FabledScribe/tests/test_lesson_convergence.py

"""Convergence named at the write (milestone 440, #4634).

When a lesson is answered "no rule fits", the response looks for other
no-rule lessons in the same situation and, once there are enough, names the
group and the rule it may be missing. The bar is pinned here as a pure
function; the search around it with the database and embedder stubbed; the
doors by what they return.
"""
from __future__ import annotations

from types import SimpleNamespace
from unittest.mock import AsyncMock, patch

import pytest

from scribe.mcp._context import _user_id_ctx
from scribe.mcp.tools import lessons as lesson_tools
from scribe.services import lesson_rules as links_svc
from scribe.services import lessons as lessons_svc

# Bound before conftest's autouse stub replaces the module attribute.
_REAL = links_svc.convergence_for


def _m(lid, sources=(), project=None, title=None):
    return {"id": lid, "title": title or f"lesson {lid}", "sources": list(sources),
            "project_id": project}


def test_below_the_group_size_nothing_is_named():
    assert links_svc.convergence_group([_m(1), _m(2)]) is None


def test_a_group_of_distinct_lessons_names_its_members_and_the_next_step():
    group = links_svc.convergence_group([_m(1, [10], 2), _m(2, [11], 5), _m(3, [], 2)])
    assert [m["id"] for m in group["lessons"]] == [1, 2, 3]
    assert group["incidents"] == [10, 11]
    assert group["projects"] == [2, 5]
    assert "create_rule" in group["hint"] and "update_lesson" in group["hint"]
    assert "#1 “lesson 1”" in group["hint"]


def test_one_broad_lesson_cannot_reach_the_bar_alone():
    """Many incidents behind ONE lesson is still one judgment: incidents never
    stand in for lessons, and a repeated id is one member."""
    broad = _m(1, [10, 11, 12, 13, 14])
    assert links_svc.convergence_group([broad]) is None
    assert links_svc.convergence_group([broad, broad, broad]) is None


def test_one_incident_written_up_three_times_is_not_a_recurring_situation():
    same = [_m(1, [10]), _m(2, [10]), _m(3, [10])]
    assert links_svc.convergence_group(same) is None


# ── The bar is relative to each lesson's own register (#5193) ──────────────


def test_too_little_background_says_nothing_stands_out():
    few = {i: 0.6 for i in range(links_svc.CONVERGENCE_MIN_BACKGROUND - 1)}
    assert links_svc.standout(few) == {}


def test_a_background_that_never_varies_says_nothing_stands_out():
    assert links_svc.standout({i: 0.7 for i in range(20)}) == {}


def test_standing_out_is_measured_in_the_registers_own_spread():
    """The same neighbour score stands out in a tight register and not in a
    loose one — the bar moves with the corpus, not with a constant."""
    tight = {i: 0.70 + 0.01 * (i % 5) for i in range(20)} | {99: 0.80}
    loose = {i: 0.60 + 0.06 * (i % 5) for i in range(20)} | {99: 0.80}
    assert links_svc.standout(tight)[99] >= links_svc.CONVERGENCE_STANDOUT
    assert links_svc.standout(loose)[99] < links_svc.CONVERGENCE_STANDOUT


def test_a_hub_near_two_lessons_that_are_not_near_each_other_is_no_clique():
    up = links_svc.CONVERGENCE_STANDOUT + 1
    standouts = {1: {2: up, 3: up}, 2: {1: up, 3: 0.0}, 3: {1: up, 2: 0.0}}
    assert links_svc.converging(1, standouts, [2, 3]) == [1, 2]


def test_resemblance_must_run_both_ways():
    up = links_svc.CONVERGENCE_STANDOUT + 1
    standouts = {1: {2: up, 3: up}, 2: {1: 0.0, 3: up}, 3: {1: up, 2: up}}
    assert links_svc.converging(1, standouts, [2, 3]) == [1, 3]


def _note(lid, *, title=None, project=None, sources=()):
    data = {"what": title or f"lesson {lid}", "when_to_apply": "a CI run overran"}
    if sources:
        data["taught_by"] = list(sources)
    return SimpleNamespace(
        id=lid, title=title or f"lesson {lid}", note_type="lesson", body="", data=data,
        project_id=project, arose_from_id=None, tags=[],
        created_at=None, updated_at=None,
    )


# A register: lessons 1-3 resemble each other well above a background of
# filler lessons 10-29, which all sit in one flat band — the shape every
# lesson-to-lesson score has, because lessons share one register.
_TRIO = {1, 2, 3}
_FILLER = range(10, 30)
_NOTES = {i: _note(i, sources=[100 + i]) for i in (*_TRIO, *_FILLER)}


def _sim(a, b):
    if a in _TRIO and b in _TRIO:
        return 0.82
    return 0.66 + 0.01 * ((a + b) % 5)


def _search_over(sim=_sim):
    async def search(user_id, query, *, exclude_ids, **_kw):
        (me,) = exclude_ids
        return sorted(((sim(me, i), n) for i, n in _NOTES.items() if i != me),
                      key=lambda pair: -pair[0])
    return search


def _run(no_rule, sim=_sim):
    return (
        patch.object(lessons_svc, "get_lesson", AsyncMock(return_value=_NOTES[1])),
        patch("scribe.services.embeddings.semantic_search_notes", _search_over(sim)),
        patch.object(links_svc, "_no_rule_ids",
                     AsyncMock(side_effect=lambda ids: {i for i in ids if i in no_rule})),
    )


@pytest.mark.asyncio
async def test_lessons_that_stand_out_together_are_named():
    a, b, c = _run(no_rule=set(_NOTES))
    with a, b, c:
        group = await _REAL(7, 1)
    assert [m["id"] for m in group["lessons"]] == [1, 2, 3]


@pytest.mark.asyncio
async def test_a_large_no_rule_pool_in_one_flat_band_names_nothing():
    """The #5193 failure: every lesson answered "no rule", every score above
    a fixed bar, and a group the size of the pool. Nothing stands out of a
    flat band, so nothing is named however large the pool grows."""
    a, b, c = _run(no_rule=set(_NOTES), sim=lambda x, y: 0.66 + 0.01 * ((x + y) % 5))
    with a, b, c:
        assert await _REAL(7, 1) is None


@pytest.mark.asyncio
async def test_only_lessons_answered_no_rule_join_the_group():
    a, b, c = _run(no_rule={1, 2})          # #3 has a rule: a pair is no group
    with a, b, c:
        assert await _REAL(7, 1) is None


@pytest.mark.asyncio
async def test_a_failed_search_names_nothing_and_raises_nothing():
    with patch.object(lessons_svc, "get_lesson", AsyncMock(side_effect=RuntimeError("db down"))):
        assert await _REAL(7, 1) is None


@pytest.mark.asyncio
async def test_the_door_carries_convergence_only_with_a_no_rule_answer():
    _user_id_ctx.set(7)
    group = {"lessons": [], "incidents": [], "projects": [], "hint": "h"}
    named = AsyncMock(return_value=group)
    with patch.object(lessons_svc, "update_lesson", AsyncMock(return_value=_note(41))), \
         patch.object(links_svc, "set_no_rule", AsyncMock()), \
         patch.object(links_svc, "convergence_for", named):
        out = await lesson_tools.update_lesson(lesson_id=41, no_rule="stands alone")
        assert out["convergence"] == group
        out = await lesson_tools.update_lesson(lesson_id=41, what="reworded")
    assert "convergence" not in out
    assert named.await_count == 1