B2: tolerant JSON extraction from Claude responses + graceful fallback
_extract_json_payload handles a json/JSON/bare fence, raw JSON and JSON embedded in prose; any unparseable response or contract violation now degrades to the deterministic scoring_service instead of raising. Also guards the response envelope itself (content[0].text). B3: single home for recommendation scoring services/recommendation_service.py holds the rules; routers/stats.py and backend/services/stats_service.py both delegate to it. Unified rules are the union of the two old copies: same weights/threshold, substring matching (superset of the old exact match), tolerant key aliases, health_flags rule kept. Endpoint response contract unchanged. Plus: Overpass circuit breaker and concurrent zone queries in geo_service - 128 sequential calls per analysis no longer each burn a connect timeout when the host has no outbound network. Tests: 152 -> 194 passed.
This commit is contained in:
root
+10
-13
@@ -2,6 +2,7 @@ from fastapi import APIRouter
|
||||
|
||||
from backend.database import db
|
||||
from backend.schemas import DashboardResponse, RecommendationRequest
|
||||
from services.recommendation_service import score_recommendation
|
||||
|
||||
router = APIRouter(prefix='/api/v1/stats', tags=['stats'])
|
||||
|
||||
@@ -18,18 +19,14 @@ def stats_heatmap() -> dict:
|
||||
|
||||
@router.post('/recommendation')
|
||||
def stats_recommendation(payload: RecommendationRequest) -> dict:
|
||||
score = 0
|
||||
if payload.age is not None and payload.age < 12:
|
||||
score += 20
|
||||
if payload.elapsed_hours is not None and payload.elapsed_hours >= 12:
|
||||
score += 20
|
||||
if payload.terrain_primary in {'лес', 'болото', 'вода'}:
|
||||
score += 15
|
||||
if payload.weather in {'дождь', 'туман', 'снег', 'ночь'}:
|
||||
score += 15
|
||||
if len(payload.health_flags) >= 2:
|
||||
score += 15
|
||||
result = score_recommendation(
|
||||
age=payload.age,
|
||||
elapsed_hours=payload.elapsed_hours,
|
||||
terrain=payload.terrain_primary,
|
||||
weather=payload.weather,
|
||||
health_flags=payload.health_flags,
|
||||
)
|
||||
return {
|
||||
'recommendation': 'Высокий приоритет на прочёс и дрон' if score >= 40 else 'Стандартный приоритет поиска',
|
||||
'score': score,
|
||||
'recommendation': result['recommendation'],
|
||||
'score': result['score'],
|
||||
}
|
||||
|
||||
@@ -1,12 +1,17 @@
|
||||
from __future__ import annotations
|
||||
|
||||
from typing import Any
|
||||
|
||||
try:
|
||||
from backend.database import db
|
||||
except ImportError: # pragma: no cover - compatibility for direct service imports
|
||||
from database import db
|
||||
|
||||
# Scoring rules live in services/recommendation_service.py (B3) — this module
|
||||
# only re-exports them so existing `stats_service.get_statistical_recommendation`
|
||||
# callers keep working.
|
||||
from services.recommendation_service import get_statistical_recommendation
|
||||
|
||||
__all__ = ['summary', 'heatmap', 'get_statistical_recommendation']
|
||||
|
||||
|
||||
def summary() -> dict:
|
||||
return db.stats()
|
||||
@@ -14,26 +19,3 @@ def summary() -> dict:
|
||||
|
||||
def heatmap() -> list[dict]:
|
||||
return db.heatmap()
|
||||
|
||||
|
||||
def get_statistical_recommendation(case_data: dict[str, Any]) -> dict[str, Any]:
|
||||
score = 0
|
||||
age = case_data.get('age') or case_data.get('age_years')
|
||||
elapsed = case_data.get('elapsed_hours')
|
||||
terrain = str(case_data.get('terrain') or case_data.get('terrain_primary') or '').lower()
|
||||
weather = str(case_data.get('weather') or case_data.get('precipitation') or '').lower()
|
||||
|
||||
if age is not None and age < 12:
|
||||
score += 20
|
||||
if elapsed is not None and elapsed >= 12:
|
||||
score += 20
|
||||
if any(token in terrain for token in ('лес', 'болото', 'вода')):
|
||||
score += 15
|
||||
if any(token in weather for token in ('дождь', 'туман', 'снег', 'ночь')):
|
||||
score += 15
|
||||
|
||||
return {
|
||||
'score': score,
|
||||
'priority': 'high' if score >= 40 else 'normal',
|
||||
'recommendation': 'Высокий приоритет на прочёс и дрон' if score >= 40 else 'Стандартный приоритет поиска',
|
||||
}
|
||||
|
||||
@@ -11,7 +11,8 @@ from services.claude_service import (
|
||||
analyze_case,
|
||||
analyze_with_fallback,
|
||||
AnalysisResult,
|
||||
PrimaryZone
|
||||
PrimaryZone,
|
||||
_extract_json_payload
|
||||
)
|
||||
|
||||
|
||||
@@ -369,3 +370,133 @@ class TestImmediateActions:
|
||||
result = await analyze_with_fallback(case_data)
|
||||
|
||||
assert any("10-15" in action or "камер" in action for action in result.immediate_actions)
|
||||
|
||||
|
||||
class TestExtractJsonPayload:
|
||||
"""Test tolerant JSON extraction from a model response (B2)."""
|
||||
|
||||
def test_bare_json(self):
|
||||
assert _extract_json_payload('{"a": 1}') == {'a': 1}
|
||||
|
||||
def test_json_fence(self):
|
||||
assert _extract_json_payload('```json\n{"a": 1}\n```') == {'a': 1}
|
||||
|
||||
def test_uppercase_json_fence(self):
|
||||
assert _extract_json_payload('```JSON\n{"a": 1}\n```') == {'a': 1}
|
||||
|
||||
def test_bare_fence(self):
|
||||
assert _extract_json_payload('```\n{"a": 1}\n```') == {'a': 1}
|
||||
|
||||
def test_prose_around_json(self):
|
||||
content = 'Вот результат анализа:\n{"a": 1}\nНадеюсь, это поможет.'
|
||||
assert _extract_json_payload(content) == {'a': 1}
|
||||
|
||||
def test_prose_around_fenced_json(self):
|
||||
content = 'Разбор:\n```json\n{"a": 1}\n```\nКонец.'
|
||||
assert _extract_json_payload(content) == {'a': 1}
|
||||
|
||||
def test_json_array_is_rejected(self):
|
||||
"""A top-level array is not a valid AnalysisResult payload."""
|
||||
with pytest.raises(ValueError):
|
||||
_extract_json_payload('[1, 2, 3]')
|
||||
|
||||
def test_plain_text_raises(self):
|
||||
with pytest.raises(ValueError):
|
||||
_extract_json_payload('Извините, я не могу выполнить этот запрос.')
|
||||
|
||||
def test_empty_raises(self):
|
||||
with pytest.raises(ValueError):
|
||||
_extract_json_payload('')
|
||||
|
||||
def test_none_raises(self):
|
||||
with pytest.raises(ValueError):
|
||||
_extract_json_payload(None)
|
||||
|
||||
|
||||
class TestClaudeResponseFallback:
|
||||
"""Malformed Claude responses must degrade to scoring, not crash (B2)."""
|
||||
|
||||
def _mock_client(self, mock_client, text=None, payload=None):
|
||||
mock_response = MagicMock()
|
||||
mock_response.status_code = 200
|
||||
mock_response.json.return_value = (
|
||||
payload if payload is not None else {"content": [{"text": text}]}
|
||||
)
|
||||
mock_client.return_value.__aenter__.return_value.post = AsyncMock(
|
||||
return_value=mock_response
|
||||
)
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_unparseable_text_falls_back(self):
|
||||
"""Model answers in prose instead of JSON -> deterministic fallback."""
|
||||
with patch.dict(os.environ, {'ANTHROPIC_API_KEY': 'test_key'}):
|
||||
with patch('httpx.AsyncClient') as mock_client:
|
||||
self._mock_client(mock_client, text='Не могу помочь с этим.')
|
||||
|
||||
result = await analyze_case({'age': 10, 'terrain': 'лес'})
|
||||
|
||||
assert result.fallback_used is True
|
||||
assert isinstance(result, AnalysisResult)
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_truncated_json_falls_back(self):
|
||||
with patch.dict(os.environ, {'ANTHROPIC_API_KEY': 'test_key'}):
|
||||
with patch('httpx.AsyncClient') as mock_client:
|
||||
self._mock_client(mock_client, text='```json\n{"urgency": "высок')
|
||||
|
||||
result = await analyze_case({'age': 10})
|
||||
|
||||
assert result.fallback_used is True
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_valid_json_missing_required_fields_falls_back(self):
|
||||
"""Parseable JSON that violates the AnalysisResult contract."""
|
||||
with patch.dict(os.environ, {'ANTHROPIC_API_KEY': 'test_key'}):
|
||||
with patch('httpx.AsyncClient') as mock_client:
|
||||
self._mock_client(mock_client, text='{"urgency": "высокая"}')
|
||||
|
||||
result = await analyze_case({'age': 10})
|
||||
|
||||
assert result.fallback_used is True
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_unexpected_envelope_falls_back(self):
|
||||
"""API envelope without content[0].text -> fallback, not KeyError."""
|
||||
with patch.dict(os.environ, {'ANTHROPIC_API_KEY': 'test_key'}):
|
||||
with patch('httpx.AsyncClient') as mock_client:
|
||||
self._mock_client(mock_client, payload={'unexpected': 'shape'})
|
||||
|
||||
result = await analyze_case({'age': 10})
|
||||
|
||||
assert result.fallback_used is True
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_unfenced_json_with_prose_succeeds(self):
|
||||
"""Recovery path: valid payload wrapped in prose is still used."""
|
||||
import json as _json
|
||||
|
||||
payload = {
|
||||
"urgency": "высокая",
|
||||
"primary_zones": [{
|
||||
"priority": 1,
|
||||
"name": "Лес север",
|
||||
"direction": "N",
|
||||
"distance": 1.5,
|
||||
"reason": "Вероятное направление"
|
||||
}],
|
||||
"search_radius_km": 5.0,
|
||||
"key_locations": ["водоёмы"],
|
||||
"behavioral_prediction": "Движение по тропам",
|
||||
"immediate_actions": ["Организовать поиск"],
|
||||
"summary": "Резюме"
|
||||
}
|
||||
text = 'Результат:\n' + _json.dumps(payload, ensure_ascii=False) + '\nГотово.'
|
||||
|
||||
with patch.dict(os.environ, {'ANTHROPIC_API_KEY': 'test_key'}):
|
||||
with patch('httpx.AsyncClient') as mock_client:
|
||||
self._mock_client(mock_client, text=text)
|
||||
|
||||
result = await analyze_case({'age': 10})
|
||||
|
||||
assert result.fallback_used is False
|
||||
assert result.urgency == "высокая"
|
||||
|
||||
@@ -0,0 +1,161 @@
|
||||
"""
|
||||
Tests for the unified recommendation scoring (B3).
|
||||
|
||||
Covers the shared scorer directly, the dict-based alias entry point, and the
|
||||
/api/v1/stats/recommendation endpoint that now delegates to it.
|
||||
"""
|
||||
import pytest
|
||||
from fastapi.testclient import TestClient
|
||||
|
||||
from backend.main import app
|
||||
from services.recommendation_service import (
|
||||
HIGH_PRIORITY_TEXT,
|
||||
NORMAL_PRIORITY_TEXT,
|
||||
get_statistical_recommendation,
|
||||
score_recommendation,
|
||||
)
|
||||
|
||||
client = TestClient(app)
|
||||
|
||||
|
||||
class TestScoreRecommendation:
|
||||
"""Weights and threshold of the shared scorer."""
|
||||
|
||||
def test_empty_case_scores_zero(self):
|
||||
result = score_recommendation()
|
||||
assert result['score'] == 0
|
||||
assert result['priority'] == 'normal'
|
||||
assert result['recommendation'] == NORMAL_PRIORITY_TEXT
|
||||
|
||||
def test_young_child(self):
|
||||
assert score_recommendation(age=8)['score'] == 20
|
||||
|
||||
def test_age_at_threshold_not_counted(self):
|
||||
assert score_recommendation(age=12)['score'] == 0
|
||||
|
||||
def test_long_elapsed(self):
|
||||
assert score_recommendation(elapsed_hours=12)['score'] == 20
|
||||
|
||||
def test_short_elapsed_not_counted(self):
|
||||
assert score_recommendation(elapsed_hours=11)['score'] == 0
|
||||
|
||||
def test_risky_terrain(self):
|
||||
assert score_recommendation(terrain='лес')['score'] == 15
|
||||
|
||||
def test_adverse_weather(self):
|
||||
assert score_recommendation(weather='дождь')['score'] == 15
|
||||
|
||||
def test_multiple_health_flags(self):
|
||||
assert score_recommendation(health_flags=['эпилепсия', 'РАС'])['score'] == 15
|
||||
|
||||
def test_single_health_flag_not_counted(self):
|
||||
assert score_recommendation(health_flags=['эпилепсия'])['score'] == 0
|
||||
|
||||
def test_terrain_matches_as_substring(self):
|
||||
"""Substring match — the router previously required an exact match."""
|
||||
assert score_recommendation(terrain='смешанный лес')['score'] == 15
|
||||
|
||||
def test_weather_matches_as_substring(self):
|
||||
assert score_recommendation(weather='сильный дождь')['score'] == 15
|
||||
|
||||
def test_terrain_case_insensitive(self):
|
||||
assert score_recommendation(terrain='ЛЕС')['score'] == 15
|
||||
|
||||
def test_unknown_terrain_scores_zero(self):
|
||||
assert score_recommendation(terrain='поле')['score'] == 0
|
||||
|
||||
def test_high_priority_at_threshold(self):
|
||||
result = score_recommendation(age=8, elapsed_hours=14)
|
||||
assert result['score'] == 40
|
||||
assert result['priority'] == 'high'
|
||||
assert result['recommendation'] == HIGH_PRIORITY_TEXT
|
||||
|
||||
def test_just_below_threshold_is_normal(self):
|
||||
result = score_recommendation(age=8, terrain='лес')
|
||||
assert result['score'] == 35
|
||||
assert result['priority'] == 'normal'
|
||||
|
||||
def test_all_factors(self):
|
||||
result = score_recommendation(
|
||||
age=6,
|
||||
elapsed_hours=24,
|
||||
terrain='болото',
|
||||
weather='туман',
|
||||
health_flags=['РАС', 'эпилепсия'],
|
||||
)
|
||||
assert result['score'] == 85
|
||||
assert result['priority'] == 'high'
|
||||
|
||||
|
||||
class TestGetStatisticalRecommendation:
|
||||
"""Dict entry point and its field aliases."""
|
||||
|
||||
def test_age_alias(self):
|
||||
assert get_statistical_recommendation({'age_years': 8})['score'] == 20
|
||||
|
||||
def test_age_preferred_over_alias(self):
|
||||
assert get_statistical_recommendation({'age': 8, 'age_years': 30})['score'] == 20
|
||||
|
||||
def test_terrain_alias(self):
|
||||
assert get_statistical_recommendation({'terrain_primary': 'лес'})['score'] == 15
|
||||
|
||||
def test_weather_alias(self):
|
||||
assert get_statistical_recommendation({'precipitation': 'снег'})['score'] == 15
|
||||
|
||||
def test_health_flags_counted(self):
|
||||
payload = {'health_flags': ['РАС', 'эпилепсия']}
|
||||
assert get_statistical_recommendation(payload)['score'] == 15
|
||||
|
||||
def test_missing_keys_are_safe(self):
|
||||
assert get_statistical_recommendation({})['score'] == 0
|
||||
|
||||
def test_none_values_are_safe(self):
|
||||
payload = {'age': None, 'elapsed_hours': None, 'terrain': None, 'weather': None}
|
||||
assert get_statistical_recommendation(payload)['score'] == 0
|
||||
|
||||
|
||||
class TestRecommendationEndpoint:
|
||||
"""The endpoint keeps its response contract while delegating."""
|
||||
|
||||
def test_returns_score_and_recommendation(self):
|
||||
response = client.post(
|
||||
'/api/v1/stats/recommendation',
|
||||
json={'age': 8, 'elapsed_hours': 14},
|
||||
)
|
||||
assert response.status_code == 200
|
||||
body = response.json()
|
||||
assert body['score'] == 40
|
||||
assert body['recommendation'] == HIGH_PRIORITY_TEXT
|
||||
|
||||
def test_normal_priority_case(self):
|
||||
response = client.post('/api/v1/stats/recommendation', json={'age': 30})
|
||||
assert response.status_code == 200
|
||||
body = response.json()
|
||||
assert body['score'] == 0
|
||||
assert body['recommendation'] == NORMAL_PRIORITY_TEXT
|
||||
|
||||
def test_empty_payload_accepted(self):
|
||||
response = client.post('/api/v1/stats/recommendation', json={})
|
||||
assert response.status_code == 200
|
||||
assert response.json()['score'] == 0
|
||||
|
||||
def test_endpoint_matches_shared_scorer(self):
|
||||
payload = {
|
||||
'age': 6,
|
||||
'elapsed_hours': 24,
|
||||
'terrain_primary': 'болото',
|
||||
'weather': 'туман',
|
||||
'health_flags': ['РАС', 'эпилепсия'],
|
||||
}
|
||||
response = client.post('/api/v1/stats/recommendation', json=payload)
|
||||
assert response.status_code == 200
|
||||
|
||||
expected = score_recommendation(
|
||||
age=payload['age'],
|
||||
elapsed_hours=payload['elapsed_hours'],
|
||||
terrain=payload['terrain_primary'],
|
||||
weather=payload['weather'],
|
||||
health_flags=payload['health_flags'],
|
||||
)
|
||||
assert response.json()['score'] == expected['score']
|
||||
assert response.json()['recommendation'] == expected['recommendation']
|
||||
Reference in New Issue
Block a user