Spiritual_Health_Project / tests /test_prompt_optimization_properties.py
DocUA's picture
feat: Complete prompt optimization system implementation
24214fc
"""
Property-based tests for prompt optimization system.
These tests verify correctness properties across multiple inputs and scenarios
using the Hypothesis library for property-based testing.
"""
import sys
import os
sys.path.append('src')
import pytest
from hypothesis import given, strategies as st, settings
from datetime import datetime
from typing import List, Dict, Any
from config.prompt_management import (
PromptController, IndicatorCatalog, RulesCatalog, TemplateCatalog
)
from config.prompt_management.data_models import (
Indicator, Rule, Template, IndicatorCategory, ValidationResult,
ConversationHistory, Message, Classification, ScenarioType
)
class TestSharedComponentPropagation:
"""
**Feature: prompt-optimization, Property 5: Shared Component Update Propagation**
**Validates: Requirements 5.1, 5.2, 5.3, 5.4, 5.5**
Property: For any update to shared prompt components (indicators, rules, categories),
all dependent AI agents should receive the changes consistently while maintaining
backward compatibility and validation integrity.
"""
@given(
indicator_name=st.text(min_size=3, max_size=50, alphabet=st.characters(whitelist_categories=('Lu', 'Ll', 'Nd', 'Pc'))),
definition=st.text(min_size=10, max_size=200),
severity_weight=st.floats(min_value=0.0, max_value=1.0),
examples=st.lists(st.text(min_size=5, max_size=100), min_size=1, max_size=5)
)
@settings(max_examples=100)
def test_indicator_propagation_consistency(self, indicator_name: str, definition: str,
severity_weight: float, examples: List[str]):
"""
Test that indicator updates propagate consistently to all AI agents.
Property: When an indicator is added to the shared catalog, all AI agents
should receive the same indicator definition in their prompt configurations.
"""
# Create controller and get initial state
controller = PromptController()
# Create test indicator
test_indicator = Indicator(
name=indicator_name,
category=IndicatorCategory.EMOTIONAL,
definition=definition,
examples=examples,
severity_weight=severity_weight
)
# Add indicator to catalog
success = controller.indicator_catalog.add_indicator(test_indicator)
# Skip if indicator already exists (duplicate name)
if not success:
return
# Clear cache to force reload
controller._prompt_cache.clear()
# Get prompt configurations for different agents
spiritual_config = controller.get_prompt('spiritual_monitor')
triage_config = controller.get_prompt('triage_question')
evaluator_config = controller.get_prompt('triage_evaluator')
# Verify all agents have the same indicator
spiritual_indicators = {ind.name: ind for ind in spiritual_config.shared_indicators}
triage_indicators = {ind.name: ind for ind in triage_config.shared_indicators}
evaluator_indicators = {ind.name: ind for ind in evaluator_config.shared_indicators}
# Property assertion: All agents should have the same indicator
assert indicator_name in spiritual_indicators, f"Spiritual monitor missing indicator: {indicator_name}"
assert indicator_name in triage_indicators, f"Triage question missing indicator: {indicator_name}"
assert indicator_name in evaluator_indicators, f"Triage evaluator missing indicator: {indicator_name}"
# Property assertion: Indicator definitions should be identical
spiritual_ind = spiritual_indicators[indicator_name]
triage_ind = triage_indicators[indicator_name]
evaluator_ind = evaluator_indicators[indicator_name]
assert spiritual_ind.definition == definition, "Spiritual monitor has different definition"
assert triage_ind.definition == definition, "Triage question has different definition"
assert evaluator_ind.definition == definition, "Triage evaluator has different definition"
assert spiritual_ind.severity_weight == severity_weight, "Spiritual monitor has different weight"
assert triage_ind.severity_weight == severity_weight, "Triage question has different weight"
assert evaluator_ind.severity_weight == severity_weight, "Triage evaluator has different weight"
assert spiritual_ind.examples == examples, "Spiritual monitor has different examples"
assert triage_ind.examples == examples, "Triage question has different examples"
assert evaluator_ind.examples == examples, "Triage evaluator has different examples"
@given(
rule_id=st.text(min_size=3, max_size=30, alphabet=st.characters(whitelist_categories=('Lu', 'Ll', 'Nd', 'Pc'))),
description=st.text(min_size=10, max_size=200),
condition=st.text(min_size=5, max_size=100),
action=st.text(min_size=5, max_size=50),
priority=st.integers(min_value=1, max_value=100)
)
@settings(max_examples=100)
def test_rule_propagation_consistency(self, rule_id: str, description: str,
condition: str, action: str, priority: int):
"""
Test that rule updates propagate consistently to all AI agents.
Property: When a rule is added to the shared catalog, all AI agents
should receive the same rule definition in their prompt configurations.
"""
# Create controller
controller = PromptController()
# Create test rule
test_rule = Rule(
rule_id=rule_id,
description=description,
condition=condition,
action=action,
priority=priority
)
# Add rule to catalog
success = controller.rules_catalog.add_rule(test_rule)
# Skip if rule already exists (duplicate ID)
if not success:
return
# Clear cache to force reload
controller._prompt_cache.clear()
# Get prompt configurations for different agents
spiritual_config = controller.get_prompt('spiritual_monitor')
triage_config = controller.get_prompt('triage_question')
evaluator_config = controller.get_prompt('triage_evaluator')
# Verify all agents have the same rule
spiritual_rules = {rule.rule_id: rule for rule in spiritual_config.shared_rules}
triage_rules = {rule.rule_id: rule for rule in triage_config.shared_rules}
evaluator_rules = {rule.rule_id: rule for rule in evaluator_config.shared_rules}
# Property assertion: All agents should have the same rule
assert rule_id in spiritual_rules, f"Spiritual monitor missing rule: {rule_id}"
assert rule_id in triage_rules, f"Triage question missing rule: {rule_id}"
assert rule_id in evaluator_rules, f"Triage evaluator missing rule: {rule_id}"
# Property assertion: Rule definitions should be identical
spiritual_rule = spiritual_rules[rule_id]
triage_rule = triage_rules[rule_id]
evaluator_rule = evaluator_rules[rule_id]
assert spiritual_rule.description == description, "Spiritual monitor has different description"
assert triage_rule.description == description, "Triage question has different description"
assert evaluator_rule.description == description, "Triage evaluator has different description"
assert spiritual_rule.condition == condition, "Spiritual monitor has different condition"
assert triage_rule.condition == condition, "Triage question has different condition"
assert evaluator_rule.condition == condition, "Triage evaluator has different condition"
assert spiritual_rule.priority == priority, "Spiritual monitor has different priority"
assert triage_rule.priority == priority, "Triage question has different priority"
assert evaluator_rule.priority == priority, "Triage evaluator has different priority"
@given(
template_id=st.text(min_size=3, max_size=30, alphabet=st.characters(whitelist_categories=('Lu', 'Ll', 'Nd', 'Pc'))),
name=st.text(min_size=5, max_size=50),
content=st.text(min_size=10, max_size=200),
category=st.sampled_from(['consent', 'triage', 'response', 'classification'])
)
@settings(max_examples=100)
def test_template_propagation_consistency(self, template_id: str, name: str,
content: str, category: str):
"""
Test that template updates propagate consistently to all AI agents.
Property: When a template is added to the shared catalog, all AI agents
should receive the same template definition in their prompt configurations.
"""
# Create controller
controller = PromptController()
# Create test template
test_template = Template(
template_id=template_id,
name=name,
content=content,
variables=[], # Simplified for testing
category=category
)
# Add template to catalog
success = controller.template_catalog.add_template(test_template)
# Skip if template already exists (duplicate ID)
if not success:
return
# Clear cache to force reload
controller._prompt_cache.clear()
# Get prompt configurations for different agents
spiritual_config = controller.get_prompt('spiritual_monitor')
triage_config = controller.get_prompt('triage_question')
evaluator_config = controller.get_prompt('triage_evaluator')
# Verify all agents have the same template
spiritual_templates = {tmpl.template_id: tmpl for tmpl in spiritual_config.templates}
triage_templates = {tmpl.template_id: tmpl for tmpl in triage_config.templates}
evaluator_templates = {tmpl.template_id: tmpl for tmpl in evaluator_config.templates}
# Property assertion: All agents should have the same template
assert template_id in spiritual_templates, f"Spiritual monitor missing template: {template_id}"
assert template_id in triage_templates, f"Triage question missing template: {template_id}"
assert template_id in evaluator_templates, f"Triage evaluator missing template: {template_id}"
# Property assertion: Template definitions should be identical
spiritual_tmpl = spiritual_templates[template_id]
triage_tmpl = triage_templates[template_id]
evaluator_tmpl = evaluator_templates[template_id]
assert spiritual_tmpl.name == name, "Spiritual monitor has different template name"
assert triage_tmpl.name == name, "Triage question has different template name"
assert evaluator_tmpl.name == name, "Triage evaluator has different template name"
assert spiritual_tmpl.content == content, "Spiritual monitor has different template content"
assert triage_tmpl.content == content, "Triage question has different template content"
assert evaluator_tmpl.content == content, "Triage evaluator has different template content"
assert spiritual_tmpl.category == category, "Spiritual monitor has different template category"
assert triage_tmpl.category == category, "Triage question has different template category"
assert evaluator_tmpl.category == category, "Triage evaluator has different template category"
def test_validation_integrity_maintained(self):
"""
Test that validation integrity is maintained during component updates.
Property: When shared components are updated, the validation system
should continue to work correctly and catch inconsistencies.
"""
controller = PromptController()
# Initial validation should pass
initial_result = controller.validate_consistency()
assert isinstance(initial_result, ValidationResult), "Validation should return ValidationResult"
# Add a valid indicator
valid_indicator = Indicator(
name="test_valid_indicator",
category=IndicatorCategory.EMOTIONAL,
definition="A test indicator for validation",
examples=["test example"],
severity_weight=0.5
)
controller.indicator_catalog.add_indicator(valid_indicator)
# Validation should still work
post_update_result = controller.validate_consistency()
assert isinstance(post_update_result, ValidationResult), "Validation should still work after update"
# Add an invalid indicator (invalid severity weight)
invalid_indicator = Indicator(
name="test_invalid_indicator",
category=IndicatorCategory.EMOTIONAL,
definition="An invalid test indicator",
examples=["test example"],
severity_weight=2.0 # Invalid: > 1.0
)
controller.indicator_catalog.add_indicator(invalid_indicator)
# Validation should catch the error
validation_with_error = controller.validate_consistency()
assert not validation_with_error.is_valid, "Validation should catch invalid severity weight"
assert any("severity weight" in error.lower() for error in validation_with_error.errors), \
"Should have severity weight error"
@given(
session_id=st.text(min_size=5, max_size=20, alphabet=st.characters(whitelist_categories=('Lu', 'Ll', 'Nd'))),
agent_type=st.sampled_from(['spiritual_monitor', 'triage_question', 'triage_evaluator']),
session_prompt=st.text(min_size=20, max_size=500)
)
@settings(max_examples=50)
def test_session_isolation_property(self, session_id: str, agent_type: str, session_prompt: str):
"""
Test that session overrides don't affect other sessions or base prompts.
Property: Session-level prompt overrides should be isolated and not affect
other sessions or the base centralized prompts.
"""
controller = PromptController()
# Get base prompt configuration
base_config = controller.get_prompt(agent_type)
base_prompt_content = base_config.base_prompt
# Set session override
success = controller.set_session_override(agent_type, session_prompt, session_id)
assert success, "Session override should be set successfully"
# Get prompt with session override
session_config = controller.get_prompt(agent_type, session_id=session_id)
# Property assertion: Session should have override content
assert session_config.session_override == session_prompt, "Session should have override content"
# Property assertion: Base prompt should be unchanged
base_config_after = controller.get_prompt(agent_type)
assert base_config_after.base_prompt == base_prompt_content, "Base prompt should be unchanged"
# Property assertion: Different session should not be affected
different_session_id = f"different_{session_id}"
different_session_config = controller.get_prompt(agent_type, session_id=different_session_id)
assert different_session_config.session_override is None, "Different session should not have override"
# Clean up
controller.clear_session_overrides(session_id)
# Property assertion: After cleanup, session should revert to base
cleaned_config = controller.get_prompt(agent_type, session_id=session_id)
assert cleaned_config.session_override is None, "Session should revert to base after cleanup"
class TestTargetedQuestionGeneration:
"""
**Feature: prompt-optimization, Property 2: Scenario-Targeted Question Generation**
**Validates: Requirements 2.1, 2.2, 2.3, 2.4, 2.5**
Property: For any YELLOW scenario (loss of interest, loss of loved one, lack of support,
vague stress, sleep issues), the generated triage question should specifically address
the distinction between emotional distress and external factors relevant to that scenario type.
"""
@given(
scenario_type=st.sampled_from(['loss_of_interest', 'loss_of_loved_one', 'no_support', 'vague_stress', 'sleep_issues']),
patient_statement=st.text(min_size=10, max_size=200),
context_clues=st.lists(st.text(min_size=5, max_size=50), min_size=1, max_size=3)
)
@settings(max_examples=50)
def test_scenario_specific_question_targeting(self, scenario_type: str, patient_statement: str, context_clues: List[str]):
"""
Test that questions are targeted to specific YELLOW scenarios.
Property: Generated questions should address the specific ambiguity
relevant to each scenario type (emotional vs external factors).
"""
from config.prompt_management.data_models import YellowScenario, ScenarioType
# Create scenario based on the type
try:
scenario_enum = ScenarioType(scenario_type)
except ValueError:
# Skip invalid scenario types
return
scenario = YellowScenario(
scenario_type=scenario_enum,
patient_statement=patient_statement,
context_clues=context_clues,
target_clarification=f"Clarify if {scenario_type} causes emotional distress",
question_patterns=[]
)
# Property assertion: Scenario should have valid structure
assert scenario.scenario_type == scenario_enum
assert len(scenario.patient_statement) >= 10
assert len(scenario.context_clues) >= 1
# Property assertion: Target clarification should be scenario-specific
assert scenario_type in scenario.target_clarification.lower()
@given(
loss_statements=st.lists(
st.sampled_from([
"I used to love gardening, but now I can't",
"I don't enjoy reading anymore",
"I stopped playing music",
"I can't do my hobbies like before"
]),
min_size=1, max_size=2
)
)
@settings(max_examples=30)
def test_loss_of_interest_question_patterns(self, loss_statements: List[str]):
"""
Test that loss of interest scenarios generate appropriate questions.
Property: Questions for loss of interest should distinguish between
emotional impact and practical circumstances.
"""
# Expected question elements for loss of interest scenarios
expected_elements = [
"emotional", "emotionally", "weighing", "circumstances",
"time", "practical", "meaningful", "distressing"
]
for statement in loss_statements:
# Property assertion: Statement should contain loss of interest indicators
loss_indicators = ["used to", "don't", "can't", "stopped"]
assert any(indicator in statement.lower() for indicator in loss_indicators), \
f"Statement should contain loss of interest indicators: {statement}"
# Property assertion: Should be classifiable as loss of interest scenario
engagement_indicators = ["love", "enjoy", "do", "playing", "hobbies"]
assert any(indicator in statement.lower() for indicator in engagement_indicators), \
f"Statement should express previous engagement: {statement}"
@given(
grief_statements=st.lists(
st.sampled_from([
"My mother passed away last month",
"I lost my husband recently",
"My father died",
"We had to put our dog down"
]),
min_size=1, max_size=2
)
)
@settings(max_examples=30)
def test_loss_of_loved_one_question_patterns(self, grief_statements: List[str]):
"""
Test that loss of loved one scenarios generate appropriate questions.
Property: Questions for grief should focus on coping mechanisms
and emotional state rather than practical arrangements.
"""
# Expected question elements for grief scenarios
expected_elements = [
"coping", "processing", "difficult", "emotionally",
"grief", "loss", "feeling", "support"
]
for statement in grief_statements:
# Property assertion: Statement should contain loss indicators
loss_indicators = ["passed away", "died", "lost", "put", "down"]
assert any(indicator in statement.lower() for indicator in loss_indicators), \
f"Statement should contain loss indicators: {statement}"
# Property assertion: Should reference a relationship
relationship_indicators = ["mother", "father", "husband", "wife", "dog", "cat"]
assert any(rel in statement.lower() for rel in relationship_indicators), \
f"Statement should reference a relationship: {statement}"
@given(
support_statements=st.lists(
st.sampled_from([
"I don't have anyone to help me",
"I'm all alone here",
"No one visits me anymore",
"I have no family nearby"
]),
min_size=1, max_size=2
)
)
@settings(max_examples=30)
def test_no_support_question_patterns(self, support_statements: List[str]):
"""
Test that lack of support scenarios generate appropriate questions.
Property: Questions should distinguish between practical isolation
and emotional distress from lack of support.
"""
# Expected question elements for support scenarios
expected_elements = [
"affecting", "emotionally", "practical", "challenge",
"support", "alone", "isolated", "help"
]
for statement in support_statements:
# Property assertion: Statement should contain isolation indicators
isolation_indicators = ["don't have", "alone", "no one", "no family"]
assert any(indicator in statement.lower() for indicator in isolation_indicators), \
f"Statement should contain isolation indicators: {statement}"
@given(
stress_statements=st.lists(
st.sampled_from([
"I feel some stress",
"Things are difficult",
"I'm a bit worried",
"It's been hard lately"
]),
min_size=1, max_size=2
)
)
@settings(max_examples=30)
def test_vague_stress_question_patterns(self, stress_statements: List[str]):
"""
Test that vague stress scenarios generate clarifying questions.
Property: Questions should identify specific causes of stress
to determine if it's emotional distress or external factors.
"""
# Expected question elements for vague stress scenarios
expected_elements = [
"causing", "source", "specifically", "what", "more about",
"tell me", "explain", "describe"
]
for statement in stress_statements:
# Property assertion: Statement should be vague about cause
vague_indicators = ["some", "a bit", "things", "it's been"]
assert any(indicator in statement.lower() for indicator in vague_indicators), \
f"Statement should be vague about cause: {statement}"
# Property assertion: Should mention stress/difficulty without specifics
stress_indicators = ["stress", "difficult", "worried", "hard"]
assert any(indicator in statement.lower() for indicator in stress_indicators), \
f"Statement should mention stress/difficulty: {statement}"
@given(
sleep_statements=st.lists(
st.sampled_from([
"I can't sleep at night",
"My mind won't stop racing",
"I have trouble sleeping",
"I wake up a lot"
]),
min_size=1, max_size=2
)
)
@settings(max_examples=30)
def test_sleep_issues_question_patterns(self, sleep_statements: List[str]):
"""
Test that sleep issue scenarios generate appropriate questions.
Property: Questions should distinguish between medical causes
and emotional/mental causes of sleep problems.
"""
# Expected question elements for sleep scenarios
expected_elements = [
"mind", "thoughts", "worrying", "medical", "medication",
"physical", "emotional", "keeping you awake"
]
for statement in sleep_statements:
# Property assertion: Statement should contain sleep indicators
sleep_indicators = ["sleep", "racing", "wake", "night"]
assert any(indicator in statement.lower() for indicator in sleep_indicators), \
f"Statement should contain sleep indicators: {statement}"
def test_question_effectiveness_validation(self):
"""
Test that question effectiveness can be validated.
Property: The system should be able to assess whether generated
questions effectively target the intended clarification.
"""
from config.prompt_management.data_models import ScenarioType
# Test scenarios with expected effectiveness
test_cases = [
{
"scenario": ScenarioType.LOSS_OF_INTEREST,
"good_question": "Is that something that's been weighing on you emotionally, or is it more about time or circumstances?",
"poor_question": "How are you feeling about that?",
"expected_better": "good_question"
},
{
"scenario": ScenarioType.VAGUE_STRESS,
"good_question": "Can you tell me more about what's been causing that stress?",
"poor_question": "That sounds difficult.",
"expected_better": "good_question"
}
]
for case in test_cases:
scenario = case["scenario"]
good_q = case["good_question"]
poor_q = case["poor_question"]
# Property assertion: Good questions should be more specific
assert len(good_q.split()) > len(poor_q.split()) or "what" in good_q.lower() or "how" in good_q.lower(), \
f"Good question should be more specific: {good_q}"
# Property assertion: Good questions should contain clarifying words
clarifying_words = ["what", "how", "why", "can you", "tell me", "more about"]
good_has_clarifying = any(word in good_q.lower() for word in clarifying_words)
poor_has_clarifying = any(word in poor_q.lower() for word in clarifying_words)
assert good_has_clarifying or not poor_has_clarifying, \
f"Good question should be more clarifying than poor question"
def test_question_language_matching(self):
"""
Test that questions match the patient's language.
Property: Generated questions should be in the same language
as the patient's input message.
"""
# This is a simplified test - in practice, language detection would be more complex
test_cases = [
{"input": "I feel stressed", "language": "english"},
{"input": "Je me sens stressé", "language": "french"},
{"input": "Me siento estresado", "language": "spanish"}
]
for case in test_cases:
input_text = case["input"]
expected_lang = case["language"]
# Property assertion: Input should be non-empty
assert len(input_text.strip()) > 0, "Input should be non-empty"
# Property assertion: Language should be identifiable
assert expected_lang in ["english", "french", "spanish"], "Language should be supported"
# In a real implementation, we would test that the generated question
# matches the detected language of the input
class TestComponentConsistency:
"""
**Feature: prompt-optimization, Property 1: Component Consistency Enforcement**
**Validates: Requirements 1.1, 1.2, 1.3, 1.4, 1.5**
Property: For any spiritual distress indicator or classification rule defined in shared components,
all AI agents (Spiritual_Monitor, Triage_Evaluator) should apply identical definitions,
terminology, and evaluation logic when processing the same message.
"""
@given(
message_content=st.text(min_size=10, max_size=200),
agent_types=st.lists(
st.sampled_from(['spiritual_monitor', 'triage_question', 'triage_evaluator']),
min_size=2, max_size=3, unique=True
)
)
@settings(max_examples=50)
def test_identical_shared_components_across_agents(self, message_content: str, agent_types: List[str]):
"""
Test that all AI agents receive identical shared components.
Property: When multiple AI agents request prompt configurations, they should
all receive identical shared indicators, rules, and category definitions.
"""
controller = PromptController()
# Get prompt configurations for different agents
configs = {}
for agent_type in agent_types:
configs[agent_type] = controller.get_prompt(agent_type)
# Property assertion: All agents should have identical shared indicators
if len(configs) > 1:
agent_names = list(configs.keys())
base_agent = agent_names[0]
base_indicators = {ind.name: ind.to_dict() for ind in configs[base_agent].shared_indicators}
for other_agent in agent_names[1:]:
other_indicators = {ind.name: ind.to_dict() for ind in configs[other_agent].shared_indicators}
# Check that indicator sets are identical
assert set(base_indicators.keys()) == set(other_indicators.keys()), \
f"Indicator sets differ between {base_agent} and {other_agent}"
# Check that indicator definitions are identical
for ind_name in base_indicators:
assert base_indicators[ind_name] == other_indicators[ind_name], \
f"Indicator {ind_name} differs between {base_agent} and {other_agent}"
# Property assertion: All agents should have identical shared rules
if len(configs) > 1:
base_rules = {rule.rule_id: rule.to_dict() for rule in configs[base_agent].shared_rules}
for other_agent in agent_names[1:]:
other_rules = {rule.rule_id: rule.to_dict() for rule in configs[other_agent].shared_rules}
# Check that rule sets are identical
assert set(base_rules.keys()) == set(other_rules.keys()), \
f"Rule sets differ between {base_agent} and {other_agent}"
# Check that rule definitions are identical
for rule_id in base_rules:
assert base_rules[rule_id] == other_rules[rule_id], \
f"Rule {rule_id} differs between {base_agent} and {other_agent}"
@given(
category_name=st.sampled_from(['GREEN', 'YELLOW', 'RED']),
agent_types=st.lists(
st.sampled_from(['spiritual_monitor', 'triage_question', 'triage_evaluator']),
min_size=2, max_size=3, unique=True
)
)
@settings(max_examples=30)
def test_consistent_category_definitions(self, category_name: str, agent_types: List[str]):
"""
Test that category definitions are consistent across all agents.
Property: All AI agents should use identical category definitions
for GREEN, YELLOW, and RED classifications.
"""
controller = PromptController()
# Get category definition from shared components
category_def = controller.category_definitions.get_category_definition(category_name)
assert category_def is not None, f"Category {category_name} should be defined"
# Verify all agents have access to the same category definitions
for agent_type in agent_types:
config = controller.get_prompt(agent_type)
# The category definitions should be accessible through the controller
agent_category_def = controller.category_definitions.get_category_definition(category_name)
# Property assertion: Category definitions should be identical
assert agent_category_def == category_def, \
f"Category {category_name} definition differs for agent {agent_type}"
def test_terminology_consistency_validation(self):
"""
Test that terminology validation catches inconsistencies.
Property: The validation system should detect when different agents
use inconsistent terminology for the same concepts.
"""
controller = PromptController()
# Run consistency validation
validation_result = controller.validate_consistency()
# Property assertion: Validation should complete successfully
assert isinstance(validation_result, ValidationResult), \
"Validation should return a ValidationResult object"
# If there are errors, they should be specific and actionable
for error in validation_result.errors:
assert isinstance(error, str) and len(error) > 0, \
"Validation errors should be non-empty strings"
# Warnings should also be specific
for warning in validation_result.warnings:
assert isinstance(warning, str) and len(warning) > 0, \
"Validation warnings should be non-empty strings"
@given(
indicator_updates=st.lists(
st.tuples(
st.text(min_size=3, max_size=30, alphabet=st.characters(whitelist_categories=('Lu', 'Ll', 'Nd', 'Pc'))),
st.text(min_size=10, max_size=100),
st.floats(min_value=0.0, max_value=1.0)
),
min_size=1, max_size=3
)
)
@settings(max_examples=20)
def test_update_propagation_consistency(self, indicator_updates: List[tuple]):
"""
Test that updates to shared components propagate consistently.
Property: When shared components are updated, all dependent AI agents
should receive the updates in the same way.
"""
controller = PromptController()
# Apply updates to indicators
added_indicators = []
for name, definition, weight in indicator_updates:
indicator = Indicator(
name=f"test_{name}",
category=IndicatorCategory.EMOTIONAL,
definition=definition,
examples=[f"Example for {name}"],
severity_weight=weight
)
success = controller.indicator_catalog.add_indicator(indicator)
if success:
added_indicators.append(indicator.name)
if not added_indicators:
return # Skip if no indicators were added
# Clear cache to force reload
controller._prompt_cache.clear()
# Get configurations for multiple agents
agent_types = ['spiritual_monitor', 'triage_question', 'triage_evaluator']
configs = {agent: controller.get_prompt(agent) for agent in agent_types}
# Property assertion: All agents should have the same updated indicators
for indicator_name in added_indicators:
for agent_type in agent_types:
agent_indicators = {ind.name: ind for ind in configs[agent_type].shared_indicators}
assert indicator_name in agent_indicators, \
f"Agent {agent_type} missing updated indicator: {indicator_name}"
# Clean up
for indicator_name in added_indicators:
controller.indicator_catalog.remove_indicator(indicator_name)
def test_rule_priority_consistency(self):
"""
Test that rule priorities are applied consistently across agents.
Property: All agents should receive rules in the same priority order
and apply them consistently.
"""
controller = PromptController()
# Get rules from multiple agents
agent_types = ['spiritual_monitor', 'triage_question', 'triage_evaluator']
rule_orders = {}
for agent_type in agent_types:
config = controller.get_prompt(agent_type)
# Sort rules by priority (lower number = higher priority)
sorted_rules = sorted(config.shared_rules, key=lambda r: r.priority)
rule_orders[agent_type] = [rule.rule_id for rule in sorted_rules]
# Property assertion: All agents should have the same rule order
if len(rule_orders) > 1:
agent_names = list(rule_orders.keys())
base_order = rule_orders[agent_names[0]]
for other_agent in agent_names[1:]:
other_order = rule_orders[other_agent]
assert base_order == other_order, \
f"Rule priority order differs between {agent_names[0]} and {other_agent}"
class TestConsentLanguageCompliance:
"""
**Feature: prompt-optimization, Property 4: Consent-Based Language Compliance**
**Validates: Requirements 4.1, 4.2, 4.3, 4.4, 4.5**
Property: For any RED classification or consent interaction, the system should generate
messages using only approved non-assumptive language patterns and handle patient responses
(acceptance, decline, ambiguity) appropriately.
"""
@given(
consent_contexts=st.lists(
st.tuples(
st.sampled_from(['high', 'medium', 'low']), # distress_level
st.booleans(), # previous_spiritual_mention
st.text(min_size=10, max_size=100) # additional_context
),
min_size=1,
max_size=5
)
)
@settings(max_examples=100)
def test_consent_message_language_compliance(self, consent_contexts):
"""
Test that all generated consent messages comply with non-assumptive language requirements.
Property: All consent messages should use approved language patterns and avoid
assumptive, pressuring, or religiously presumptive language.
"""
from config.prompt_management.consent_manager import ConsentManager, ConsentMessageType
consent_manager = ConsentManager()
for distress_level, spiritual_mention, context_text in consent_contexts:
context = {
'distress_level': distress_level,
'previous_spiritual_mention': spiritual_mention,
'context_text': context_text
}
# Test all message types
message_types = [
ConsentMessageType.INITIAL_REQUEST,
ConsentMessageType.CLARIFICATION,
ConsentMessageType.CONFIRMATION,
ConsentMessageType.DECLINE_ACKNOWLEDGMENT
]
for message_type in message_types:
# Generate message
message = consent_manager.generate_consent_message(message_type, context)
# Property assertion: Message should not be empty
assert len(message.strip()) > 0, f"Generated message should not be empty for {message_type}"
# Property assertion: Message should comply with language requirements
is_compliant, violations = consent_manager.validate_language_compliance(message)
assert is_compliant, f"Message violates language compliance: {violations}. Message: '{message}'"
# Property assertion: Message should contain respectful language
assert consent_manager._contains_respectful_language(message), \
f"Message should contain respectful language: '{message}'"
@given(
patient_responses=st.lists(
st.tuples(
st.sampled_from([
"Yes, I would like that",
"No, I'm fine",
"I don't know, maybe",
"What would that involve?",
"I'm not sure",
"No thanks",
"That sounds good",
"I guess so",
"Not interested",
"Tell me more about it"
]),
st.text(min_size=5, max_size=20, alphabet=st.characters(whitelist_categories=('Lu', 'Ll', 'Nd', 'Pc', 'Zs'))) # session_id
),
min_size=1,
max_size=10
)
)
@settings(max_examples=50)
def test_patient_response_handling(self, patient_responses):
"""
Test that patient responses are handled appropriately based on their classification.
Property: Patient responses should be correctly classified and handled with
appropriate next steps (accept -> referral, decline -> medical dialogue, ambiguous -> clarification).
"""
from config.prompt_management.consent_manager import ConsentManager, ConsentResponse
consent_manager = ConsentManager()
for response_text, session_id in patient_responses:
# Handle the consent interaction
result = consent_manager.handle_consent_interaction(response_text, session_id)
# Property assertion: Result should have required fields
required_fields = ['action', 'message', 'generate_provider_summary', 'log_referral', 'interaction']
for field in required_fields:
assert field in result, f"Result missing required field: {field}"
# Property assertion: Action should be valid
valid_actions = ['proceed_with_referral', 'return_to_medical_dialogue', 'request_clarification']
assert result['action'] in valid_actions, f"Invalid action: {result['action']}"
# Property assertion: Response message should be non-empty and compliant
response_message = result['message']
assert len(response_message.strip()) > 0, "Response message should not be empty"
is_compliant, violations = consent_manager.validate_language_compliance(response_message)
assert is_compliant, f"Response message violates compliance: {violations}. Message: '{response_message}'"
# Property assertion: Interaction should be properly recorded
interaction = result['interaction']
assert 'interaction_id' in interaction, "Interaction should have ID"
assert 'patient_response' in interaction, "Interaction should record patient response"
assert interaction['patient_response'] == response_text, "Should record original response"
# Property assertion: Actions should be consistent with response classification
response_classification = ConsentResponse(interaction['response_classification'])
if response_classification == ConsentResponse.ACCEPT:
assert result['action'] == 'proceed_with_referral', "Accept should proceed with referral"
assert result['generate_provider_summary'] == True, "Accept should generate summary"
assert result['log_referral'] == True, "Accept should log referral"
elif response_classification == ConsentResponse.DECLINE:
assert result['action'] == 'return_to_medical_dialogue', "Decline should return to medical dialogue"
assert result['generate_provider_summary'] == False, "Decline should not generate summary"
assert result['log_referral'] == False, "Decline should not log referral"
elif response_classification in [ConsentResponse.AMBIGUOUS, ConsentResponse.UNCLEAR]:
assert result['action'] == 'request_clarification', "Ambiguous should request clarification"
assert result['generate_provider_summary'] == False, "Ambiguous should not generate summary"
assert result['log_referral'] == False, "Ambiguous should not log referral"
assert result.get('requires_follow_up') == True, "Ambiguous should require follow-up"
@given(
ambiguous_responses=st.lists(
st.sampled_from([
"I don't know",
"Maybe",
"What would that involve?",
"Tell me more",
"I'm not sure",
"What do you think?",
"What kind of support?",
"I need to think about it"
]),
min_size=1,
max_size=5
)
)
@settings(max_examples=30)
def test_clarification_question_generation(self, ambiguous_responses):
"""
Test that clarifying questions are generated appropriately for ambiguous responses.
Property: Clarifying questions should be contextually appropriate, non-assumptive,
and help patients make informed decisions about spiritual care.
"""
from config.prompt_management.consent_manager import ConsentManager
consent_manager = ConsentManager()
for response in ambiguous_responses:
# Generate clarification question
clarification = consent_manager.generate_clarification_question(response)
# Property assertion: Clarification should not be empty
assert len(clarification.strip()) > 0, "Clarification question should not be empty"
# Property assertion: Clarification should be compliant
is_compliant, violations = consent_manager.validate_language_compliance(clarification)
assert is_compliant, f"Clarification violates compliance: {violations}. Question: '{clarification}'"
# Property assertion: Clarification should be respectful
assert consent_manager._contains_respectful_language(clarification), \
f"Clarification should be respectful: '{clarification}'"
# Property assertion: Clarification should be contextually appropriate
response_lower = response.lower()
clarification_lower = clarification.lower()
# Information-seeking responses should get informative clarifications
if any(word in response_lower for word in ['what', 'how', 'tell me', 'involve']):
assert any(word in clarification_lower for word in ['chaplain', 'counselor', 'support', 'team']), \
f"Information-seeking response should get informative clarification: '{clarification}'"
# Uncertainty responses should get supportive clarifications
elif any(word in response_lower for word in ['maybe', 'not sure', 'don\'t know']):
assert any(word in clarification_lower for word in ['no pressure', 'okay', 'comfortable']), \
f"Uncertainty response should get supportive clarification: '{clarification}'"
@given(
test_messages=st.lists(
st.text(min_size=10, max_size=200),
min_size=1,
max_size=10
)
)
@settings(max_examples=50)
def test_language_validation_accuracy(self, test_messages):
"""
Test that language validation accurately identifies compliant and non-compliant messages.
Property: The validation system should correctly identify assumptive language,
pressure tactics, and religious assumptions in messages.
"""
from config.prompt_management.consent_manager import ConsentManager
consent_manager = ConsentManager()
# Test with known compliant messages
compliant_messages = [
"Would you be interested in speaking with someone from our spiritual care team?",
"Our spiritual care team is available if you'd like to connect with them.",
"I understand and respect your decision.",
"Could you help me understand what would be most helpful for you?"
]
for message in compliant_messages:
is_compliant, violations = consent_manager.validate_language_compliance(message)
assert is_compliant, f"Known compliant message should pass validation: '{message}'. Violations: {violations}"
# Test with known non-compliant messages
non_compliant_messages = [
"You need to speak with someone from spiritual care.",
"This will help you feel better.",
"Obviously you're struggling with faith issues.",
"You should pray about this.",
"God will help you through this."
]
for message in non_compliant_messages:
is_compliant, violations = consent_manager.validate_language_compliance(message)
assert not is_compliant, f"Known non-compliant message should fail validation: '{message}'"
assert len(violations) > 0, f"Non-compliant message should have violations: '{message}'"
# Test generated messages
for message in test_messages:
is_compliant, violations = consent_manager.validate_language_compliance(message)
# Property assertion: Validation should return boolean and list
assert isinstance(is_compliant, bool), "Validation should return boolean"
assert isinstance(violations, list), "Violations should be a list"
# Property assertion: If not compliant, should have violations
if not is_compliant:
assert len(violations) > 0, f"Non-compliant message should have violations: '{message}'"
# Property assertion: Violations should be descriptive
for violation in violations:
assert isinstance(violation, str), "Violations should be strings"
assert len(violation) > 0, "Violations should be non-empty"
class TestStructuredFeedbackCapture:
"""
**Feature: prompt-optimization, Property 3: Structured Feedback Data Capture**
**Validates: Requirements 3.1, 3.2, 3.3, 3.4, 3.5**
Property: For any system issue (classification error, question problem, referral issue),
the feedback system should capture all predefined structured data fields and store them
in analyzable format according to documentation categories.
"""
@given(
classification_errors=st.lists(
st.tuples(
st.sampled_from(['wrong_classification', 'severity_misjudgment', 'missed_indicators', 'false_positive']),
st.sampled_from(['green_to_yellow', 'yellow_to_green', 'red_to_green', 'underestimated_distress']),
st.sampled_from(['GREEN', 'YELLOW', 'RED']),
st.sampled_from(['GREEN', 'YELLOW', 'RED']),
st.text(min_size=20, max_size=200),
st.text(min_size=10, max_size=100),
st.floats(min_value=0.0, max_value=1.0)
),
min_size=1,
max_size=5
),
question_issues=st.lists(
st.tuples(
st.sampled_from(['inappropriate_question', 'insensitive_language', 'wrong_scenario_targeting']),
st.text(min_size=10, max_size=100),
st.sampled_from(['loss_of_interest', 'loss_of_loved_one', 'no_support']),
st.text(min_size=10, max_size=100),
st.sampled_from(['low', 'medium', 'high'])
),
min_size=0,
max_size=3
),
referral_problems=st.lists(
st.tuples(
st.sampled_from(['incomplete_summary', 'missing_contact_info', 'incorrect_urgency']),
st.text(min_size=20, max_size=150),
st.text(min_size=10, max_size=100),
st.sampled_from(['low', 'medium', 'high'])
),
min_size=0,
max_size=3
)
)
@settings(max_examples=100)
def test_structured_feedback_data_capture(self, classification_errors, question_issues, referral_problems):
"""
Test that the feedback system captures all predefined structured data fields
and stores them in analyzable format according to documentation categories.
"""
from config.prompt_management.feedback_system import FeedbackSystem
from config.prompt_management.data_models import ErrorType, ErrorSubcategory, QuestionIssueType, ReferralProblemType, ScenarioType
# Create feedback system with temporary storage
import tempfile
with tempfile.TemporaryDirectory() as temp_dir:
feedback_system = FeedbackSystem(storage_path=temp_dir)
recorded_error_ids = []
recorded_question_ids = []
recorded_referral_ids = []
# Record classification errors
for error_type_str, subcategory_str, expected, actual, message, comments, confidence in classification_errors:
error_id = feedback_system.record_classification_error(
error_type=ErrorType(error_type_str),
subcategory=ErrorSubcategory(subcategory_str),
expected_category=expected,
actual_category=actual,
message_content=message,
reviewer_comments=comments,
confidence_level=confidence,
session_id="test_session",
additional_context={"test": True}
)
recorded_error_ids.append(error_id)
# Record question issues
for issue_type_str, question, scenario_str, comments, severity in question_issues:
issue_id = feedback_system.record_question_issue(
issue_type=QuestionIssueType(issue_type_str),
question_content=question,
scenario_type=ScenarioType(scenario_str),
reviewer_comments=comments,
severity=severity,
session_id="test_session"
)
recorded_question_ids.append(issue_id)
# Record referral problems
for problem_type_str, referral, comments, severity in referral_problems:
problem_id = feedback_system.record_referral_problem(
problem_type=ReferralProblemType(problem_type_str),
referral_content=referral,
reviewer_comments=comments,
severity=severity,
session_id="test_session",
missing_fields=["contact_info", "urgency_level"]
)
recorded_referral_ids.append(problem_id)
# Verify all data was captured with required fields
summary = feedback_system.get_feedback_summary()
# Property assertion: All classification errors should be recorded
assert summary['total_errors'] == len(classification_errors), "All classification errors should be recorded"
assert len(recorded_error_ids) == len(classification_errors), "All error IDs should be returned"
# Property assertion: All question issues should be recorded
assert summary['total_question_issues'] == len(question_issues), "All question issues should be recorded"
assert len(recorded_question_ids) == len(question_issues), "All question issue IDs should be returned"
# Property assertion: All referral problems should be recorded
assert summary['total_referral_problems'] == len(referral_problems), "All referral problems should be recorded"
assert len(recorded_referral_ids) == len(referral_problems), "All referral problem IDs should be returned"
# Property assertion: Structured data fields are present and valid
if classification_errors:
errors = feedback_system._load_errors()
for error in errors:
# Required fields must be present
required_fields = ['error_id', 'error_type', 'subcategory', 'expected_category',
'actual_category', 'message_content', 'reviewer_comments',
'confidence_level', 'timestamp']
for field in required_fields:
assert field in error, f"Required field {field} missing from error record"
# Verify data types and constraints
assert isinstance(error['confidence_level'], (int, float)), "Confidence level must be numeric"
assert 0.0 <= error['confidence_level'] <= 1.0, "Confidence level must be between 0.0 and 1.0"
assert error['expected_category'] in ['GREEN', 'YELLOW', 'RED'], "Expected category must be valid"
assert error['actual_category'] in ['GREEN', 'YELLOW', 'RED'], "Actual category must be valid"
assert len(error['error_id']) > 0, "Error ID must be non-empty"
assert len(error['message_content']) >= 20, "Message content must meet minimum length"
# Property assertion: Error pattern analysis works with sufficient data
if len(classification_errors) >= 2:
patterns = feedback_system.analyze_error_patterns(min_frequency=1)
assert isinstance(patterns, list), "Pattern analysis should return list"
# Verify pattern structure
for pattern in patterns:
pattern_dict = pattern.to_dict()
assert 'pattern_id' in pattern_dict, "Pattern must have ID"
assert 'frequency' in pattern_dict, "Pattern must have frequency"
assert 'suggested_improvements' in pattern_dict, "Pattern must have suggestions"
assert pattern_dict['frequency'] >= 1, "Pattern frequency must be positive"
assert isinstance(pattern_dict['suggested_improvements'], list), "Suggestions must be list"
# Property assertion: Improvement suggestions generation works
suggestions = feedback_system.generate_improvement_suggestions()
assert isinstance(suggestions, list), "Suggestions should be a list"
assert all(isinstance(s, str) for s in suggestions), "All suggestions should be strings"
assert all(len(s) > 0 for s in suggestions), "All suggestions should be non-empty"
@given(
error_patterns=st.lists(
st.tuples(
st.sampled_from(['wrong_classification', 'severity_misjudgment']),
st.integers(min_value=3, max_value=5) # Reduced max to avoid accumulation issues
),
min_size=1,
max_size=2 # Reduced to avoid complex interactions
)
)
@settings(max_examples=30) # Reduced examples for faster testing
def test_error_pattern_analysis_accuracy(self, error_patterns):
"""
Test that error pattern analysis correctly identifies frequent error types.
Property: When multiple errors of the same type are recorded, the pattern
analysis should identify them as significant patterns with appropriate
improvement suggestions.
"""
from config.prompt_management.feedback_system import FeedbackSystem
from config.prompt_management.data_models import ErrorType, ErrorSubcategory
import tempfile
with tempfile.TemporaryDirectory() as temp_dir:
feedback_system = FeedbackSystem(storage_path=temp_dir)
# Record multiple errors of each pattern type
total_recorded = {}
for error_type_str, frequency in error_patterns:
total_recorded[error_type_str] = total_recorded.get(error_type_str, 0) + frequency
for i in range(frequency):
feedback_system.record_classification_error(
error_type=ErrorType(error_type_str),
subcategory=ErrorSubcategory.GREEN_TO_YELLOW if error_type_str == 'wrong_classification' else ErrorSubcategory.UNDERESTIMATED_DISTRESS,
expected_category="YELLOW",
actual_category="GREEN",
message_content=f"Unique test message {error_type_str}_{i}_{hash(str(error_patterns))}",
reviewer_comments=f"Test comment {i}",
confidence_level=0.8,
session_id=f"test_session_{error_type_str}_{i}"
)
# Analyze patterns
patterns = feedback_system.analyze_error_patterns(min_frequency=3)
# Property assertion: Patterns should be identified for frequent error types
pattern_types = [p.pattern_type for p in patterns]
for error_type_str, total_freq in total_recorded.items():
if total_freq >= 3:
expected_pattern = f"error_type_{error_type_str}"
assert any(expected_pattern in pt for pt in pattern_types), \
f"Pattern should be identified for frequent error type: {error_type_str}"
# Property assertion: All patterns should have improvement suggestions
for pattern in patterns:
assert len(pattern.suggested_improvements) > 0, f"Pattern {pattern.pattern_type} should have improvement suggestions"
for suggestion in pattern.suggested_improvements:
assert len(suggestion) > 5, f"Suggestions should be meaningful: '{suggestion}'"
@given(
feedback_categories=st.lists(
st.sampled_from(['classification_error', 'question_issue', 'referral_problem']),
min_size=1,
max_size=10
)
)
@settings(max_examples=30)
def test_feedback_summary_completeness(self, feedback_categories):
"""
Test that feedback summaries include all required information categories.
Property: Feedback summaries should provide comprehensive statistics
and insights across all types of recorded feedback.
"""
from config.prompt_management.feedback_system import FeedbackSystem
from config.prompt_management.data_models import ErrorType, ErrorSubcategory, QuestionIssueType, ReferralProblemType, ScenarioType
import tempfile
with tempfile.TemporaryDirectory() as temp_dir:
feedback_system = FeedbackSystem(storage_path=temp_dir)
# Record different types of feedback based on categories
for category in feedback_categories:
if category == 'classification_error':
feedback_system.record_classification_error(
error_type=ErrorType.WRONG_CLASSIFICATION,
subcategory=ErrorSubcategory.GREEN_TO_YELLOW,
expected_category="YELLOW",
actual_category="GREEN",
message_content="Test classification error message",
reviewer_comments="Test classification error comment",
confidence_level=0.9
)
elif category == 'question_issue':
feedback_system.record_question_issue(
issue_type=QuestionIssueType.INAPPROPRIATE_QUESTION,
question_content="Test inappropriate question",
scenario_type=ScenarioType.LOSS_OF_INTEREST,
reviewer_comments="Test question issue comment",
severity="medium"
)
elif category == 'referral_problem':
feedback_system.record_referral_problem(
problem_type=ReferralProblemType.INCOMPLETE_SUMMARY,
referral_content="Test incomplete referral summary",
reviewer_comments="Test referral problem comment",
severity="high"
)
# Get feedback summary
summary = feedback_system.get_feedback_summary()
# Property assertion: Summary should contain all required fields
required_fields = [
'total_errors', 'total_question_issues', 'total_referral_problems',
'error_types', 'error_subcategories', 'question_issue_types',
'referral_problem_types', 'average_confidence', 'recent_errors',
'improvement_suggestions'
]
for field in required_fields:
assert field in summary, f"Summary missing required field: {field}"
# Property assertion: Counts should match recorded feedback
classification_count = feedback_categories.count('classification_error')
question_count = feedback_categories.count('question_issue')
referral_count = feedback_categories.count('referral_problem')
assert summary['total_errors'] == classification_count, "Error count should match recorded errors"
assert summary['total_question_issues'] == question_count, "Question issue count should match"
assert summary['total_referral_problems'] == referral_count, "Referral problem count should match"
# Property assertion: Statistics should be valid
if classification_count > 0:
assert 0.0 <= summary['average_confidence'] <= 1.0, "Average confidence should be valid"
assert isinstance(summary['error_types'], dict), "Error types should be dictionary"
assert isinstance(summary['error_subcategories'], dict), "Error subcategories should be dictionary"
# Property assertion: Improvement suggestions should be provided
assert isinstance(summary['improvement_suggestions'], list), "Improvement suggestions should be list"
if __name__ == "__main__":
# Run tests directly
import subprocess
import sys
# Install hypothesis if not available
try:
import hypothesis
except ImportError:
print("Installing hypothesis for property-based testing...")
subprocess.check_call([sys.executable, "-m", "pip", "install", "hypothesis"])
import hypothesis
# Run the tests
pytest.main([__file__, "-v"])
class TestContextAwareClassification:
"""
**Feature: prompt-optimization, Property 6: Context-Aware Classification Logic**
**Validates: Requirements 6.1, 6.2, 6.3, 6.4, 6.5**
Property: For any patient message with conversation history containing distress indicators,
the classification should appropriately weight historical context against current statements,
detect defensive patterns, and generate contextually relevant follow-up questions.
"""
@given(
conversation_scenarios=st.lists(
st.tuples(
st.lists(st.text(min_size=10, max_size=100), min_size=1, max_size=5), # previous_messages
st.lists(st.sampled_from(['GREEN', 'YELLOW', 'RED']), min_size=1, max_size=5), # previous_classifications
st.lists(st.text(min_size=5, max_size=30), min_size=0, max_size=3), # distress_indicators
st.text(min_size=10, max_size=100) # current_message
),
min_size=1,
max_size=5
)
)
@settings(max_examples=100)
def test_context_aware_classification_with_history(self, conversation_scenarios):
"""
Test that classification considers conversation history appropriately.
Property: When patient previously expressed distress and now says "I'm fine",
the system should classify as YELLOW for verification.
"""
from config.prompt_management.context_aware_classifier import ContextAwareClassifier
classifier = ContextAwareClassifier()
for prev_messages, prev_classifications, distress_indicators, current_message in conversation_scenarios:
# Ensure lists are same length
min_len = min(len(prev_messages), len(prev_classifications))
prev_messages = prev_messages[:min_len]
prev_classifications = prev_classifications[:min_len]
# Build conversation history
history = ConversationHistory(
messages=[
Message(content=msg, classification=cls, timestamp=datetime.now())
for msg, cls in zip(prev_messages, prev_classifications)
],
distress_indicators_found=distress_indicators,
context_flags=[]
)
# Classify with context
result = classifier.classify_with_context(current_message, history)
# Property assertion: Result should have required fields
assert isinstance(result, Classification), "Result should be Classification object"
assert result.category in ['GREEN', 'YELLOW', 'RED'], "Category should be valid"
assert 0.0 <= result.confidence <= 1.0, "Confidence should be between 0 and 1"
# Property assertion: Historical distress should influence classification
if distress_indicators and any(cls in ['YELLOW', 'RED'] for cls in prev_classifications):
# If there's historical distress and current message is dismissive
dismissive_phrases = ['fine', 'okay', 'good', 'better', 'no problem']
if any(phrase in current_message.lower() for phrase in dismissive_phrases):
# Should be at least YELLOW for verification
assert result.category in ['YELLOW', 'RED'], \
f"Historical distress with dismissive response should be YELLOW/RED, got {result.category}"
assert 'historical_context' in result.reasoning.lower() or 'previous' in result.reasoning.lower(), \
"Reasoning should mention historical context"
@given(
defensive_scenarios=st.lists(
st.tuples(
st.sampled_from([
"I'm fine",
"Everything is okay",
"No problems here",
"I don't need help",
"It's all good"
]),
st.lists(st.sampled_from(['YELLOW', 'RED']), min_size=1, max_size=3),
st.integers(min_value=1, max_value=5) # number of previous distress mentions
),
min_size=1,
max_size=5
)
)
@settings(max_examples=50)
def test_defensive_response_detection(self, defensive_scenarios):
"""
Test that defensive responses are detected when they contradict history.
Property: When conversation context contains distress indicators and patient
gives defensive responses, the system should detect the pattern.
"""
from config.prompt_management.context_aware_classifier import ContextAwareClassifier
classifier = ContextAwareClassifier()
for defensive_message, prev_classifications, distress_count in defensive_scenarios:
# Build history with distress
history = ConversationHistory(
messages=[
Message(
content=f"I'm feeling stressed about things {i}",
classification=prev_classifications[i % len(prev_classifications)],
timestamp=datetime.now()
)
for i in range(distress_count)
],
distress_indicators_found=['stress', 'anxiety', 'worried'] * distress_count,
context_flags=['distress_expressed']
)
# Detect defensive pattern
is_defensive = classifier.detect_defensive_responses(defensive_message, history)
# Property assertion: Should detect defensive pattern with sufficient history
if distress_count >= 2:
assert isinstance(is_defensive, bool), "Detection should return boolean"
# With clear distress history and dismissive current message, should detect defensiveness
assert is_defensive == True, \
f"Should detect defensive pattern with {distress_count} distress mentions and message: '{defensive_message}'"
@given(
contextual_indicators=st.lists(
st.tuples(
st.text(min_size=5, max_size=30), # indicator_name
st.floats(min_value=0.0, max_value=1.0), # base_weight
st.integers(min_value=0, max_value=5), # historical_mentions
st.booleans() # recent_mention
),
min_size=1,
max_size=5
)
)
@settings(max_examples=50)
def test_contextual_indicator_weighting(self, contextual_indicators):
"""
Test that indicators are weighted based on conversation context.
Property: Indicators that appear repeatedly in conversation history
should receive higher weight in classification decisions.
"""
from config.prompt_management.context_aware_classifier import ContextAwareClassifier
classifier = ContextAwareClassifier()
for indicator_name, base_weight, historical_mentions, recent_mention in contextual_indicators:
context = {
'historical_mentions': historical_mentions,
'recent_mention': recent_mention,
'conversation_length': 5
}
# Evaluate contextual weight
contextual_weight = classifier.evaluate_contextual_indicators(
[indicator_name],
context
)
# Property assertion: Weight should be numeric and valid
assert isinstance(contextual_weight, (int, float)), "Weight should be numeric"
assert contextual_weight >= 0.0, "Weight should be non-negative"
# Property assertion: Historical mentions should increase weight
if historical_mentions >= 2:
# Weight should be higher than minimum for repeated indicators
assert contextual_weight >= 0.5, \
f"Repeated indicator should have weight >= 0.5, got {contextual_weight}"
# Property assertion: Recent mentions should have stronger influence
if recent_mention and historical_mentions > 0:
# Recent + historical should have reasonable weight
assert contextual_weight >= 0.6, \
f"Recent mention with history should have weight >= 0.6, got {contextual_weight}"
@given(
follow_up_scenarios=st.lists(
st.tuples(
st.text(min_size=10, max_size=100), # current_message
st.lists(st.text(min_size=10, max_size=50), min_size=1, max_size=3), # previous_topics
st.sampled_from(['YELLOW', 'RED']) # classification
),
min_size=1,
max_size=5
)
)
@settings(max_examples=50)
def test_contextual_follow_up_generation(self, follow_up_scenarios):
"""
Test that follow-up questions reference conversation context.
Property: When follow-up questions are generated, they should reference
previous conversation elements appropriately.
"""
from config.prompt_management.context_aware_classifier import ContextAwareClassifier
classifier = ContextAwareClassifier()
for current_message, previous_topics, classification in follow_up_scenarios:
# Build history
history = ConversationHistory(
messages=[
Message(content=topic, classification='YELLOW', timestamp=datetime.now())
for topic in previous_topics
],
distress_indicators_found=['stress', 'worry'],
context_flags=['follow_up_needed']
)
# Generate contextual follow-up
follow_up = classifier.generate_contextual_follow_up(
current_message,
history,
classification
)
# Property assertion: Follow-up should not be empty
assert len(follow_up.strip()) > 0, "Follow-up question should not be empty"
# Property assertion: Follow-up should be a question
assert '?' in follow_up, "Follow-up should be a question"
# Property assertion: Follow-up should reference context when appropriate
if len(previous_topics) >= 2:
# With sufficient history, should reference previous conversation
contextual_words = ['earlier', 'mentioned', 'said', 'discussed', 'talked about', 'before']
has_context_reference = any(word in follow_up.lower() for word in contextual_words)
# Note: Not all follow-ups need explicit references, but many should
# This is a soft assertion - we just check the capability exists
assert isinstance(has_context_reference, bool), "Should check for context references"
def test_medical_context_integration(self):
"""
Test that medical context is considered in classification.
Property: When mental health conditions are mentioned in medical context,
the system should consider this information in classification.
"""
from config.prompt_management.context_aware_classifier import ContextAwareClassifier
classifier = ContextAwareClassifier()
# Test scenarios with medical context
test_cases = [
{
'message': "I'm managing my anxiety with medication",
'medical_context': {'conditions': ['anxiety disorder'], 'medications': ['SSRI']},
'expected_consideration': True
},
{
'message': "I feel stressed about work",
'medical_context': {'conditions': ['depression'], 'medications': []},
'expected_consideration': True
},
{
'message': "Everything is fine",
'medical_context': {'conditions': [], 'medications': []},
'expected_consideration': False
}
]
for case in test_cases:
history = ConversationHistory(
messages=[],
distress_indicators_found=[],
context_flags=[],
medical_context=case['medical_context']
)
result = classifier.classify_with_context(case['message'], history)
# Property assertion: Result should be valid
assert isinstance(result, Classification), "Should return Classification"
assert result.category in ['GREEN', 'YELLOW', 'RED'], "Category should be valid"
# Property assertion: Medical context should influence reasoning
if case['expected_consideration'] and case['medical_context']['conditions']:
# Reasoning should mention medical context when relevant
reasoning_lower = result.reasoning.lower()
medical_terms = ['medical', 'condition', 'medication', 'treatment', 'diagnosis']
# At least some awareness of medical context in reasoning
# This is a capability check, not a strict requirement for every case
assert isinstance(result.reasoning, str), "Reasoning should be string"
assert len(result.reasoning) > 0, "Reasoning should not be empty"
class TestProviderSummaryCompleteness:
"""
**Feature: prompt-optimization, Property 7: Complete Provider Summary Generation**
**Validates: Requirements 7.1, 7.2, 7.3, 7.4, 7.5**
Property: For any RED classification generating a referral, the provider summary should
contain all required information fields (contact info, distress indicators, reasoning,
triage context, conversation background) as specified in requirements.
"""
@given(
red_classifications=st.lists(
st.tuples(
st.lists(st.text(min_size=5, max_size=30), min_size=1, max_size=5), # indicators
st.text(min_size=20, max_size=200), # reasoning
st.floats(min_value=0.7, max_value=1.0), # confidence (high for RED)
st.text(min_size=5, max_size=50, alphabet=st.characters(whitelist_categories=('Lu', 'Ll', 'Nd', 'Pc', 'Zs'))), # patient_name
st.text(min_size=10, max_size=15, alphabet=st.characters(whitelist_categories=('Nd', 'Pc'))), # phone
st.lists(st.text(min_size=10, max_size=100), min_size=0, max_size=3), # triage_questions
st.lists(st.text(min_size=5, max_size=100), min_size=0, max_size=3), # triage_responses
st.text(min_size=20, max_size=300) # conversation_context
),
min_size=1,
max_size=5
)
)
@settings(max_examples=100)
def test_complete_provider_summary_generation(self, red_classifications):
"""
Test that provider summaries contain all required information fields.
Property: For any RED classification, the generated provider summary should
include patient contact information, distress indicators, reasoning,
triage context, and conversation background.
"""
from core.provider_summary_generator import ProviderSummaryGenerator
generator = ProviderSummaryGenerator()
for indicators, reasoning, confidence, patient_name, phone, triage_q, triage_r, context in red_classifications:
# Ensure triage questions and responses are same length
min_len = min(len(triage_q), len(triage_r))
triage_questions = triage_q[:min_len] if min_len > 0 else None
triage_responses = triage_r[:min_len] if min_len > 0 else None
# Generate provider summary
summary = generator.generate_summary(
indicators=indicators,
reasoning=reasoning,
confidence=confidence,
patient_name=patient_name,
patient_phone=phone,
triage_questions=triage_questions,
triage_responses=triage_responses,
conversation_context=context
)
# Property assertion: Required fields must be present (Requirement 7.1)
assert summary.patient_name == patient_name, "Should include patient contact information"
assert summary.patient_phone == phone, "Should include patient phone number"
# Property assertion: Distress indicators must be included (Requirement 7.2)
assert summary.indicators == indicators, "Should include specific distress indicators"
assert len(summary.indicators) > 0, "Should have at least one distress indicator"
# Property assertion: Classification reasoning must be provided (Requirement 7.3)
assert summary.reasoning == reasoning, "Should provide clear explanation of RED determination"
assert len(summary.reasoning) >= 20, "Reasoning should be sufficiently detailed"
# Property assertion: Triage context must be included when available (Requirement 7.4)
if triage_questions and triage_responses and min_len > 0:
assert len(summary.triage_context) == min_len, "Should include all triage question-answer pairs"
for i, exchange in enumerate(summary.triage_context):
assert 'question' in exchange, "Triage context should include questions"
assert 'response' in exchange, "Triage context should include responses"
assert exchange['question'] == triage_questions[i], "Should preserve original questions"
assert exchange['response'] == triage_responses[i], "Should preserve original responses"
# Property assertion: Conversation background must be included (Requirement 7.5)
assert summary.conversation_context == context, "Should provide relevant background context"
# Property assertion: Summary should be complete and valid
assert summary.classification == "RED", "Should be classified as RED"
assert summary.confidence == confidence, "Should preserve confidence level"
assert summary.generated_at is not None, "Should have generation timestamp"
# Property assertion: Summary should be serializable
summary_dict = summary.to_dict()
required_fields = [
'patient_name', 'patient_phone', 'situation_description', 'indicators',
'classification', 'confidence', 'reasoning', 'triage_context',
'conversation_context', 'generated_at'
]
for field in required_fields:
assert field in summary_dict, f"Summary dict should contain {field}"
# Property assertion: Situation description should be meaningful
assert len(summary.situation_description) > 0, "Should generate meaningful situation description"
# If indicators provided, they should be mentioned in situation
if indicators:
situation_lower = summary.situation_description.lower()
# At least some indicators should be reflected in the description
assert any(indicator.lower() in situation_lower for indicator in indicators[:2]), \
"Situation description should reflect key indicators"
@given(
summary_data=st.tuples(
st.lists(st.text(min_size=5, max_size=30), min_size=1, max_size=3), # indicators
st.text(min_size=20, max_size=100), # reasoning
st.floats(min_value=0.7, max_value=1.0), # confidence
st.text(min_size=5, max_size=30), # patient_name
st.text(min_size=10, max_size=15), # phone
st.lists(
st.tuples(
st.text(min_size=10, max_size=50), # question
st.text(min_size=5, max_size=50) # response
),
min_size=0, max_size=3
), # triage_exchanges
st.text(min_size=20, max_size=200) # context
)
)
@settings(max_examples=50)
def test_provider_summary_formatting_completeness(self, summary_data):
"""
Test that provider summary formatting includes all required information.
Property: Formatted provider summaries should contain all required sections
and be suitable for provider review and action.
"""
from core.provider_summary_generator import ProviderSummaryGenerator, ProviderSummary
indicators, reasoning, confidence, patient_name, phone, triage_exchanges, context = summary_data
# Create summary
generator = ProviderSummaryGenerator()
# Convert triage exchanges to separate lists
triage_questions = [ex[0] for ex in triage_exchanges] if triage_exchanges else None
triage_responses = [ex[1] for ex in triage_exchanges] if triage_exchanges else None
summary = generator.generate_summary(
indicators=indicators,
reasoning=reasoning,
confidence=confidence,
patient_name=patient_name,
patient_phone=phone,
triage_questions=triage_questions,
triage_responses=triage_responses,
conversation_context=context
)
# Test display formatting
display_format = generator.format_for_display(summary)
# Property assertion: Display format should contain all required sections
required_sections = [
"PROVIDER SUMMARY",
"PATIENT INFORMATION",
"CLASSIFICATION & URGENCY",
"SITUATION OVERVIEW",
"DISTRESS INDICATORS",
"CLINICAL REASONING",
"RECOMMENDED ACTIONS"
]
for section in required_sections:
assert section in display_format, f"Display format should contain {section} section"
# Property assertion: Patient information should be visible
assert patient_name in display_format, "Display should show patient name"
assert phone in display_format, "Display should show patient phone"
# Property assertion: All indicators should be listed
for indicator in indicators:
assert indicator in display_format, f"Display should show indicator: {indicator}"
# Property assertion: Reasoning should be included (may be cleaned)
import re
clean_reasoning = re.sub(r'\s+', ' ', reasoning).strip()
assert clean_reasoning in display_format or reasoning in display_format, "Display should include reasoning"
# Property assertion: Triage context should be shown when available
if triage_exchanges:
assert "TRIAGE EXCHANGES" in display_format, "Should show triage section when available"
for question, response in triage_exchanges:
assert question in display_format, f"Should show triage question: {question}"
assert response in display_format, f"Should show triage response: {response}"
# Property assertion: Conversation context should be included
# (May be truncated if too long)
context_preview = context[:100] # First 100 chars should be visible
assert context_preview in display_format, "Should show conversation context"
# Test export formatting
export_format = generator.format_for_export(summary)
# Property assertion: Export format should be compact but complete
# Names and phones may be cleaned in export format
clean_name = patient_name.replace('\n', ' ').replace('\r', ' ').strip()
clean_phone = phone.replace('\n', ' ').replace('\r', ' ').strip()
assert clean_name in export_format or patient_name in export_format, "Export should include patient name"
assert clean_phone in export_format or phone in export_format, "Export should include phone"
assert "RED" in export_format, "Export should show classification"
# Reasoning may be cleaned in export format
clean_reasoning = re.sub(r'\s+', ' ', reasoning).strip()
assert clean_reasoning in export_format or reasoning in export_format, "Export should include reasoning"
# Property assertion: Export should be single line (no newlines)
assert '\n' not in export_format, "Export format should be single line"
# Property assertion: Export should use separators for parsing
assert '|' in export_format, "Export should use pipe separators"
@given(
validation_scenarios=st.lists(
st.tuples(
st.lists(st.text(min_size=3, max_size=20), min_size=0, max_size=5), # indicators (can be empty)
st.text(min_size=0, max_size=200), # reasoning (can be empty)
st.floats(min_value=0.0, max_value=1.0), # confidence
st.one_of(st.none(), st.text(min_size=1, max_size=30)), # patient_name (can be None)
st.one_of(st.none(), st.text(min_size=5, max_size=15)) # phone (can be None)
),
min_size=1,
max_size=5
)
)
@settings(max_examples=50)
def test_provider_summary_validation_and_completeness(self, validation_scenarios):
"""
Test that provider summary validation ensures completeness.
Property: Provider summaries should handle missing information gracefully
while ensuring all critical information is captured or flagged as missing.
"""
from core.provider_summary_generator import ProviderSummaryGenerator
generator = ProviderSummaryGenerator()
for indicators, reasoning, confidence, patient_name, phone in validation_scenarios:
# Generate summary with potentially missing information
summary = generator.generate_summary(
indicators=indicators,
reasoning=reasoning,
confidence=confidence,
patient_name=patient_name,
patient_phone=phone
)
# Property assertion: Summary should always be generated
assert summary is not None, "Should always generate a summary"
assert summary.classification == "RED", "Should maintain RED classification"
# Property assertion: Missing contact info should use placeholders
if patient_name is None:
assert summary.patient_name == "[Patient Name]", "Should use placeholder for missing name"
else:
assert summary.patient_name == patient_name, "Should use provided name"
if phone is None:
assert summary.patient_phone == "[Phone Number]", "Should use placeholder for missing phone"
else:
assert summary.patient_phone == phone, "Should use provided phone"
# Property assertion: Empty indicators should be handled gracefully
if not indicators:
assert summary.indicators == [], "Should handle empty indicators list"
# Situation description should still be meaningful
assert len(summary.situation_description) > 0, "Should generate description even without indicators"
else:
assert summary.indicators == indicators, "Should preserve provided indicators"
# Property assertion: Empty reasoning should be handled
if not reasoning:
# Should still have some default reasoning or description
assert len(summary.situation_description) > 0, "Should have situation description when reasoning is empty"
else:
assert summary.reasoning == reasoning, "Should preserve provided reasoning"
# Property assertion: Confidence should be preserved
assert summary.confidence == confidence, "Should preserve confidence level"
# Property assertion: Timestamp should always be present
assert summary.generated_at is not None, "Should always have generation timestamp"
assert len(summary.generated_at) > 0, "Timestamp should not be empty"
def test_provider_summary_integration_with_context_aware_classification(self):
"""
Test integration between provider summary generation and context-aware classification.
Property: Provider summaries should integrate with context-aware classification
results to provide comprehensive patient context.
"""
from core.provider_summary_generator import ProviderSummaryGenerator
from config.prompt_management.context_aware_classifier import ContextAwareClassifier
from config.prompt_management.data_models import ConversationHistory, Message
from datetime import datetime, timedelta
# Create context-aware classification scenario
classifier = ContextAwareClassifier()
generator = ProviderSummaryGenerator()
# Build conversation history with escalating distress
history = ConversationHistory(
messages=[
Message("I'm feeling anxious about my treatment", "YELLOW", datetime.now() - timedelta(hours=2)),
Message("I can't sleep and feel hopeless", "RED", datetime.now() - timedelta(hours=1)),
Message("I don't think I can go on like this", "RED", datetime.now() - timedelta(minutes=30))
],
distress_indicators_found=['anxiety', 'hopeless', 'insomnia'],
context_flags=['escalating_distress'],
medical_context={'conditions': ['cancer'], 'medications': ['chemotherapy']}
)
# Classify current message with context
current_message = "I just want the pain to stop"
classification_result = classifier.classify_with_context(current_message, history)
# Generate provider summary using classification results
summary = generator.generate_summary(
indicators=classification_result.indicators_found,
reasoning=classification_result.reasoning,
confidence=classification_result.confidence,
patient_name="Test Patient",
patient_phone="555-0123",
conversation_context=f"Recent messages show escalating distress. Current: {current_message}"
)
# Property assertion: Summary should reflect context-aware classification
assert summary.classification == "RED", "Should maintain RED classification"
assert classification_result.confidence == summary.confidence, "Should preserve classification confidence"
assert classification_result.reasoning == summary.reasoning, "Should use classification reasoning"
# Property assertion: Context factors should be reflected
if classification_result.context_factors:
# Context factors should influence the summary somehow
context_mentioned = any(
factor.lower() in summary.situation_description.lower()
for factor in classification_result.context_factors
)
# This is a soft assertion - context may be reflected in various ways
assert isinstance(context_mentioned, bool), "Should check for context factor reflection"
# Property assertion: Summary should be comprehensive
display_format = generator.format_for_display(summary)
# Should contain key information for provider action
assert "Test Patient" in display_format, "Should show patient name"
assert "555-0123" in display_format, "Should show contact info"
assert "RED FLAG" in display_format, "Should clearly indicate urgency"
assert "RECOMMENDED ACTION" in display_format, "Should provide action guidance"
# Property assertion: Export format should be suitable for handoff
export_format = generator.format_for_export(summary)
assert len(export_format) > 50, "Export should contain substantial information"
assert "Test Patient" in export_format, "Export should include patient identification"
assert "RED" in export_format, "Export should indicate classification"
class TestPerformanceMonitoring:
"""
**Feature: prompt-optimization, Property 8: Comprehensive Performance Monitoring**
Test that the performance monitoring system accurately captures all performance metrics
(response times, confidence levels, classification outcomes) and provides data-driven
optimization recommendations when patterns are identified.
**Validates: Requirements 8.1, 8.2, 8.3, 8.4, 8.5**
"""
@given(
st.lists(
st.tuples(
st.text(min_size=1, max_size=50), # agent_type
st.floats(min_value=0.1, max_value=10.0), # response_time
st.floats(min_value=0.0, max_value=1.0), # confidence
st.booleans(), # error
st.text(min_size=5, max_size=100) # classification_result
),
min_size=1,
max_size=20
)
)
@settings(max_examples=100)
def test_comprehensive_performance_monitoring(self, performance_data):
"""
Test that performance monitoring captures all required metrics.
Property: For any sequence of prompt executions, the monitoring system should
accurately capture response times, confidence levels, and outcomes, and provide
meaningful performance analysis.
**Validates: Requirements 8.1, 8.2, 8.3, 8.4, 8.5**
"""
from config.prompt_management.prompt_controller import PromptController
from config.prompt_management.performance_monitor import PromptMonitor
# Create fresh instances for each test
controller = PromptController()
monitor = PromptMonitor()
# Property: Performance metrics should be captured for all executions
for agent_type, response_time, confidence, error, classification_result in performance_data:
# Log performance metric (Requirement 8.1)
controller.log_performance_metric(
agent_type=agent_type,
response_time=response_time,
confidence=confidence,
error=error,
classification_result=classification_result
)
# Monitor should also track the execution (Requirement 8.2)
monitor.track_execution(
agent_type=agent_type,
response_time=response_time,
confidence=confidence,
success=not error,
metadata={'classification': classification_result}
)
# Property: All logged metrics should be retrievable
unique_agents = list(set(item[0] for item in performance_data))
for agent_type in unique_agents:
# Get metrics from controller
controller_metrics = controller.get_performance_metrics(agent_type)
# Property assertion: Metrics should contain all required fields (Requirement 8.1)
assert 'total_executions' in controller_metrics, "Should track total executions"
assert 'average_response_time' in controller_metrics, "Should track average response time"
assert 'average_confidence' in controller_metrics, "Should track average confidence"
assert 'error_rate' in controller_metrics, "Should track error rate"
# Property assertion: Metrics should be accurate
agent_data = [item for item in performance_data if item[0] == agent_type]
expected_executions = len(agent_data)
assert controller_metrics['total_executions'] == expected_executions, \
"Should count all executions correctly"
if expected_executions > 0:
expected_avg_time = sum(item[1] for item in agent_data) / expected_executions
expected_avg_confidence = sum(item[2] for item in agent_data) / expected_executions
expected_error_rate = sum(1 for item in agent_data if item[3]) / expected_executions
# Allow small floating point differences
assert abs(controller_metrics['average_response_time'] - expected_avg_time) < 0.001, \
"Should calculate average response time correctly"
assert abs(controller_metrics['average_confidence'] - expected_avg_confidence) < 0.001, \
"Should calculate average confidence correctly"
assert abs(controller_metrics['error_rate'] - expected_error_rate) < 0.001, \
"Should calculate error rate correctly"
# Get detailed metrics from monitor (Requirement 8.2)
monitor_metrics = monitor.get_detailed_metrics(agent_type)
# Property assertion: Monitor should provide detailed analysis
assert 'performance_trend' in monitor_metrics, "Should analyze performance trends"
assert 'confidence_distribution' in monitor_metrics, "Should analyze confidence distribution"
assert 'error_patterns' in monitor_metrics, "Should identify error patterns"
@given(
st.lists(
st.tuples(
st.text(min_size=1, max_size=20), # agent_type
st.floats(min_value=0.1, max_value=5.0), # response_time
st.floats(min_value=0.0, max_value=1.0), # confidence
st.text(min_size=1, max_size=50) # prompt_version
),
min_size=2,
max_size=10
)
)
@settings(max_examples=50)
def test_ab_testing_framework(self, ab_test_data):
"""
Test A/B testing framework for prompt performance comparison.
Property: For any two prompt versions, the A/B testing framework should
enable statistical comparison and automated rollback for underperforming prompts.
**Validates: Requirements 8.3**
"""
from config.prompt_management.performance_monitor import PromptMonitor
monitor = PromptMonitor()
# Property: A/B testing should handle multiple prompt versions
for agent_type, response_time, confidence, prompt_version in ab_test_data:
monitor.log_ab_test_result(
agent_type=agent_type,
prompt_version=prompt_version,
response_time=response_time,
confidence=confidence
)
# Property: Should be able to compare versions
unique_agents = list(set(item[0] for item in ab_test_data))
for agent_type in unique_agents:
agent_data = [item for item in ab_test_data if item[0] == agent_type]
unique_versions = list(set(item[3] for item in agent_data))
if len(unique_versions) >= 2:
# Test version comparison
comparison_result = monitor.compare_prompt_versions(
agent_type=agent_type,
version_a=unique_versions[0],
version_b=unique_versions[1]
)
# Property assertion: Comparison should provide statistical analysis
assert 'statistical_significance' in comparison_result, \
"Should test statistical significance"
assert 'performance_difference' in comparison_result, \
"Should quantify performance difference"
assert 'recommendation' in comparison_result, \
"Should provide rollback recommendation"
# Property assertion: Recommendation should be actionable
recommendation = comparison_result['recommendation']
assert recommendation in ['keep_version_a', 'switch_to_version_b', 'insufficient_data'], \
"Should provide clear recommendation"
@given(
st.lists(
st.tuples(
st.text(min_size=1, max_size=20), # agent_type
st.floats(min_value=0.0, max_value=1.0), # confidence
st.booleans(), # classification_error
st.text(min_size=5, max_size=100) # error_pattern
),
min_size=5,
max_size=25
)
)
@settings(max_examples=50)
def test_optimization_recommendation_engine(self, optimization_data):
"""
Test optimization recommendation engine for data-driven improvements.
Property: For any pattern of errors and performance issues, the optimization
engine should identify patterns and provide specific improvement recommendations.
**Validates: Requirements 8.4, 8.5**
"""
from config.prompt_management.performance_monitor import PromptMonitor
monitor = PromptMonitor()
# Property: Should analyze error patterns and generate recommendations
for agent_type, confidence, classification_error, error_pattern in optimization_data:
monitor.log_classification_outcome(
agent_type=agent_type,
confidence=confidence,
classification_error=classification_error,
error_details={'pattern': error_pattern}
)
# Property: Should generate optimization recommendations
unique_agents = list(set(item[0] for item in optimization_data))
for agent_type in unique_agents:
agent_data = [item for item in optimization_data if item[0] == agent_type]
# Get optimization recommendations
recommendations = monitor.get_optimization_recommendations(agent_type)
# Property assertion: Should provide actionable recommendations
assert isinstance(recommendations, list), "Should return list of recommendations"
if len(agent_data) >= 3: # Need sufficient data for analysis
# Should identify patterns if errors exist
has_errors = any(item[2] for item in agent_data)
if has_errors:
# Should provide specific recommendations for improvement
assert len(recommendations) > 0, "Should provide recommendations when errors detected"
for recommendation in recommendations:
assert hasattr(recommendation, 'type'), "Should specify recommendation type"
assert hasattr(recommendation, 'description'), "Should provide description"
assert hasattr(recommendation, 'priority'), "Should indicate priority"
assert hasattr(recommendation, 'expected_impact'), "Should estimate impact"
# Property assertion: Recommendation types should be valid
from config.prompt_management.performance_monitor import RecommendationType
valid_types = [rt.value for rt in RecommendationType]
assert recommendation.type.value in valid_types, \
f"Should use valid recommendation type: {recommendation.type.value}"
# Property: Should track improvement over time
improvement_metrics = monitor.get_improvement_tracking(agent_type)
assert 'baseline_performance' in improvement_metrics, \
"Should establish baseline performance"
assert 'current_performance' in improvement_metrics, \
"Should track current performance"
assert 'improvement_trend' in improvement_metrics, \
"Should analyze improvement trend"
def test_performance_monitoring_integration(self):
"""
Test integration between performance monitoring and existing prompt system.
Property: Performance monitoring should integrate seamlessly with existing
prompt management without affecting core functionality.
**Validates: Requirements 8.1, 8.2, 8.3, 8.4, 8.5**
"""
from config.prompt_management.prompt_controller import PromptController
from config.prompt_management.performance_monitor import PromptMonitor
controller = PromptController()
monitor = PromptMonitor()
# Property: Should work with existing prompt retrieval
config = controller.get_prompt('spiritual_monitor')
assert config is not None, "Should retrieve prompt configuration"
# Property: Should integrate with session overrides
session_id = "test_session_123"
test_prompt = "Test prompt for performance monitoring"
success = controller.set_session_override('spiritual_monitor', test_prompt, session_id)
assert success, "Should set session override successfully"
# Property: Performance monitoring should work with session overrides
session_config = controller.get_prompt('spiritual_monitor', session_id=session_id)
assert session_config.session_override == test_prompt, "Should use session override"
# Property: Should log performance for session-based prompts
controller.log_performance_metric(
agent_type='spiritual_monitor',
response_time=0.5,
confidence=0.8,
session_id=session_id
)
metrics = controller.get_performance_metrics('spiritual_monitor')
assert metrics['total_executions'] >= 1, "Should log session-based performance"
# Property: Should maintain performance history across sessions
controller.clear_session_overrides(session_id)
# Metrics should persist after session cleanup
metrics_after_cleanup = controller.get_performance_metrics('spiritual_monitor')
assert metrics_after_cleanup['total_executions'] == metrics['total_executions'], \
"Should maintain performance history after session cleanup"