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

@@ -6,9 +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 .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
-from .ab_testing import ABTestCrisisDetector, VariantRecord
 
 __all__ = [
     "detect_crisis",
@@ -17,13 +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",
-    "ABTestCrisisDetector",
-    "VariantRecord",
 ]

crisis/ab_testing.py

@@ -1,112 +0,0 @@
-"""A/B test framework for crisis detection in the-door."""
-
-from __future__ import annotations
-
-import os
-import random
-import time
-from dataclasses import dataclass
-from typing import Callable, Dict, List, Optional, Tuple
-
-from .detect import CrisisDetectionResult
-
-
-def _get_variant_override() -> Optional[str]:
-    """Return env override for deterministic testing/debugging."""
-    value = os.environ.get("CRISIS_AB_VARIANT", "").strip().upper()
-    if value in {"A", "B"}:
-        return value
-    return None
-
-
-@dataclass
-class VariantRecord:
-    """Single crisis detection event record with no user text or PII."""
-
-    variant: str
-    level: str
-    latency_ms: float
-    indicator_count: int
-    false_positive: Optional[bool] = None
-
-
-class ABTestCrisisDetector:
-    """Route crisis detection between two variants and collect comparison stats."""
-
-    def __init__(
-        self,
-        variant_a: Callable[[str], CrisisDetectionResult],
-        variant_b: Callable[[str], CrisisDetectionResult],
-        split: float = 0.5,
-    ):
-        self.variant_a = variant_a
-        self.variant_b = variant_b
-        self.split = max(0.0, min(1.0, float(split)))
-        self.records: List[VariantRecord] = []
-
-    def _select_variant(self) -> str:
-        override = _get_variant_override()
-        if override:
-            return override
-        return "A" if random.random() < self.split else "B"
-
-    def detect(self, text: str) -> Tuple[CrisisDetectionResult, str, int]:
-        variant = self._select_variant()
-        detector = self.variant_a if variant == "A" else self.variant_b
-
-        start = time.perf_counter()
-        result = detector(text)
-        latency_ms = (time.perf_counter() - start) * 1000.0
-
-        record = VariantRecord(
-            variant=variant,
-            level=result.level,
-            latency_ms=latency_ms,
-            indicator_count=len(result.indicators),
-        )
-        self.records.append(record)
-        return result, variant, len(self.records) - 1
-
-    def record_outcome(self, record_id: int, *, false_positive: bool) -> None:
-        if record_id < 0 or record_id >= len(self.records):
-            raise IndexError(f"Unknown record id: {record_id}")
-        self.records[record_id].false_positive = bool(false_positive)
-
-    def get_stats(self) -> Dict[str, dict]:
-        stats: Dict[str, dict] = {}
-        for variant in ("A", "B"):
-            records = [record for record in self.records if record.variant == variant]
-            if not records:
-                stats[variant] = {
-                    "count": 0,
-                    "reviewed_count": 0,
-                    "false_positive_rate": None,
-                }
-                continue
-
-            levels: Dict[str, int] = {}
-            for record in records:
-                levels[record.level] = levels.get(record.level, 0) + 1
-
-            reviewed = [record for record in records if record.false_positive is not None]
-            false_positive_rate = None
-            if reviewed:
-                false_positive_rate = round(
-                    sum(1 for record in reviewed if record.false_positive) / len(reviewed),
-                    4,
-                )
-
-            stats[variant] = {
-                "count": len(records),
-                "avg_latency_ms": round(sum(record.latency_ms for record in records) / len(records), 4),
-                "max_latency_ms": round(max(record.latency_ms for record in records), 4),
-                "min_latency_ms": round(min(record.latency_ms for record in records), 4),
-                "avg_indicator_count": round(sum(record.indicator_count for record in records) / len(records), 4),
-                "levels": levels,
-                "reviewed_count": len(reviewed),
-                "false_positive_rate": false_positive_rate,
-            }
-        return stats
-
-    def reset(self) -> None:
-        self.records.clear()

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,
@@ -25,17 +30,24 @@ from .response import (
 from .session_tracker import CrisisSessionTracker
 
 
-def check_crisis(text: str) -> dict:
+def check_crisis(
+    text: str,
+    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)
 
-    return {
+    result = {
         "level": detection.level,
         "score": detection.score,
         "indicators": detection.indicators,
@@ -49,6 +61,30 @@ def check_crisis(text: str) -> dict:
         "escalate": response.escalate,
     }
 
+    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:
     """

crisis/metrics.py

@@ -0,0 +1,239 @@
+"""
+Crisis metrics tracker and endpoint helpers for the-door.
+
+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 json
+from dataclasses import dataclass
+from datetime import datetime, timezone
+from typing import Any, Optional
+from urllib.parse import parse_qs
+
+
+UTC = timezone.utc
+PROMETHEUS_CONTENT_TYPE = "text/plain; version=0.0.4; charset=utf-8"
+JSON_CONTENT_TYPE = "application/json"
+
+
+@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 {
+        "status": 200,
+        "headers": {"Content-Type": JSON_CONTENT_TYPE},
+        "body": json.dumps(snapshot.to_dict()),
+    }
+
+
+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", ""),
+    )
+    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 _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 _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}")
+
+
+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")
+
+
+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 {
+        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_ab_testing.py

@@ -1,138 +0,0 @@
-"""Tests for crisis.ab_testing — A/B test framework for crisis detection (#101)."""
-
-import os
-from unittest.mock import patch
-
-import pytest
-
-from crisis.ab_testing import ABTestCrisisDetector
-from crisis.detect import CrisisDetectionResult, detect_crisis
-
-
-@pytest.fixture(autouse=True)
-def clear_variant_override():
-    old = os.environ.pop("CRISIS_AB_VARIANT", None)
-    try:
-        yield
-    finally:
-        if old is not None:
-            os.environ["CRISIS_AB_VARIANT"] = old
-        else:
-            os.environ.pop("CRISIS_AB_VARIANT", None)
-
-
-def _make_variant(level: str, indicators=None):
-    indicators = indicators or [f"mock_{level.lower()}"]
-
-    def fn(text: str) -> CrisisDetectionResult:
-        return CrisisDetectionResult(level=level, indicators=list(indicators))
-
-    return fn
-
-
-def test_detect_returns_result_variant_and_logged_record():
-    detector = ABTestCrisisDetector(
-        variant_a=_make_variant("LOW"),
-        variant_b=_make_variant("HIGH"),
-    )
-
-    with patch.object(detector, "_select_variant", return_value="A"):
-        result, variant, record_id = detector.detect("test message")
-
-    assert isinstance(result, CrisisDetectionResult)
-    assert variant == "A"
-    assert record_id == 0
-    assert len(detector.records) == 1
-    assert detector.records[0].variant == "A"
-    assert detector.records[0].level == "LOW"
-
-
-def test_env_override_forces_variant_b():
-    os.environ["CRISIS_AB_VARIANT"] = "b"
-    detector = ABTestCrisisDetector(
-        variant_a=_make_variant("LOW"),
-        variant_b=_make_variant("HIGH"),
-    )
-
-    result, variant, _ = detector.detect("test")
-
-    assert variant == "B"
-    assert result.level == "HIGH"
-
-
-def test_get_stats_reports_latency_counts_and_level_breakdown():
-    detector = ABTestCrisisDetector(
-        variant_a=_make_variant("LOW"),
-        variant_b=_make_variant("CRITICAL"),
-    )
-
-    with patch.object(detector, "_select_variant", side_effect=["A", "A", "B"]):
-        detector.detect("first")
-        detector.detect("second")
-        detector.detect("third")
-
-    stats = detector.get_stats()
-    assert stats["A"]["count"] == 2
-    assert stats["B"]["count"] == 1
-    assert stats["A"]["levels"]["LOW"] == 2
-    assert stats["B"]["levels"]["CRITICAL"] == 1
-    assert "avg_latency_ms" in stats["A"]
-    assert "avg_indicator_count" in stats["B"]
-
-
-def test_false_positive_rate_is_computed_from_reviewed_outcomes():
-    detector = ABTestCrisisDetector(
-        variant_a=_make_variant("LOW"),
-        variant_b=_make_variant("HIGH"),
-    )
-
-    with patch.object(detector, "_select_variant", side_effect=["A", "A", "B"]):
-        _, _, a0 = detector.detect("first")
-        _, _, a1 = detector.detect("second")
-        _, _, b0 = detector.detect("third")
-
-    detector.record_outcome(a0, false_positive=True)
-    detector.record_outcome(a1, false_positive=False)
-    detector.record_outcome(b0, false_positive=False)
-
-    stats = detector.get_stats()
-    assert stats["A"]["reviewed_count"] == 2
-    assert stats["A"]["false_positive_rate"] == 0.5
-    assert stats["B"]["false_positive_rate"] == 0.0
-
-
-def test_record_outcome_rejects_unknown_record():
-    detector = ABTestCrisisDetector(
-        variant_a=_make_variant("LOW"),
-        variant_b=_make_variant("HIGH"),
-    )
-
-    with pytest.raises(IndexError):
-        detector.record_outcome(99, false_positive=True)
-
-
-def test_reset_clears_records_and_stats():
-    detector = ABTestCrisisDetector(
-        variant_a=_make_variant("LOW"),
-        variant_b=_make_variant("HIGH"),
-    )
-    detector.detect("test")
-    detector.reset()
-
-    assert detector.records == []
-    stats = detector.get_stats()
-    assert stats["A"]["count"] == 0
-    assert stats["B"]["count"] == 0
-
-
-def test_with_real_detector_integration():
-    detector = ABTestCrisisDetector(
-        variant_a=detect_crisis,
-        variant_b=detect_crisis,
-    )
-
-    result, variant, record_id = detector.detect("I want to kill myself")
-
-    assert result.level == "CRITICAL"
-    assert variant in ("A", "B")
-    assert record_id == 0

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()