51b886f3f6
REQ-317: core/metrics/outcome_backfill.py backfills fact_decision.outcome pending -> succeeded/failed after run.completed/run.failed; idempotent + terminal (does not overwrite a non-pending outcome); wired into the collector. The Post-Pilot AI Decision Accuracy denominator is now grounded (fact_decision.outcome is not stuck pending). REQ-318: ai.decision.made on a block band carries escalation_reason: 'confidence' (the only value in v1.26 — a block is always confidence- driven; future milestones may add 'policy'). Persisted into fact_run by the collector. The Post-Pilot Human Escalation Frequency denominator is now grounded. ---ci--- project: acdl phase: 3 milestone: v1.26 status: execute wave: W2 ---
211 lines
8.0 KiB
Python
211 lines
8.0 KiB
Python
"""Tests for escalation_reason on ai.decision.made (REQ-318, SPEC §5.8, P3 W2).
|
|
|
|
Covers:
|
|
* block band carries escalation_reason == "confidence" + human_override True
|
|
* pass band has escalation_reason ABSENT + human_override False
|
|
* fact_run persists escalation_reason (collector wiring)
|
|
|
|
Follows the fixture pattern in tests/test_metrics_emitters.py.
|
|
"""
|
|
|
|
import json
|
|
import os
|
|
import sqlite3
|
|
import sys
|
|
from pathlib import Path
|
|
|
|
import pytest
|
|
|
|
ROOT = Path(__file__).resolve().parent.parent
|
|
sys.path.insert(0, str(ROOT))
|
|
|
|
|
|
@pytest.fixture
|
|
def tmp_metrics(tmp_path, monkeypatch):
|
|
"""Redirect metrics/ to a tmp dir for isolated testing."""
|
|
metrics_dir = tmp_path / "metrics"
|
|
metrics_dir.mkdir()
|
|
runs_dir = metrics_dir / "runs"
|
|
runs_dir.mkdir()
|
|
events_log = metrics_dir / "events.jsonl"
|
|
ledger_db = metrics_dir / "decision_ledger.db"
|
|
store_db = metrics_dir / "nova_metrics.db"
|
|
|
|
monkeypatch.setattr("core.metrics.event_envelope.METRICS_DIR", str(metrics_dir))
|
|
monkeypatch.setattr("core.metrics.event_envelope.EVENTS_LOG", str(events_log))
|
|
monkeypatch.setattr("core.metrics.run_manifest._METRICS_DIR", str(metrics_dir))
|
|
monkeypatch.setattr("core.metrics.run_manifest._RUNS_DIR", str(runs_dir))
|
|
monkeypatch.setattr("core.metrics.decision_ledger._LEDGER_PATH", str(ledger_db))
|
|
monkeypatch.setattr("core.metrics.collector._METRICS_DIR", str(metrics_dir))
|
|
monkeypatch.setattr("core.metrics.collector._STORE_PATH", str(store_db))
|
|
monkeypatch.setattr("core.metrics.collector._RUNS_DIR", str(runs_dir))
|
|
monkeypatch.setattr("core.metrics.collector._LEDGER_DB", str(ledger_db))
|
|
monkeypatch.setattr("core.metrics.outcome_backfill._METRICS_DIR", str(metrics_dir))
|
|
monkeypatch.setattr("core.metrics.outcome_backfill._STORE_PATH", str(store_db))
|
|
monkeypatch.setattr("core.metrics.outcome_backfill._LEDGER_PATH", str(ledger_db))
|
|
return {
|
|
"metrics_dir": metrics_dir,
|
|
"events_log": events_log,
|
|
"ledger_db": ledger_db,
|
|
"store_db": store_db,
|
|
"runs_dir": runs_dir,
|
|
}
|
|
|
|
|
|
def _base_inputs():
|
|
return {
|
|
"policy": [{"result": "pass", "severity": "info"}],
|
|
"validation": {"schema": True, "stack_resolved": True,
|
|
"tf_validated": True, "tf_planned": True},
|
|
"freshness": {"age_days": 0, "max_age_days": 7},
|
|
"source": {"submitter": "dev", "commit_sha": "abc"},
|
|
"history": {"prior_rollbacks": 0, "prior_policy_fails": 0},
|
|
"nfrs": {"conformance": 1.0},
|
|
}
|
|
|
|
|
|
def _block_inputs():
|
|
"""A critical PCR triggers a hard override (score=0, band=block)."""
|
|
inputs = _base_inputs()
|
|
inputs["policy"] = [{"result": "fail", "severity": "critical", "ruleId": "CKV_X"}]
|
|
return inputs
|
|
|
|
|
|
def _read_decision_event(events_log):
|
|
lines = events_log.read_text().strip().split("\n")
|
|
for line in lines:
|
|
ev = json.loads(line)
|
|
if ev["type"] == "nova.ai.decision.made":
|
|
return ev
|
|
return None
|
|
|
|
|
|
def test_block_band_has_escalation_reason(tmp_metrics):
|
|
"""A block (critical PCR hard override) carries escalation_reason='confidence'."""
|
|
from core.confidence_signal import compute
|
|
sig = compute("cid-block-1", "dev", _block_inputs())
|
|
assert sig.band == "block"
|
|
ev = _read_decision_event(tmp_metrics["events_log"])
|
|
assert ev is not None
|
|
data = ev["data"]
|
|
assert data["chosen_action"] == "block"
|
|
assert data["human_override"] is True
|
|
assert data.get("escalation_reason") == "confidence"
|
|
|
|
|
|
def test_pass_band_no_escalation_reason(tmp_metrics):
|
|
"""A clean dev apply (pass band) has NO escalation_reason + human_override False."""
|
|
from core.confidence_signal import compute
|
|
sig = compute("cid-pass-1", "dev", _base_inputs())
|
|
assert sig.band == "pass"
|
|
ev = _read_decision_event(tmp_metrics["events_log"])
|
|
assert ev is not None
|
|
data = ev["data"]
|
|
assert data["chosen_action"] == "pass"
|
|
assert data["human_override"] is False
|
|
# escalation_reason must be ABSENT on a non-block band.
|
|
assert "escalation_reason" not in data
|
|
|
|
|
|
def test_low_confidence_block_has_escalation_reason(tmp_metrics):
|
|
"""A score below (threshold - 0.10) blocks on confidence grounds."""
|
|
from core.confidence_signal import compute
|
|
# Freshness maximally stale + a high-severity policy fail drags the
|
|
# score well below the dev threshold of 0.50 - 0.10 = 0.40.
|
|
inputs = _base_inputs()
|
|
inputs["freshness"] = {"age_days": 7, "max_age_days": 7}
|
|
inputs["policy"] = [{"result": "fail", "severity": "high", "ruleId": "CKV_Y"}]
|
|
sig = compute("cid-block-2", "dev", inputs)
|
|
assert sig.band == "block"
|
|
ev = _read_decision_event(tmp_metrics["events_log"])
|
|
data = ev["data"]
|
|
assert data.get("escalation_reason") == "confidence"
|
|
|
|
|
|
def test_fact_run_persists_escalation_reason(tmp_metrics):
|
|
"""The collector persists escalation_reason into fact_run + fact_decision.
|
|
|
|
End-to-end: confidence_signal emits ai.decision.made (block) →
|
|
run_manifest.complete_run writes the manifest with escalation_reason →
|
|
collector.collect_run_manifests + collect_decision_ledger populate
|
|
fact_run.escalation_reason + fact_decision.escalation_reason.
|
|
"""
|
|
from core.confidence_signal import compute
|
|
from core.metrics.run_manifest import complete_run
|
|
from core.metrics.collector import collect_run_manifests, collect_decision_ledger
|
|
|
|
# Emit a block decision.
|
|
os.environ["NOVA_RUN_ID"] = "run-esc-1"
|
|
try:
|
|
sig = compute("cid-esc-1", "dev", _block_inputs())
|
|
assert sig.band == "block"
|
|
finally:
|
|
os.environ.pop("NOVA_RUN_ID", None)
|
|
|
|
# Complete the run with escalation_reason carried into the manifest.
|
|
manifest = complete_run(
|
|
"run-esc-1", "cid-esc-1", "dev",
|
|
stages=[{"name": "apply", "duration_ms": 100, "exit_code": 0}],
|
|
exit_code=0,
|
|
confidence={"score": sig.score, "band": sig.band, "perInput": sig.perInput},
|
|
decision_id="run-esc-1",
|
|
escalation_reason="confidence",
|
|
)
|
|
assert manifest["escalation_reason"] == "confidence"
|
|
|
|
# Collector reads the manifest → fact_run.
|
|
collect_run_manifests()
|
|
# Collector reads the ledger → fact_decision.
|
|
collect_decision_ledger()
|
|
|
|
conn = sqlite3.connect(str(tmp_metrics["store_db"]))
|
|
conn.row_factory = sqlite3.Row
|
|
run_row = conn.execute(
|
|
"SELECT run_id, escalation_reason, decision_id FROM fact_run WHERE run_id = ?",
|
|
("run-esc-1",),
|
|
).fetchone()
|
|
dec_row = conn.execute(
|
|
"SELECT decision_id, escalation_reason, human_override FROM fact_decision WHERE decision_id = ?",
|
|
("run-esc-1",),
|
|
).fetchone()
|
|
conn.close()
|
|
|
|
assert run_row is not None
|
|
assert run_row["escalation_reason"] == "confidence"
|
|
assert run_row["decision_id"] == "run-esc-1"
|
|
assert dec_row is not None
|
|
assert dec_row["escalation_reason"] == "confidence"
|
|
assert dec_row["human_override"] == 1
|
|
|
|
|
|
def test_fact_run_no_escalation_reason_on_pass(tmp_metrics):
|
|
"""A pass-band run has escalation_reason NULL in fact_run."""
|
|
from core.confidence_signal import compute
|
|
from core.metrics.run_manifest import complete_run
|
|
from core.metrics.collector import collect_run_manifests
|
|
|
|
os.environ["NOVA_RUN_ID"] = "run-esc-pass-1"
|
|
try:
|
|
sig = compute("cid-esc-pass-1", "dev", _base_inputs())
|
|
assert sig.band == "pass"
|
|
finally:
|
|
os.environ.pop("NOVA_RUN_ID", None)
|
|
|
|
complete_run(
|
|
"run-esc-pass-1", "cid-esc-pass-1", "dev",
|
|
stages=[{"name": "apply", "duration_ms": 100, "exit_code": 0}],
|
|
exit_code=0,
|
|
confidence={"score": sig.score, "band": sig.band, "perInput": sig.perInput},
|
|
decision_id="run-esc-pass-1",
|
|
# escalation_reason intentionally omitted (pass band).
|
|
)
|
|
collect_run_manifests()
|
|
conn = sqlite3.connect(str(tmp_metrics["store_db"]))
|
|
conn.row_factory = sqlite3.Row
|
|
row = conn.execute(
|
|
"SELECT run_id, escalation_reason FROM fact_run WHERE run_id = ?",
|
|
("run-esc-pass-1",),
|
|
).fetchone()
|
|
conn.close()
|
|
assert row is not None
|
|
assert row["escalation_reason"] is None |