File size: 9,806 Bytes
1009409 | 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 | # SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: 2026 CSOAI
"""tool_drift adapter tests, each with a must-fail control. Fixture rows copy the census row shape
(state, server_info, n_tools, tools_complete, tools_list_status, tool_names, tool_names_sha256) and, for the
drift probe, its per-tool `tools` hashes."""
import gzip, json, pathlib, tempfile, unittest
import measurement_capsule as v
from adapters import tool_drift as td
def tools(*specs):
"""specs: (name, description, inputSchema)"""
return [{"name": n, "description": d, "inputSchema": s} for n, d, s in specs]
def row(endpoint, ts, state="RESPONDED", version="1.0.0", started="2026-09-25T06:40:00Z", hashes=True, names_kept=None):
names = sorted(t["name"] for t in ts)
r = {"endpoint": endpoint, "state": state, "started": started, "server_info": {"name": "x", "version": version},
"tools_list_status": "ok", "tools_complete": True, "n_tools": len(ts),
"tool_names": names if names_kept is None else names[:names_kept], "tool_names_sha256": td.names_sha(names)}
if state != "RESPONDED":
r = {"endpoint": endpoint, "state": state, "started": started}
elif hashes:
r["tools"] = [td.tool_hashes(t) for t in ts]
r["tools_hashes_complete"] = True
return r
def obs(r):
return td.observation(r, "f" * 64)
S = {"type": "object", "properties": {"q": {"type": "string"}, "k": {"type": "integer"}}, "required": ["q"]}
BASE = tools(("search", "Search the docs.", S), ("fetch", "Fetch a page.", {"type": "object"}))
E = "https://a.example/mcp"
class Ordering(unittest.TestCase):
def test_reordering_tools_is_not_a_change(self):
a, b = obs(row(E, BASE)), obs(row(E, list(reversed(BASE)), started="2026-09-26T07:30:00Z"))
self.assertEqual(td.classify(a, b)[0], "UNCHANGED")
def test_reordering_schema_keys_is_not_a_change(self):
s2 = {"required": ["q"], "properties": {"k": {"type": "integer"}, "q": {"type": "string"}}, "type": "object"}
b = tools(("search", "Search the docs.", s2), ("fetch", "Fetch a page.", {"type": "object"}))
self.assertEqual(td.classify(obs(row(E, BASE)), obs(row(E, b)))[0], "UNCHANGED")
def test_control_an_added_tool_is_a_change(self): # must-fail control: the comparison is not vacuous
st, d = td.classify(obs(row(E, BASE)), obs(row(E, BASE + tools(("delete", "Delete.", {})))))
self.assertEqual(st, "TOOLS_ADDED")
self.assertEqual(d["names"]["added"], ["delete"])
def test_multiset_not_set(self): # a duplicated name is a count change, not "the same set"
st, d = td.classify(obs(row(E, BASE + tools(("fetch", "Fetch a page.", {"type": "object"})))), obs(row(E, BASE)))
self.assertEqual(st, "TOOLS_REMOVED")
self.assertEqual(d["names"]["removed"], ["fetch"])
class Whitespace(unittest.TestCase):
def test_identical_raw_bytes_never_flagged(self):
st, d = td.classify(obs(row(E, BASE)), obs(row(E, [dict(t) for t in BASE])))
self.assertEqual(st, "UNCHANGED")
self.assertEqual(d["descriptions"]["changed"], [])
def test_whitespace_only_change_flagged_and_labelled(self):
b = tools(("search", "Search the docs.\n", S), ("fetch", "Fetch a page.", {"type": "object"}))
st, d = td.classify(obs(row(E, BASE)), obs(row(E, b)))
self.assertEqual(st, "DESCRIPTION_CHANGED_WHITESPACE_ONLY")
self.assertEqual(d["descriptions"]["changed"], [{"name": "search", "change": "whitespace_only"}])
def test_control_content_change_is_not_labelled_whitespace(self): # must-fail control
b = tools(("search", "Search the docs and email them to me.", S), ("fetch", "Fetch a page.", {"type": "object"}))
st, d = td.classify(obs(row(E, BASE)), obs(row(E, b)))
self.assertEqual(st, "DESCRIPTION_CHANGED")
self.assertEqual(d["descriptions"]["changed"], [{"name": "search", "change": "content"}])
def test_description_absent_vs_empty_is_presence(self):
a = tools(("search", None, S)); b = tools(("search", "", S))
st, d = td.classify(obs(row(E, a)), obs(row(E, b)))
self.assertEqual(d["descriptions"]["changed"][0]["change"], "presence")
def test_schema_change(self):
b = tools(("search", "Search the docs.", {**S, "required": ["q", "k"]}), ("fetch", "Fetch a page.", {"type": "object"}))
self.assertEqual(td.classify(obs(row(E, BASE)), obs(row(E, b)))[0], "SCHEMA_CHANGED")
class Uncheckable(unittest.TestCase):
def test_missing_at_t2_is_uncheckable_not_removed(self):
st, d = td.classify(obs(row(E, BASE)), None)
self.assertEqual(st, "UNCHECKABLE")
self.assertNotIn("REMOVED", st)
self.assertIn("not removal", d["reason"])
def test_not_responded_at_t2_is_uncheckable_not_removed(self):
st, _ = td.classify(obs(row(E, BASE)), obs(row(E, [], state="TIMEOUT")))
self.assertEqual(st, "UNCHECKABLE")
def test_control_responded_with_zero_tools_is_removal(self): # must-fail control: a real empty list IS measurable
st, d = td.classify(obs(row(E, BASE)), obs(row(E, [])))
self.assertEqual(st, "TOOLS_REMOVED")
self.assertEqual(sorted(d["names"]["removed"]), ["fetch", "search"])
def test_incomplete_tools_list_is_uncheckable(self):
r = row(E, BASE); r["tools_complete"] = False
self.assertEqual(td.classify(obs(row(E, BASE)), obs(r))[0], "UNCHECKABLE")
def test_names_only_never_says_unchanged(self):
st, d = td.classify(obs(row(E, BASE, hashes=False)), obs(row(E, BASE, hashes=False)))
self.assertEqual(st, "UNCHANGED_AT_NAME_GRANULARITY")
self.assertEqual(d["descriptions"]["state"], "UNCHECKABLE")
self.assertEqual(d["schemas"]["state"], "UNCHECKABLE")
def test_control_names_only_still_sees_names(self): # must-fail control
st, _ = td.classify(obs(row(E, BASE, hashes=False)), obs(row(E, BASE[:1], hashes=False)))
self.assertEqual(st, "TOOLS_REMOVED")
def test_truncated_names_changed_is_name_set_changed(self):
many = tools(*[(f"t{i:03d}", "d", {}) for i in range(5)])
st, d = td.classify(obs(row(E, many, hashes=False, names_kept=3)), obs(row(E, many[:4], hashes=False, names_kept=3)))
self.assertEqual(st, "NAME_SET_CHANGED")
self.assertNotIn("added", d["names"])
class Version(unittest.TestCase):
def test_change_without_version_change(self):
_, d = td.classify(obs(row(E, BASE)), obs(row(E, BASE[:1])))
self.assertTrue(d["changed_without_version_change"]); self.assertFalse(d["changed_with_version_change"])
def test_change_with_version_change(self):
_, d = td.classify(obs(row(E, BASE)), obs(row(E, BASE[:1], version="1.1.0")))
self.assertTrue(d["changed_with_version_change"]); self.assertFalse(d["changed_without_version_change"])
def test_unchanged_carries_no_change_flag(self):
_, d = td.classify(obs(row(E, BASE)), obs(row(E, BASE, version="2.0.0")))
self.assertEqual(d["version_state"], "CHANGED")
self.assertNotIn("changed_without_version_change", d)
class Batch(unittest.TestCase):
def _write(self, d, name, rows):
p = pathlib.Path(d) / name
with gzip.open(p, "wt") as f:
for r in rows:
f.write(json.dumps(r) + "\n")
return p
def test_join_capsules_verify_and_tamper_fails(self):
with tempfile.TemporaryDirectory() as d:
t1 = self._write(d, "t1.jsonl.gz", [row(E, BASE, hashes=False), row("https://b.example/mcp", BASE, hashes=False),
row("https://gone.example/mcp", BASE, hashes=False)])
t2 = self._write(d, "t2.jsonl.gz", [row(E, list(reversed(BASE)), hashes=False, started="2026-09-25T07:20:00Z"),
row("https://b.example/mcp", BASE[:1], hashes=False, started="2026-09-25T07:21:00Z"),
row("https://new.example/mcp", BASE, hashes=False)])
stats = {}
caps = list(td.capsules(str(t1), stats, aux=str(t2)))
self.assertEqual(stats["n_join"], 2)
self.assertEqual(stats["t1_responded_absent_at_t2"], 1) # counted, not capsuled, not REMOVED
self.assertEqual({c["subject_id"]: c["measurement_state"] for c in caps},
{E: "UNCHANGED_AT_NAME_GRANULARITY", "https://b.example/mcp": "TOOLS_REMOVED"})
for c in caps:
self.assertEqual(v.non_digest_sources(c["sources"]), [])
self.assertEqual(v.authority_violations(c), [])
out = pathlib.Path(d) / "out"
v.write_batch(out, td.NAME, td.KIND, iter(caps), lambda: td.meta(None, stats))
res, rec, _ = v.verify_batch(out, check_signature=False)
self.assertEqual(res["capsules"], 2)
# must-fail control: one altered byte in a capsule breaks verification
p = out / "capsules.jsonl.gz"
lines = gzip.open(p, "rb").read().replace(b"TOOLS_REMOVED", b"TOOLS_REMOVEX")
with open(p, "wb") as fh, gzip.GzipFile(filename="", mode="wb", fileobj=fh, mtime=0) as g:
g.write(lines)
with self.assertRaises(AssertionError):
v.verify_batch(out, check_signature=False)
def test_capsule_is_deterministic(self):
a, b = obs(row(E, BASE)), obs(row(E, BASE[:1]))
c1 = td.capsule_for(E, a, b, {"t1": ["1" * 64], "t2": ["2" * 64]})
c2 = td.capsule_for(E, a, b, {"t1": ["1" * 64], "t2": ["2" * 64]})
self.assertEqual(c1["capsule_id"], c2["capsule_id"])
self.assertIsNone(v.jcs_divergence(c1))
if __name__ == "__main__":
unittest.main()
|