tests/test_council.py

"""
PhD Research OS — AI Model Council Tests
==========================================
Tests for the 4-member council pipeline: Query Planner → Extractor → Critic → Chairman
"""

import os
import sys
import json
import pytest

sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))

from phd_research_os.council import (
    ModelCouncil, CouncilRound, COUNCIL_PROMPTS
)
from phd_research_os.db import init_db, get_db, create_claim
from phd_research_os.taxonomy import TaxonomyManager

TEST_DB = "test_council.db"

SAMPLE_TEXT = """
We investigated the sensitivity of graphene field-effect transistor (GFET) biosensors 
to cardiac troponin I (cTnI) in phosphate-buffered saline (PBS). Our measurements 
show that the Dirac point shifts by 45 ± 3 mV upon binding of 1 pM cTnI in 10 mM PBS 
(n=5, p<0.001). The limit of detection was determined to be 0.8 fM using the 3σ/slope 
method. We interpret these results as evidence that aptamer-functionalized GFETs can 
achieve clinically relevant sensitivity for cardiac biomarker detection. However, ionic 
strength effects at physiological conditions (150 mM) significantly reduce sensitivity, 
suggesting that a desalting step may be necessary for clinical translation. We hypothesize 
that a PEG spacer of 5 kDa molecular weight could mitigate Debye screening while 
maintaining binding affinity, though this remains to be experimentally validated.
"""


@pytest.fixture(autouse=True)
def setup_teardown():
    init_db(TEST_DB)
    TaxonomyManager(db_path=TEST_DB)
    yield
    for suffix in ["", "-wal", "-shm"]:
        p = TEST_DB + suffix
        if os.path.exists(p):
            os.remove(p)


# ============================================================
# Council Prompt Tests
# ============================================================

def test_all_4_council_members_defined():
    assert "query_planner" in COUNCIL_PROMPTS
    assert "extractor" in COUNCIL_PROMPTS
    assert "critic" in COUNCIL_PROMPTS
    assert "chairman" in COUNCIL_PROMPTS

def test_query_planner_prompt_has_json_instruction():
    assert "JSON array" in COUNCIL_PROMPTS["query_planner"]

def test_extractor_prompt_has_epistemic_tags():
    prompt = COUNCIL_PROMPTS["extractor"]
    assert "Fact" in prompt
    assert "Interpretation" in prompt
    assert "Hypothesis" in prompt
    assert "Conflict_Hypothesis" in prompt

def test_critic_prompt_checks_5_things():
    prompt = COUNCIL_PROMPTS["critic"]
    assert "Missing important claims" in prompt
    assert "Incorrect epistemic tags" in prompt
    assert "Overly confident" in prompt
    assert "Taxonomy correctness" in prompt
    assert "Missing fields" in prompt

def test_chairman_prompt_has_completeness_penalty():
    assert "0.7 completeness penalty" in COUNCIL_PROMPTS["chairman"]

def test_chairman_output_schema():
    prompt = COUNCIL_PROMPTS["chairman"]
    assert "epistemic_tag" in prompt
    assert "confidence" in prompt
    assert "missing_fields" in prompt
    assert "status" in prompt


# ============================================================
# Mock Council Pipeline Tests (no brain)
# ============================================================

def test_council_deliberate_without_brain():
    council = ModelCouncil(brain=None, db_path=TEST_DB)
    result = council.deliberate(SAMPLE_TEXT)

    assert isinstance(result, CouncilRound)
    assert result.round_id.startswith("CNCL_")
    assert len(result.final_claims) >= 1
    assert result.started_at
    assert result.completed_at

def test_council_produces_valid_claims():
    council = ModelCouncil(brain=None, db_path=TEST_DB)
    result = council.deliberate(SAMPLE_TEXT)

    for claim in result.final_claims:
        assert "text" in claim
        assert "epistemic_tag" in claim
        assert "confidence" in claim
        assert "missing_fields" in claim
        assert "status" in claim
        assert claim["epistemic_tag"] in ["Fact", "Interpretation", "Hypothesis", "Conflict_Hypothesis"]
        assert 0.0 <= claim["confidence"] <= 1.0
        assert claim["status"] in ["Complete", "Incomplete"]

def test_council_claims_have_consistent_status():
    council = ModelCouncil(brain=None, db_path=TEST_DB)
    result = council.deliberate(SAMPLE_TEXT)

    for claim in result.final_claims:
        if claim["missing_fields"]:
            assert claim["status"] == "Incomplete"

def test_council_with_query():
    council = ModelCouncil(brain=None, db_path=TEST_DB)
    result = council.deliberate(SAMPLE_TEXT, query="What is the LOD for GFET cTnI sensors?")

    assert isinstance(result.query_plan, list)
    assert len(result.final_claims) >= 1

def test_council_query_planner_standalone():
    council = ModelCouncil(brain=None, db_path=TEST_DB)
    queries = council.deliberate_query("What are the ionic strength effects on GFET sensitivity?")
    assert isinstance(queries, list)
    assert len(queries) >= 1

def test_council_metadata():
    council = ModelCouncil(brain=None, db_path=TEST_DB)
    result = council.deliberate(SAMPLE_TEXT)

    assert "council_version" in result.metadata
    assert "taxonomy_domain" in result.metadata
    assert "extractor_claim_count" in result.metadata
    assert "final_claim_count" in result.metadata

def test_council_critique_structure():
    council = ModelCouncil(brain=None, db_path=TEST_DB)
    result = council.deliberate(SAMPLE_TEXT)

    if result.critique:
        assert "feedback" in result.critique

def test_council_raw_extraction_preserved():
    council = ModelCouncil(brain=None, db_path=TEST_DB)
    result = council.deliberate(SAMPLE_TEXT)

    assert isinstance(result.raw_extraction, list)

def test_council_round_logged_to_db():
    council = ModelCouncil(brain=None, db_path=TEST_DB)
    result = council.deliberate(SAMPLE_TEXT)

    history = council.get_council_history()
    assert len(history) >= 1
    assert history[0]["round_id"] == result.round_id

def test_council_history_retrieval():
    council = ModelCouncil(brain=None, db_path=TEST_DB)
    council.deliberate(SAMPLE_TEXT)
    council.deliberate(SAMPLE_TEXT)

    history = council.get_council_history(limit=10)
    assert len(history) >= 2

def test_council_confidence_clamped():
    council = ModelCouncil(brain=None, db_path=TEST_DB)
    result = council.deliberate(SAMPLE_TEXT)

    for claim in result.final_claims:
        assert 0.0 <= claim["confidence"] <= 1.0

def test_council_invalid_tag_defaulted():
    """Claims with invalid epistemic tags should be defaulted to Interpretation."""
    council = ModelCouncil(brain=None, db_path=TEST_DB)
    # Mock produces "Interpretation" by default, so this validates the path
    result = council.deliberate(SAMPLE_TEXT)
    for claim in result.final_claims:
        assert claim["epistemic_tag"] in ["Fact", "Interpretation", "Hypothesis", "Conflict_Hypothesis"]

def test_multiple_deliberations_independent():
    council = ModelCouncil(brain=None, db_path=TEST_DB)
    r1 = council.deliberate("Text about graphene sensors.")
    r2 = council.deliberate("Text about lithium batteries.")

    assert r1.round_id != r2.round_id


if __name__ == "__main__":
    pytest.main([__file__, "-v"])