Files
chat/tests/test_state_update.py
T
Joseph Doherty 5aab98e4d7 fix: classifier robustness — schema in prompt, retries, kickoff fallback
The kickoff parse-and-confirm route was 500-ing intermittently because
Hermes-3 + Featherless's response_format={"type":"json_object"} only
guarantees JSON output, NOT a particular schema. The model was inventing
its own field names (sceneTime, entities, settingDetails) instead of
the KickoffParse fields, causing Pydantic validation to fail on both
classify() retries.

Three changes:

1. Include the Pydantic JSON schema in the system prompt so the model
   knows exactly which keys to produce. Affects every classify() call
   (kickoff parse, turn parse, scene-close detect, significance,
   state-update, scene summarize). Strip ```json fences if the model
   wraps its output. Bump retries 2 → 3 (model is stochastic; one extra
   attempt closes most of the remaining gap).

2. parse_kickoff() now passes a default empty KickoffParse so the
   route degrades to a fillable form instead of 500 when the classifier
   ultimately fails. The confirm form is the human-in-the-loop; an
   empty form is strictly better UX than a stack trace.

3. Tests updated: bumped canned-failure arrays from 2 → 3 entries to
   match the new attempt count; renamed kickoff test from
   "raises_when_classifier_fails_twice" to
   "falls_back_to_empty_when_classifier_fails" reflecting the new
   degraded-but-usable behavior.

Verified live with all 3 sample bots (maya/eli/sam) — kickoff route
returns 200 across multiple attempts. Full suite: 168 passed.
2026-04-26 15:03:13 -04:00

238 lines
7.7 KiB
Python

"""Post-turn state-update pass (T20).
Per Requirements §3.4, after each utterance we run a classifier on every
present entity (silent witnesses too) to extract directed-edge deltas
(``affinity_delta``, ``trust_delta``, ``knowledge_facts``). The deltas
land as ``edge_update`` events and project into the ``edges`` table.
These tests cover:
- The unit-level :func:`compute_state_update` happy path: classifier
returns valid JSON, the wrapper returns a populated ``StateUpdate``.
- The unit-level fallback path: classifier fails twice, the wrapper
returns a no-op ``StateUpdate`` (zeros + empty facts) per §3.3.
- The integration path: a successful POST appends two ``edge_update``
events (one per direction) after the ``assistant_turn`` and the edge
projections reflect the deltas.
"""
from __future__ import annotations
import json
from pathlib import Path
import pytest
from fastapi.testclient import TestClient
from chat.app import app
from chat.db.connection import open_db
from chat.eventlog.log import append_event
from chat.eventlog.projector import project
from chat.llm.mock import MockLLMClient
from chat.services.state_update import StateUpdate, compute_state_update
@pytest.mark.asyncio
async def test_compute_state_update_parses_classifier_output():
canned = json.dumps(
{"affinity_delta": 2, "trust_delta": 1, "knowledge_facts": ["likes coffee"]}
)
mock = MockLLMClient(canned=[canned])
result = await compute_state_update(
mock,
model="x",
source_id="bot_a",
target_id="you",
source_name="BotA",
source_persona="thoughtful",
target_name="Me",
prior_affinity=50,
prior_trust=50,
prior_summary="",
recent_dialogue=[
{"speaker": "you", "text": "hi"},
{"speaker": "BotA", "text": "Hello!"},
],
)
assert isinstance(result, StateUpdate)
assert result.affinity_delta == 2
assert result.trust_delta == 1
assert result.knowledge_facts == ["likes coffee"]
@pytest.mark.asyncio
async def test_compute_state_update_returns_default_on_failure():
"""Two malformed classifier responses -> default StateUpdate (zeros)."""
mock = MockLLMClient(canned=["nope", "still nope", "nope3"])
result = await compute_state_update(
mock,
model="x",
source_id="bot_a",
target_id="you",
source_name="BotA",
source_persona="",
target_name="Me",
prior_affinity=50,
prior_trust=50,
prior_summary="",
recent_dialogue=[],
)
assert result.affinity_delta == 0
assert result.trust_delta == 0
assert result.knowledge_facts == []
# --- integration test --------------------------------------------------------
@pytest.fixture
def client(tmp_path, monkeypatch):
cfg = tmp_path / "config.toml"
cfg.write_text('featherless_api_key = "test"\n')
monkeypatch.setenv("CHAT_CONFIG_PATH", str(cfg))
db = tmp_path / "test.db"
monkeypatch.setenv("CHAT_DB_PATH", str(db))
canned_parse = json.dumps(
{"segments": [{"kind": "dialogue", "text": "hello"}]}
)
canned_response = "Hi there."
canned_state_b2y = json.dumps(
{"affinity_delta": 2, "trust_delta": 1, "knowledge_facts": ["greets warmly"]}
)
canned_state_y2b = json.dumps(
{"affinity_delta": 3, "trust_delta": 0, "knowledge_facts": []}
)
from chat.web.kickoff import get_llm_client
mock = MockLLMClient(
canned=[canned_parse, canned_response, canned_state_b2y, canned_state_y2b]
)
app.dependency_overrides[get_llm_client] = lambda: mock
with TestClient(app) as c:
c.mock_llm = mock # type: ignore[attr-defined]
yield c
app.dependency_overrides.clear()
def _seed(db_path: Path) -> None:
"""Author a bot, create a chat, and seed enough state for prompt assembly."""
with open_db(db_path) as conn:
append_event(
conn,
kind="bot_authored",
payload={
"id": "bot_a",
"name": "BotA",
"persona": "thoughtful, observant",
"voice_samples": [],
"traits": [],
"backstory": "",
"initial_relationship_to_you": "",
"kickoff_prose": "...",
},
)
append_event(
conn,
kind="chat_created",
payload={
"id": "chat_bot_a",
"host_bot_id": "bot_a",
"initial_time": "2026-04-26T20:00:00+00:00",
"narrative_anchor": "Day 1",
"weather": "",
},
)
append_event(
conn,
kind="edge_update",
payload={
"source_id": "bot_a",
"target_id": "you",
"chat_id": "chat_bot_a",
"knowledge_facts": ["coworker"],
},
)
append_event(
conn,
kind="activity_change",
payload={
"entity_id": "you",
"posture": "sitting",
"action": {
"verb": "talking",
"interruptible": True,
"required_attention": "low",
"expected_duration": "ongoing",
},
"attention": "",
"holding": [],
"status": {},
},
)
append_event(
conn,
kind="activity_change",
payload={
"entity_id": "bot_a",
"posture": "sitting",
"action": {
"verb": "listening",
"interruptible": True,
"required_attention": "low",
"expected_duration": "ongoing",
},
"attention": "",
"holding": [],
"status": {},
},
)
project(conn)
def test_post_turn_appends_edge_updates_and_applies_deltas(client, tmp_path):
"""After a turn, edge_update events fire for both directions and project."""
db_path = tmp_path / "test.db"
_seed(db_path)
response = client.post("/chats/chat_bot_a/turns", data={"prose": "hello"})
assert response.status_code == 204
with open_db(db_path) as conn:
# Two new edge_update events should land *after* the assistant_turn.
cur = conn.execute(
"SELECT kind, payload_json FROM event_log "
"WHERE kind = 'edge_update' "
"AND id > (SELECT MAX(id) FROM event_log WHERE kind = 'assistant_turn') "
"ORDER BY id"
)
rows = cur.fetchall()
assert len(rows) == 2
kinds = [r[0] for r in rows]
assert kinds == ["edge_update", "edge_update"]
# Inspect the two payloads — one per direction.
payloads = [json.loads(r[1]) for r in rows]
directions = {(p["source_id"], p["target_id"]) for p in payloads}
assert ("bot_a", "you") in directions
assert ("you", "bot_a") in directions
# Edge bot_a -> you: seeded affinity=50, plus delta 2 -> 52.
from chat.state.edges import get_edge
edge_b2y = get_edge(conn, "bot_a", "you")
assert edge_b2y is not None
assert edge_b2y["affinity"] == 52
assert edge_b2y["trust"] == 51
# Existing fact preserved, new fact appended.
assert "coworker" in edge_b2y["knowledge"]
assert "greets warmly" in edge_b2y["knowledge"]
# Edge you -> bot_a: defaults (50/50) plus delta +3 affinity -> 53.
edge_y2b = get_edge(conn, "you", "bot_a")
assert edge_y2b is not None
assert edge_y2b["affinity"] == 53
assert edge_y2b["trust"] == 50