Files
FabledScribe/tests/test_note_usage.py
T
bvandeusen 2b85443dd1
CI & Build / Python lint (push) Successful in 2s
CI & Build / integration (push) Successful in 31s
CI & Build / TypeScript typecheck (push) Successful in 34s
CI & Build / Python tests (push) Successful in 56s
CI & Build / Build & push image (push) Successful in 44s
feat(snippets): usage signal — was a surfaced record ever actually pulled?
retrieval_logs answers "what did the ranker return, at what scores" — the
right substrate for tuning a threshold. It cannot answer the question the
snippet corpus actually needs: did anyone open this? A snippet nobody
opens is not neutral. It takes a slot in every future auto-inject menu and
crowds out something useful.

Adds note_usage_events (migration 0071): one row per note per event,
either 'surfaced' (we put its title in front of an agent) or 'pulled'
(someone opened it in full), tagged with which surface produced it.

Closes the gap #2082 recorded against this work. The write-path PLACE arm
carries no score, so it has no home in retrieval_logs — folding it in
would corrupt the score distribution that table exists to capture. The
result was that the arm firing on the STRONGEST claim ("there is already a
canonical helper in this exact file") was the one arm nobody could
measure. Both arms now emit usage events under distinct sources, so their
pull-through rates are finally comparable.

Deliberate departures from the task as written:

- Not in-session correlation. The original framing was "correlate
  result_ids against a later get_note in the same session." There is no
  session identity server-side — the MCP endpoint is stateless and the
  hooks send no session id — and adding one would mean threading an
  opaque client-supplied token through every read path. Two independent
  counters answer the question without it: surfaced 40×, pulled 0 is dead
  weight regardless of how those events distribute across sessions.

- Pulls record at the ENTRY POINTS (MCP tools, REST detail route), not in
  snippets_svc.get_snippet, which update and merge also reach. Counting
  those would inflate precisely the number meant to say "someone chose to
  look at this."

- get_note records for every note kind, not just snippets. The auto-inject
  menu surfaces tasks and processes too; scoping this to snippets would
  pin those at zero pulls forever and make them read as dead weight next
  to snippets that merely had a counter.

Surfaced in the Snippets list as an "N/M used" badge, warning-toned once a
record has been offered 3+ times and never opened, with the tooltip saying
what to do about it (usually: its "when to reach for it" doesn't say
when). No badge at all below one surfacing — "0/0" reads as a verdict when
it's an absence of evidence. Also returned from MCP list_snippets so the
agent can see dead weight without opening the UI.

Telemetry keeps the retrieval_telemetry contract throughout: writes are
fire-and-forget, reads degrade to zeroes, and no path can raise into the
surface it observes.

Refs #2085

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UaYUaouG9jjhATyuxCKrQs
2026-07-28 18:09:17 -04:00

186 lines
7.1 KiB
Python

"""Tests for the usage signal (#2085) — was a surfaced record ever pulled?
Covers the payload shaping and the two contracts that make this safe to leave in
the hot path: telemetry never raises, and telemetry never blocks. Plus the thing
the feature exists for — that the write-path PLACE arm is now recorded, since
before this it surfaced snippets while leaving no trace anywhere.
"""
from unittest.mock import AsyncMock, MagicMock, patch
import pytest
from scribe.services import note_usage
from scribe.services.note_usage import (
empty_usage,
record_pulled,
record_surfaced,
usage_for_notes,
)
def _note(nid, title, user_id=1):
n = MagicMock()
n.id, n.title, n.user_id = nid, title, user_id
n.note_type = "snippet"
return n
# --- recording ------------------------------------------------------------
async def test_record_surfaced_writes_one_row_per_note():
with patch.object(note_usage, "_schedule") as sched:
record_surfaced(user_id=7, note_ids=[11, 12], source="auto_inject")
rows = sched.call_args[0][0]
assert [r["note_id"] for r in rows] == [11, 12]
assert {r["event"] for r in rows} == {"surfaced"}
assert {r["source"] for r in rows} == {"auto_inject"}
assert {r["user_id"] for r in rows} == {7}
async def test_record_pulled_writes_a_single_row():
with patch.object(note_usage, "_schedule") as sched:
record_pulled(user_id=7, note_id=11, source="mcp_get_snippet")
rows = sched.call_args[0][0]
assert rows == [
{"user_id": 7, "note_id": 11, "event": "pulled", "source": "mcp_get_snippet"}
]
async def test_empty_menu_schedules_nothing():
"""No notes surfaced is not an event — it must not cost a write."""
with patch.object(note_usage, "_insert_events") as ins:
record_surfaced(user_id=1, note_ids=[], source="auto_inject")
ins.assert_not_called()
async def test_recording_never_raises_on_bad_input():
"""Telemetry sits in the hot path of every retrieval. A malformed id must
cost a data point, never the operator's request."""
with patch.object(note_usage, "_schedule"):
record_surfaced(user_id=1, note_ids=["not-an-int"], source="auto_inject")
record_pulled(user_id=1, note_id=None, source="mcp_get_note")
async def test_recording_without_an_event_loop_is_skipped_not_raised():
"""Called from a sync context outside the app (a script, a test helper),
there is no loop to schedule on. Skip rather than blow up."""
with patch.object(
note_usage.asyncio, "get_running_loop", side_effect=RuntimeError
):
record_pulled(user_id=1, note_id=5, source="mcp_get_note")
# --- readout --------------------------------------------------------------
async def test_usage_for_notes_zero_fills_every_requested_id():
"""The caller renders this shape unconditionally, so a note with no events
must come back as zeroes, not as a missing key."""
session = MagicMock()
session.execute = AsyncMock(return_value=MagicMock(all=MagicMock(return_value=[])))
ctx = MagicMock()
ctx.__aenter__ = AsyncMock(return_value=session)
ctx.__aexit__ = AsyncMock(return_value=False)
with patch.object(note_usage, "async_session", return_value=ctx):
out = await usage_for_notes([3, 4])
assert out == {3: empty_usage(), 4: empty_usage()}
async def test_usage_for_notes_splits_counts_by_event():
from datetime import datetime, timezone
ts = datetime(2026, 7, 28, tzinfo=timezone.utc)
rows = [(3, "surfaced", 9, ts), (3, "pulled", 2, ts)]
session = MagicMock()
session.execute = AsyncMock(
return_value=MagicMock(all=MagicMock(return_value=rows))
)
ctx = MagicMock()
ctx.__aenter__ = AsyncMock(return_value=session)
ctx.__aexit__ = AsyncMock(return_value=False)
with patch.object(note_usage, "async_session", return_value=ctx):
out = await usage_for_notes([3])
assert out[3]["surfaced_count"] == 9
assert out[3]["pull_count"] == 2
assert out[3]["last_pulled_at"] == ts.isoformat()
async def test_usage_readout_failure_degrades_to_zeroes():
"""A telemetry readout must not be able to break the list it decorates."""
with patch.object(note_usage, "async_session", side_effect=RuntimeError("boom")):
out = await usage_for_notes([3])
assert out == {3: empty_usage()}
async def test_no_ids_short_circuits_without_a_query():
with patch.object(note_usage, "async_session") as sess:
assert await usage_for_notes([]) == {}
sess.assert_not_called()
# --- the gap this closes --------------------------------------------------
@pytest.mark.parametrize(
"marker, expected_source",
[("here", "write_path_place"), ("nearby", "write_path_place")],
)
async def test_write_path_place_arm_is_recorded(marker, expected_source):
"""The place arm carries no score, so it has no home in retrieval_logs — it
surfaced snippets while leaving no trace anywhere. That was the blocker
#2082 recorded against this task; this is the assertion that it's closed."""
from scribe.services import plugin_context
here = [{"id": 42, "title": "helper", "user_id": 1, "note_type": "snippet"}]
with (
patch.object(
plugin_context,
"get_writepath_config",
AsyncMock(return_value={"enabled": True, "threshold": 0.55, "top_k": 3}),
),
patch.object(
plugin_context.snippets_svc,
"list_snippets",
AsyncMock(side_effect=[(here, 1), ([], 0)]),
),
patch.object(
plugin_context, "semantic_search_notes", AsyncMock(return_value=[])
),
patch.object(plugin_context, "owner_names_for", AsyncMock(return_value={})),
patch.object(plugin_context, "record_retrieval"),
patch.object(plugin_context, "record_surfaced") as surfaced,
):
out = await plugin_context.build_write_path_hint(
1, "src/a.py", code="def f(): pass"
)
assert out["note_ids"] == [42]
sources = {c.kwargs["source"] for c in surfaced.call_args_list}
assert expected_source in sources
async def test_auto_inject_records_what_survived_the_margin_gate():
"""Not what the ranker returned — retrieval_logs already holds that. These
two numbers must not silently mean different things per surface."""
from scribe.services import plugin_context
hits = [(0.90, _note(1, "kept")), (0.40, _note(2, "cut by the margin gate"))]
with (
patch.object(
plugin_context,
"get_autoinject_config",
AsyncMock(return_value={"enabled": True, "threshold": 0.3, "top_k": 5}),
),
patch.object(
plugin_context, "semantic_search_notes", AsyncMock(return_value=hits)
),
patch.object(plugin_context, "owner_names_for", AsyncMock(return_value={})),
patch.object(plugin_context, "record_retrieval"),
patch.object(plugin_context, "record_surfaced") as surfaced,
):
await plugin_context.build_autoinject_hint(1, "a query")
assert surfaced.call_args.kwargs["note_ids"] == [1]
assert surfaced.call_args.kwargs["source"] == "auto_inject"