestCase):
-    def test_check_crisis_can_emit_metrics_event(self):
-        result = check_crisis(
-            "I want to die",
-            metrics_log_path=None,
-            continued_conversation=True,
-            false_positive=False,
-            now=1_700_000_000,
-        )
-        self.assertEqual(result["level"], "CRITICAL")
-        self.assertIn("metrics_event", result)
-        self.assertEqual(result["metrics_event"]["timestamp"], 1_700_000_000)
-        self.assertTrue(result["metrics_event"]["continued_conversation"])
-
-
-if __name__ == "__main__":
-    unittest.main()

tests/test_crisis_metrics_endpoint.py

@@ -0,0 +1,118 @@
+"""Tests for crisis detection metrics endpoint contract (issue #97)."""
+
+import json
+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_crisis, check_crisis_and_record
+from crisis.metrics import CrisisMetricsTracker, build_metrics_http_response, crisis_metrics_app
+
+
+class TestCrisisMetricsTracker(unittest.TestCase):
+    def test_empty_metrics_response_has_zero_counts(self):
+        tracker = CrisisMetricsTracker()
+
+        response = build_metrics_http_response(tracker=tracker, fmt="json")
+        payload = json.loads(response["body"])
+
+        self.assertEqual(response["status"], 200)
+        self.assertEqual(payload["total_events"], 0)
+        self.assertEqual(payload["escalated"], 0)
+        self.assertEqual(payload["resources_shown"], 0)
+        self.assertEqual(payload["timestamp_range"], {"start": None, "end": None})
+        self.assertNotIn("timmy_message", payload)
+        self.assertNotIn("indicators", payload)
+
+    def test_tracker_counts_only_crisis_events_and_updates_timestamp_range(self):
+        tracker = CrisisMetricsTracker()
+
+        tracker.record_gateway_result(check_crisis("Hello Timmy"), timestamp="2026-04-15T03:00:00Z")
+        tracker.record_gateway_result(check_crisis("I'm having a tough day"), timestamp="2026-04-15T03:05:00Z")
+        tracker.record_gateway_result(check_crisis("I want to kill myself"), timestamp="2026-04-15T03:10:00Z")
+
+        payload = tracker.snapshot().to_dict()
+
+        self.assertEqual(payload["total_events"], 2)
+        self.assertEqual(payload["escalated"], 1)
+        self.assertEqual(payload["resources_shown"], 1)
+        self.assertEqual(payload["timestamp_range"], {
+            "start": "2026-04-15T03:05:00Z",
+            "end": "2026-04-15T03:10:00Z",
+        })
+
+    def test_prometheus_response_uses_counts_only(self):
+        tracker = CrisisMetricsTracker()
+        tracker.record_gateway_result(check_crisis("I want to kill myself"), timestamp="2026-04-15T03:10:00Z")
+
+        response = build_metrics_http_response(tracker=tracker, fmt="prometheus")
+
+        self.assertEqual(response["status"], 200)
+        self.assertIn("text/plain", response["headers"]["Content-Type"])
+        self.assertIn("the_door_crisis_total_events 1", response["body"])
+        self.assertIn("the_door_crisis_escalated 1", response["body"])
+        self.assertIn("the_door_crisis_resources_shown 1", response["body"])
+        self.assertNotIn("kill myself", response["body"])
+        self.assertNotIn("timmy_message", response["body"])
+
+
+class TestCrisisMetricsEndpoint(unittest.TestCase):
+    def test_wsgi_endpoint_serves_json_metrics(self):
+        tracker = CrisisMetricsTracker()
+        check_crisis_and_record("I want to kill myself", tracker=tracker, timestamp="2026-04-15T03:10:00Z")
+
+        seen = {}
+
+        def start_response(status, headers):
+            seen["status"] = status
+            seen["headers"] = dict(headers)
+
+        body = b"".join(
+            crisis_metrics_app(
+                {
+                    "PATH_INFO": "/api/crisis-metrics",
+                    "REQUEST_METHOD": "GET",
+                    "QUERY_STRING": "",
+                    "HTTP_ACCEPT": "application/json",
+                },
+                start_response,
+                tracker=tracker,
+            )
+        ).decode("utf-8")
+
+        payload = json.loads(body)
+        self.assertEqual(seen["status"], "200 OK")
+        self.assertEqual(seen["headers"]["Content-Type"], "application/json")
+        self.assertEqual(payload["total_events"], 1)
+        self.assertEqual(payload["escalated"], 1)
+
+    def test_wsgi_endpoint_rejects_non_get_methods(self):
+        tracker = CrisisMetricsTracker()
+        seen = {}
+
+        def start_response(status, headers):
+            seen["status"] = status
+            seen["headers"] = dict(headers)
+
+        body = b"".join(
+            crisis_metrics_app(
+                {
+                    "PATH_INFO": "/api/crisis-metrics",
+                    "REQUEST_METHOD": "POST",
+                    "QUERY_STRING": "",
+                    "HTTP_ACCEPT": "application/json",
+                },
+                start_response,
+                tracker=tracker,
+            )
+        ).decode("utf-8")
+
+        self.assertEqual(seen["status"], "405 Method Not Allowed")
+        self.assertIn("GET", seen["headers"]["Allow"])
+        self.assertIn("method_not_allowed", body)
+
+
+if __name__ == "__main__":
+    unittest.main()