FabledScribe/tests/test_services_retrieval_telemetry.py

"""Tests for services.retrieval_telemetry.

_build_payload is pure (no DB, no loop) and gets unit coverage. The persistence
path (_insert_retrieval_log + the RetrievalLog model / JSONB roundtrip) is an
integration test against real Postgres.
"""
from types import SimpleNamespace

import pytest

from scribe.services.retrieval_telemetry import (
    _build_payload,
    record_retrieval,
)


def _note(nid):
    """Minimal stand-in — _build_payload only reads .id."""
    return SimpleNamespace(id=nid)


# ─── _build_payload (pure) ───────────────────────────────────────────────────


def test_build_payload_ranks_and_score_bounds():
    results = [(0.91, _note(11)), (0.72, _note(22)), (0.55, _note(33))]
    p = _build_payload(
        user_id=7, source="mcp_search", query="hello", threshold=0.45,
        limit=10, project_id=3, is_task=None, results=results, duration_ms=12.345,
    )
    assert p["result_count"] == 3
    assert p["top_score"] == 0.91
    assert p["min_score"] == 0.55
    assert [it["rank"] for it in p["result_ids"]] == [0, 1, 2]
    assert [it["id"] for it in p["result_ids"]] == [11, 22, 33]
    assert p["duration_ms"] == 12.35  # rounded to 2dp
    assert p["user_id"] == 7 and p["project_id"] == 3 and p["threshold"] == 0.45


def test_build_payload_empty_results():
    p = _build_payload(
        user_id=1, source="rest_search", query="x", threshold=0.3,
        limit=5, project_id=None, is_task=False, results=[], duration_ms=None,
    )
    assert p["result_count"] == 0
    assert p["top_score"] is None and p["min_score"] is None
    assert p["result_ids"] == []
    assert p["duration_ms"] is None


def test_build_payload_rounds_scores_to_5dp():
    p = _build_payload(
        user_id=1, source="mcp_search", query="q", threshold=0.45,
        limit=1, project_id=None, is_task=None,
        results=[(0.123456789, _note(1))], duration_ms=0.0,
    )
    assert p["result_ids"][0]["score"] == 0.12346


def test_record_retrieval_without_event_loop_is_safe():
    """Called from a sync context (no running loop) it must swallow and return,
    never raise — telemetry can't be allowed to break a caller."""
    # No event loop running in this plain sync test.
    assert record_retrieval(
        user_id=1, source="mcp_search", query="q", threshold=0.45,
        limit=10, project_id=None, is_task=None,
        results=[(0.9, _note(1))],
    ) is None


# ─── persistence (integration) ───────────────────────────────────────────────


@pytest.mark.integration
@pytest.mark.asyncio
async def test_insert_retrieval_log_roundtrip(_dispose_engine):
    from sqlalchemy import delete, select

    from scribe.models import async_session
    from scribe.models.retrieval_log import RetrievalLog
    from scribe.services.retrieval_telemetry import _insert_retrieval_log

    payload = _build_payload(
        user_id=990001, source="mcp_search", query="pgvector tuning",
        threshold=0.45, limit=10, project_id=None, is_task=None,
        results=[(0.88, _note(501)), (0.61, _note(502))], duration_ms=9.9,
    )
    await _insert_retrieval_log(payload)

    async with async_session() as s:
        row = (
            await s.execute(
                select(RetrievalLog).where(RetrievalLog.user_id == 990001)
            )
        ).scalars().first()
        assert row is not None
        assert row.source == "mcp_search"
        assert row.result_count == 2
        assert row.top_score == 0.88
        # JSONB roundtrips as a list of dicts with the expected shape.
        assert row.result_ids[0] == {"id": 501, "score": 0.88, "rank": 0}
        assert row.created_at is not None  # server_default now()
        await s.execute(delete(RetrievalLog).where(RetrievalLog.user_id == 990001))
        await s.commit()


# ─── the read half: retrieval_summary (integration) ──────────────────────────
# Integration, not mocked, and deliberately so. #2663 is the bug where a
# GROUP BY the database rejected was swallowed by a broad except, so every
# counter read zero in production while the writes landed fine and the mocked
# tests passed. `retrieval_summary` runs a grouped aggregate with
# percentile_cont ... WITHIN GROUP and a two-label CASE — precisely the shape
# that failed then. Only a real Postgres can say it parses.


@pytest.mark.integration
@pytest.mark.asyncio
async def test_retrieval_summary_reads_what_the_writer_wrote(_dispose_engine):
    from sqlalchemy import delete

    from scribe.models import async_session
    from scribe.models.note_usage import NoteUsageEvent
    from scribe.models.retrieval_log import RetrievalLog
    from scribe.services.retrieval_telemetry import (
        _insert_retrieval_log, retrieval_summary,
    )

    UID = 990002
    # Three auto_inject calls at a 0.55 bar: two clear it, one does not.
    # Plus one call that returned nothing at all — a different failure from a
    # low-scoring one, and the readout must not blend them.
    for score in (0.91, 0.72, 0.40):
        await _insert_retrieval_log(_build_payload(
            user_id=UID, source="auto_inject", query="q", threshold=0.55,
            limit=3, project_id=None, is_task=None,
            results=[(score, _note(1))], duration_ms=5.0,
        ))
    await _insert_retrieval_log(_build_payload(
        user_id=UID, source="auto_inject", query="q", threshold=0.55,
        limit=3, project_id=None, is_task=None, results=[], duration_ms=5.0,
    ))
    # A second surface, so the GROUP BY has something to separate.
    await _insert_retrieval_log(_build_payload(
        user_id=UID, source="mcp_search", query="q", threshold=0.45,
        limit=10, project_id=None, is_task=None,
        results=[(0.80, _note(2))], duration_ms=11.0,
    ))
    # Corpus side: two ranked surfacings, one ambient, one pull.
    async with async_session() as s:
        s.add_all([
            NoteUsageEvent(user_id=UID, note_id=1, event="surfaced", source="auto_inject"),
            NoteUsageEvent(user_id=UID, note_id=2, event="surfaced", source="auto_inject"),
            NoteUsageEvent(user_id=UID, note_id=3, event="surfaced", source="enter_project"),
            NoteUsageEvent(user_id=UID, note_id=1, event="pulled", source="mcp_get_note"),
        ])
        await s.commit()

    try:
        out = await retrieval_summary(UID, days=30)

        assert out["read_failed"] is False, "the aggregate did not execute"
        ai = out["sources"]["auto_inject"]
        assert ai["calls"] == 4
        assert ai["zero_result_calls"] == 1
        assert ai["cleared_threshold"] == 2      # 0.91 and 0.72, not 0.40
        # p50 over the three scored calls; the empty one contributes no score.
        assert ai["top_score"]["p50"] == pytest.approx(0.72, abs=1e-4)
        assert ai["top_score"]["min"] == pytest.approx(0.40, abs=1e-4)
        assert ai["top_score"]["max"] == pytest.approx(0.91, abs=1e-4)
        assert out["sources"]["mcp_search"]["calls"] == 1

        u = out["usage"]
        assert u["surfaced"] == 2 and u["ambient"] == 1 and u["pulled"] == 1
        assert u["distinct_notes_surfaced"] == 2
        # The pull came from `mcp_get_note`, so it counts as an AGENT pull
        # and drives pull_through; a human `rest_*` pull would not.
        assert u["pulled_by_agent"] == 1 and u["pulled_by_human"] == 0
        assert u["pull_through"] == pytest.approx(0.5)
    finally:
        async with async_session() as s:
            await s.execute(delete(RetrievalLog).where(RetrievalLog.user_id == UID))
            await s.execute(delete(NoteUsageEvent).where(NoteUsageEvent.user_id == UID))
            await s.commit()


@pytest.mark.integration
@pytest.mark.asyncio
async def test_retrieval_summary_is_empty_not_broken_for_a_fresh_install(_dispose_engine):
    """Rule #115: an install with no telemetry gets a coherent zero readout,
    and `read_failed` stays False — the distinction #2663 says must exist."""
    from scribe.services.retrieval_telemetry import retrieval_summary

    out = await retrieval_summary(990003, days=30)
    assert out["read_failed"] is False
    assert out["sources"] == {}
    assert out["usage"]["pull_through"] is None   # no division by zero
    assert out["usage"]["surfaced"] == 0


@pytest.mark.integration
@pytest.mark.asyncio
async def test_retrieval_summary_sees_only_its_own_users_telemetry(_dispose_engine):
    """A retrieval log records what one user's agent asked for, query text
    included. The owner filter is the access rule, so it gets a test."""
    from sqlalchemy import delete

    from scribe.models import async_session
    from scribe.models.retrieval_log import RetrievalLog
    from scribe.services.retrieval_telemetry import (
        _insert_retrieval_log, retrieval_summary,
    )

    await _insert_retrieval_log(_build_payload(
        user_id=990004, source="auto_inject", query="theirs", threshold=0.55,
        limit=3, project_id=None, is_task=None, results=[(0.9, _note(1))],
        duration_ms=1.0,
    ))
    try:
        assert (await retrieval_summary(990005, days=30))["sources"] == {}
        assert (await retrieval_summary(990004, days=30))["sources"]["auto_inject"]["calls"] == 1
    finally:
        async with async_session() as s:
            await s.execute(delete(RetrievalLog).where(RetrievalLog.user_id == 990004))
            await s.commit()