add deterministic pilot benchmarks
This commit is contained in:
@@ -0,0 +1,82 @@
|
||||
{
|
||||
"seed": 1601,
|
||||
"limitations": [
|
||||
"Synthetic, small, English-heavy cases are not representative of production traffic.",
|
||||
"Contact labels cover parser false positives but do not establish consent, identity, or deliverability.",
|
||||
"Website fixtures classify supplied HTML only; they do not measure network, DNS, redirects, or adversarial pages.",
|
||||
"Latency and cache measurements are local pilot signals and vary by host load."
|
||||
],
|
||||
"normalization": [
|
||||
{"name": "SA contact fields", "input": {"name": " Acme Solar ", "website": "HTTPS://WWW.ACME.TEST/path", "phone": "082 555 1234", "email": " SALES@ACME.TEST "}, "expected": {"name": "Acme Solar", "website_domain": "acme.test", "email": "sales@acme.test"}},
|
||||
{"name": "accented location", "input": {"name": "Cafe", "province": " Western Cape ", "city": "CAPE TOWN", "suburb": "Sea Point"}, "expected": {"province": "western cape", "city": "cape town", "suburb": "sea point"}}
|
||||
],
|
||||
"matching": [
|
||||
{
|
||||
"name": "shared exact domain is a positive",
|
||||
"source": {"name": "Acme Solar", "website": "https://www.acme.test", "city": "Johannesburg"},
|
||||
"candidates": [
|
||||
{"id": 1, "name": "Acme Solar (Pty) Ltd", "website": "http://acme.test/", "city": "Johannesburg"},
|
||||
{"id": 2, "name": "Acme Supplies", "website": "https://supplies.test", "city": "Johannesburg"}
|
||||
],
|
||||
"threshold": 0.72,
|
||||
"expected_ids": [1],
|
||||
"forbidden_ids": [2]
|
||||
},
|
||||
{
|
||||
"name": "similar name without corroboration is not a positive",
|
||||
"source": {"name": "Brightline Consulting", "city": "Cape Town"},
|
||||
"candidates": [
|
||||
{"id": 3, "name": "Brightline Consulting", "city": "Durban"},
|
||||
{"id": 4, "name": "Brightline Consulting", "city": "Cape Town"}
|
||||
],
|
||||
"threshold": 0.72,
|
||||
"expected_ids": [4],
|
||||
"forbidden_ids": [3]
|
||||
},
|
||||
{
|
||||
"name": "unrelated businesses are negatives",
|
||||
"source": {"name": "Green Oak Architects", "website": "green-oak.test"},
|
||||
"candidates": [
|
||||
{"id": 5, "name": "Green Oak Accounting", "website": "accounting.test"},
|
||||
{"id": 6, "name": "Red River Bakery", "website": "redriver.test"}
|
||||
],
|
||||
"threshold": 0.72,
|
||||
"expected_ids": [],
|
||||
"forbidden_ids": [5, 6]
|
||||
}
|
||||
],
|
||||
"contacts": [
|
||||
{
|
||||
"name": "public contacts and ignored markup",
|
||||
"source_url": "https://acme.test/contact",
|
||||
"html": "<p>Sales: sales@acme.test</p><p>Call +27 (12) 345-6789</p><script>secret@acme.test</script><img src=\"https://tracker.test/pixel?email=tracker@tracker.test\"><p>API key: token@acme.test</p>",
|
||||
"expected": [["email", "sales@acme.test"], ["phone", "+27123456789"]],
|
||||
"forbidden": [["email", "secret@acme.test"], ["email", "tracker@tracker.test"], ["email", "token@acme.test"]]
|
||||
},
|
||||
{
|
||||
"name": "forms are provenance not email destinations",
|
||||
"source_url": "https://bright.test/contact",
|
||||
"html": "<form><input name=\"email\" placeholder=\"Your email\"></form><p>hello [at] bright.test</p>",
|
||||
"expected": [["email", "hello@bright.test"]],
|
||||
"forbidden": [["email", "email@bright.test"]]
|
||||
}
|
||||
],
|
||||
"websites": [
|
||||
{"name": "healthy", "status": 200, "url": "https://acme.test", "body": "<html><h1>Acme Solar</h1><p>Welcome to our business.</p></html>", "expected": "healthy"},
|
||||
{"name": "parked", "status": 200, "url": "https://parked.test", "body": "Domain for sale - buy this domain", "expected": "parked"},
|
||||
{"name": "construction", "status": 200, "url": "https://new.test", "body": "Website coming soon", "expected": "under_construction"},
|
||||
{"name": "broken", "status": 404, "url": "https://gone.test", "body": "", "expected": "broken"},
|
||||
{"name": "blocked", "status": null, "url": "https://private.test", "body": "", "error": "unsafe_address", "expected": "blocked"},
|
||||
{"name": "unknown is not healthy", "status": null, "url": "https://unknown.test", "body": "", "expected": "unknown"}
|
||||
],
|
||||
"scoring": [
|
||||
{"name": "complete active business", "signals": {"business": {"name": "Acme", "email": "a@acme.test", "phone": "+27123456789"}, "website": {"classification": "healthy"}, "contacts": {"public_count": 1}, "domain": {"status": "resolved"}, "state": {"verified": true, "suppressed": false, "merge_status": "active"}}},
|
||||
{"name": "suppressed remains ineligible", "signals": {"business": {"name": "Suppressed"}, "state": {"suppressed": true, "merge_status": "active"}}, "must_be_ineligible": true},
|
||||
{"name": "stale evidence does not create positive", "signals": {"business": {"name": "Stale"}, "website": {"classification": "healthy", "stale": true}, "state": {"suppressed": false, "merge_status": "active"}}}
|
||||
],
|
||||
"tenant_isolation": {
|
||||
"tenant_a": {"id": 101, "business_ids": [1, 2]},
|
||||
"tenant_b": {"id": 202, "business_ids": [3, 4]},
|
||||
"cross_tenant_ids": [3, 4]
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,61 @@
|
||||
import json
|
||||
import sys
|
||||
import unittest
|
||||
from pathlib import Path
|
||||
|
||||
from app.domain import match_businesses
|
||||
from app.contact_extractor import extract_contacts
|
||||
from app.website_scanner import classify_website
|
||||
from app.scoring import DEFAULT_RULES, evaluate_score
|
||||
|
||||
ROOT = Path(__file__).resolve().parents[1]
|
||||
FIXTURE_PATH = ROOT / "fixtures" / "phase16.json"
|
||||
sys.path.insert(0, str(ROOT.parent.parent / "scripts"))
|
||||
from benchmark_phase16 import run_benchmark
|
||||
|
||||
|
||||
class Phase16BenchmarkRegressionTests(unittest.TestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.fixtures = json.loads(FIXTURE_PATH.read_text())
|
||||
|
||||
def test_fixture_is_stable_and_benchmark_cases_are_deterministic(self):
|
||||
self.assertEqual(self.fixtures["seed"], 1601)
|
||||
first = json.dumps(self.fixtures, sort_keys=True, separators=(",", ":"))
|
||||
second = json.dumps(json.loads(FIXTURE_PATH.read_text()), sort_keys=True, separators=(",", ":"))
|
||||
self.assertEqual(first, second)
|
||||
case = self.fixtures["matching"][0]
|
||||
self.assertEqual(
|
||||
match_businesses(case["source"], case["candidates"], threshold=case["threshold"]),
|
||||
match_businesses(case["source"], case["candidates"], threshold=case["threshold"]),
|
||||
)
|
||||
|
||||
def test_benchmark_semantic_report_is_deterministic_and_passes_thresholds(self):
|
||||
first = run_benchmark(measure_latency=False)
|
||||
second = run_benchmark(measure_latency=False)
|
||||
self.assertEqual(first, second)
|
||||
self.assertTrue(first["passed"])
|
||||
self.assertEqual(first["results"]["matching"]["false_positive"], 0)
|
||||
self.assertEqual(first["results"]["contacts"]["false_positive"], 0)
|
||||
|
||||
def test_labeled_false_positives_stay_suppressed(self):
|
||||
for case in self.fixtures["matching"]:
|
||||
predicted = {item["id"] for item in match_businesses(case["source"], case["candidates"], threshold=case["threshold"])}
|
||||
self.assertTrue(predicted.isdisjoint(set(case["forbidden_ids"])), case["name"])
|
||||
|
||||
for case in self.fixtures["contacts"]:
|
||||
values = {(item["kind"], item["value"]) for item in extract_contacts(case["html"], case["source_url"])}
|
||||
self.assertTrue(values.isdisjoint({tuple(item) for item in case["forbidden"]}), case["name"])
|
||||
|
||||
def test_website_labels_and_scores_have_no_unsafe_positive(self):
|
||||
for case in self.fixtures["websites"]:
|
||||
self.assertEqual(classify_website(case["status"], case["url"], case["body"], error=case.get("error")), case["expected"], case["name"])
|
||||
for case in self.fixtures["scoring"]:
|
||||
result = evaluate_score(case["signals"], DEFAULT_RULES)
|
||||
self.assertEqual(result, evaluate_score(case["signals"], DEFAULT_RULES))
|
||||
if case.get("must_be_ineligible"):
|
||||
self.assertFalse(result["eligible"])
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
Reference in New Issue
Block a user