Spaces:
Running
Running
File size: 9,548 Bytes
69e310f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 | """End-to-end: the whole governance chain, against live market data.
trigger → supervisor → parallel workers → writer → verify → human gate → deliver
Every assertion here is about the *chain*, not about a particular market value:
that a figure reconciles, that a fault is caught, that nothing ships without a
human. Those hold whatever the market did today.
"""
from __future__ import annotations
import asyncio
from typing import Any
import pytest
from app.core.settings import Settings
from app.models.run import HumanDecision, RunMode, RunStatus
from app.services.repository import ActiveRunExistsError
from app.services.runner import RunService
SETTLE_TIMEOUT_S = 180.0
async def _settle(service: RunService, run_id: str) -> dict[str, Any]:
"""Wait until the run leaves RUNNING."""
deadline = asyncio.get_running_loop().time() + SETTLE_TIMEOUT_S
while asyncio.get_running_loop().time() < deadline:
record = await service.get_run(run_id)
if record and record["status"] not in ("RUNNING", "QUEUED"):
return record
await asyncio.sleep(0.25)
pytest.fail(f"run {run_id} did not settle within {SETTLE_TIMEOUT_S}s")
@pytest.fixture
def service(settings: Settings) -> RunService:
return RunService(settings=settings)
class TestHappyPath:
async def test_full_chain_from_trigger_to_delivery(self, service: RunService) -> None:
run_id = await service.start_run(tickers=["AAPL"], trigger="test")
record = await _settle(service, run_id)
# 1. It paused for a human rather than delivering itself.
assert record["status"] == RunStatus.AWAITING_APPROVAL
gate = await service.gate_payload(run_id)
assert gate is not None
report = gate["verification"]
# 2. Every numeric claim was recomputed and matched.
assert gate["verified"] is True
assert report["checked_claims"] > 0
assert report["matched"] == report["checked_claims"]
assert report["mismatched"] == 0
assert report["coverage"] == 1.0
# 3. The brief is well-formed and renders.
assert gate["brief"]["snapshot"]
assert gate["markdown"].startswith("# AlphaBrief")
# 4. Approval releases it.
result = await service.submit_decision(
run_id, HumanDecision(action="approve", reviewer="tester")
)
assert result["status"] == RunStatus.DELIVERED
final = await service.get_run(run_id)
assert final is not None
assert final["status"] == RunStatus.DELIVERED
assert final["verified"] is True
assert final["iterations"] >= 1
async def test_telemetry_covers_the_whole_chain(self, service: RunService) -> None:
run_id = await service.start_run(tickers=["MSFT"], trigger="test")
await _settle(service, run_id)
kinds = {str(event.kind) for event in await service.bus.history(run_id)}
for expected in (
"run.started",
"supervisor.plan",
"agent.started",
"agent.completed",
"mcp.tool_call",
"writer.started",
"verify.started",
"verify.claim",
"verify.completed",
"gate.awaiting",
):
assert expected in kinds, f"missing telemetry: {expected}"
await service.submit_decision(run_id, HumanDecision(action="approve"))
async def test_tool_calls_carry_arguments_and_timing(self, service: RunService) -> None:
run_id = await service.start_run(tickers=["NVDA"], trigger="test")
await _settle(service, run_id)
events = [e for e in await service.bus.history(run_id) if str(e.kind) == "mcp.tool_call"]
assert events
for event in events:
assert "tool" in event.payload
assert "arguments" in event.payload
assert isinstance(event.payload["duration_ms"], float)
await service.submit_decision(run_id, HumanDecision(action="approve"))
class TestHumanGate:
async def test_rejection_blocks_delivery(self, service: RunService) -> None:
run_id = await service.start_run(tickers=["TSLA"], trigger="test")
await _settle(service, run_id)
result = await service.submit_decision(
run_id, HumanDecision(action="reject", note="not today")
)
assert result["status"] == RunStatus.REJECTED
record = await service.get_run(run_id)
assert record is not None
assert record["status"] == RunStatus.REJECTED
async def test_edit_applies_narrative_only(self, service: RunService) -> None:
run_id = await service.start_run(tickers=["AMZN"], trigger="test")
await _settle(service, run_id)
await service.submit_decision(
run_id,
HumanDecision(
action="edit",
edited_headline="Reviewer rewrote this headline",
reviewer="tester",
),
)
stored = await service.repository.latest_brief(run_id)
assert stored is not None
assert stored["brief"]["headline"] == "Reviewer rewrote this headline"
async def test_an_edit_containing_a_numeral_is_refused(self, service: RunService) -> None:
"""Reviewers may reword; they may not introduce an unverified figure."""
run_id = await service.start_run(tickers=["META"], trigger="test")
await _settle(service, run_id)
gate_before = await service.gate_payload(run_id)
assert gate_before is not None
original = gate_before["brief"]["headline"]
await service.submit_decision(
run_id,
HumanDecision(action="edit", edited_headline="Stock jumped 42 percent today"),
)
stored = await service.repository.latest_brief(run_id)
assert stored is not None
assert stored["brief"]["headline"] == original
async def test_decision_on_an_unknown_run_is_refused(self, service: RunService) -> None:
from app.services.runner import RunNotFoundError
with pytest.raises(RunNotFoundError):
await service.submit_decision("run_does_not_exist", HumanDecision(action="approve"))
async def test_a_second_decision_is_refused(self, service: RunService) -> None:
from app.services.runner import RunNotFoundError
run_id = await service.start_run(tickers=["GOOGL"], trigger="test")
await _settle(service, run_id)
await service.submit_decision(run_id, HumanDecision(action="approve"))
with pytest.raises(RunNotFoundError):
await service.submit_decision(run_id, HumanDecision(action="approve"))
class TestFaultInjectionModes:
async def test_demo_fault_degrades_without_crashing(self, service: RunService) -> None:
"""The interview kill shot: a fake ticker must not break the run."""
run_id = await service.start_run(tickers=["AAPL"], mode=RunMode.DEMO_FAULT)
record = await _settle(service, run_id)
assert record["status"] == RunStatus.AWAITING_APPROVAL
assert record["status"] != RunStatus.FAILED
assert record["partial"] is True
assert record["error_count"] > 0
gate = await service.gate_payload(run_id)
assert gate is not None
brief = gate["brief"]
assert brief["partial"] is True
assert brief["data_gaps"]
# The dead ticker is in the watchlist but not in the snapshot.
assert "ZZZZQQQQ" in brief["watchlist"]
assert "ZZZZQQQQ" not in [row["ticker"] for row in brief["snapshot"]]
# Everything that *did* resolve still verified cleanly.
assert gate["verified"] is True
await service.submit_decision(run_id, HumanDecision(action="approve"))
async def test_demo_mismatch_is_caught_and_escalated(self, service: RunService) -> None:
"""Injected bad figure: one regeneration, then HUMAN_REVIEW. Never delivered."""
run_id = await service.start_run(tickers=["MSFT"], mode=RunMode.DEMO_MISMATCH)
record = await _settle(service, run_id)
assert record["status"] == RunStatus.HUMAN_REVIEW
gate = await service.gate_payload(run_id)
assert gate is not None
assert gate["verified"] is False
assert gate["requires_review"] is True
assert gate["verification"]["mismatched"] >= 1
events = await service.bus.history(run_id)
# Exactly one regeneration was attempted before escalating.
assert sum(1 for e in events if str(e.kind) == "writer.started") == 2
assert sum(1 for e in events if str(e.kind) == "verify.completed") == 2
await service.submit_decision(run_id, HumanDecision(action="reject", note="bad numbers"))
final = await service.get_run(run_id)
assert final is not None
assert final["status"] == RunStatus.REJECTED
class TestConcurrencyGuard:
async def test_one_active_run_per_watchlist(self, service: RunService) -> None:
run_id = await service.start_run(tickers=["INTC"], trigger="test")
with pytest.raises(ActiveRunExistsError):
await service.start_run(tickers=["intc"], trigger="test")
await _settle(service, run_id)
await service.submit_decision(run_id, HumanDecision(action="approve"))
# Slot released.
second = await service.start_run(tickers=["INTC"], trigger="test")
await _settle(service, second)
await service.submit_decision(second, HumanDecision(action="approve"))
|