tests/test_conservation.py

"""Audit #17 + #18: multiple alignment, and conservation counted from it.

ESM-2 gives a learned opinion about whether a substitution looks plausible.
Conservation across real homologs gives an OBSERVED fact: in 40 orthologs this
position is serine 40 times. They fail differently — the model is weakest
exactly where the audit says (membrane, disordered, multi-domain) and a
frequency count is unaffected by any of that.

The tests that matter are about the INPUT, not the algorithm. Conservation is
only as meaningful as the homolog set, and a set of near-identical sequences
makes every column look invariant. A tool that reports that as "conserved"
without comment is worse than no tool.
"""
import pytest

from dee.core import agent_tools as t
from dee.core import conservation as C
from dee.core import orchestrator as orch

# Position 1,3,4,5,6 invariant; position 2 varies across all five.
HOMOLOGS = ["MAKQWLTVEG", "MSKQWLTVEG", "MTKQWLSVEG", "MAKQWLTVDG", "MGKQWLTIEG"]


def test_an_invariant_column_is_found_and_a_variable_one_is_not():
    r = C.score(HOMOLOGS)
    by_pos = {c["anchor_position"]: c for c in r["conservation"]}
    assert by_pos[5]["call"] == "invariant" and by_pos[5]["agreement_pct"] == 100.0
    assert by_pos[2]["call"] == "variable" and by_pos[2]["agreement_pct"] < 60


def test_entropy_is_never_reported_as_negative_zero():
    """-sum(...) over an all-agreeing column yields IEEE -0.0, which prints as
    '-0.0 bits' and reads like a bug in a number users are asked to trust."""
    r = C.score(HOMOLOGS)
    for c in r["conservation"]:
        assert c["entropy_bits"] >= 0.0
        assert str(c["entropy_bits"]) != "-0.0"


def test_a_redundant_homolog_set_is_flagged_and_marked_untrustworthy():
    """The failure that would make this tool actively misleading. Forty
    sequences at 99% identity are ONE sequence counted forty times, and every
    column will look invariant."""
    r = C.score(["MAKQWLTVEG"] * 4 + ["MAKQWLTVEA"])
    assert r["trustworthy"] is False
    assert r["diversity_warning"] and "redundancy" in r["diversity_warning"]
    assert "more divergent" in r["diversity_warning"]


def test_a_diverse_set_is_trusted():
    """A checker that never trusts anything is a checker nobody reads."""
    r = C.score(HOMOLOGS)
    assert r["trustworthy"] is True and r["diversity_warning"] is None


def test_too_few_sequences_is_refused_with_a_way_forward():
    r = C.score(["MAKQ", "MSKQ"])
    assert r["ok"] is False and r["kind"] == "too_few"
    assert "BLAST" in r["next"]


def test_positions_are_numbered_along_the_anchor_the_user_designs_against():
    r = C.score(HOMOLOGS, positions=[5])
    assert [(c["anchor_position"], c["anchor_residue"]) for c in r["conservation"]] \
        == [(5, "W")]


def test_indels_do_not_shift_the_anchor_numbering():
    """A homolog with a deletion must not renumber the sequence the user is
    designing against — that is how a conservation call lands on the wrong
    residue."""
    r = C.score(["MAKQWLTVEG", "MAKWLTVEG", "MAKQWLTVEG"])
    positions = [c["anchor_position"] for c in r["conservation"]
                 if c["anchor_position"]]
    assert positions == sorted(positions)
    assert max(positions) == 10          # anchor length, unchanged by the gap


def test_the_alignment_says_it_is_approximate():
    """Progressive alignment is not a simultaneous optimum. Every practical
    tool approximates; pretending otherwise is the dishonest part."""
    m = C.align_many(HOMOLOGS)
    assert m["ok"] and "Approximate" in m["method"]
    assert all(len(r["aligned"]) == m["columns"] for r in m["rows"])


def test_the_anchor_is_the_longest_sequence():
    m = C.align_many(["MAK", "MAKQWLTVEG", "MAKQ"], ["a", "b", "c"])
    assert m["anchor"] == "b"


def test_it_positions_itself_against_esm2_rather_than_as_a_confirmation():
    """Two independent signals. Averaging them would destroy the only thing
    that makes having both worthwhile."""
    r = C.score(HOMOLOGS)
    assert "complements ESM-2" in r["caveat"]
    assert "disagreement is worth investigating" in r["caveat"]


def test_conservation_is_framed_as_an_observation_not_a_property():
    r = C.score(HOMOLOGS)
    assert "OBSERVATION" in r["caveat"]
    assert "not a property of the protein" in r["caveat"]


def test_it_is_reachable_ungated_and_specced():
    assert "check_conservation" in t._TOOLS
    assert any(s["function"]["name"] == "check_conservation" for s in orch.TOOL_SPECS)
    assert orch._requires_confirm("check_conservation") is False


def test_the_summary_leads_with_untrustworthiness_when_it_applies():
    r = t.execute_tool("check_conservation",
                       {"sequences": ["MAKQWLTVEG"] * 4 + ["MAKQWLTVEA"]},
                       auth_anonymous=True)
    assert "too redundant to trust" in orch._summarize("check_conservation", r)


def test_the_spec_makes_the_agent_relay_both_limits():
    d = next(s["function"]["description"] for s in orch.TOOL_SPECS
             if s["function"]["name"] == "check_conservation")
    assert "approximate" in d
    assert "diversity_warning" in d and "trustworthy" in d
    assert "complements ESM-2" in d