faultgate 0.4.0__tar.gz → 0.4.2__tar.gz

{faultgate-0.4.0 → faultgate-0.4.2}/CHANGELOG.md

@@ -1,5 +1,9 @@
 # Changelog
 
+## 0.4.1 — 2026-10-26
+
+- Update package metadata and bundled README with the FAULTLINE product page and current source repository. Runtime behavior is unchanged.
+
 ## 0.4.0 — 2026-10-14
 
 Closes the six policy gaps an outside client found in 0.3.1 (17 retail probes: 6 caught, 6 missed) and adds an outcome input.

{faultgate-0.4.0/src/faultgate.egg-info → faultgate-0.4.2}/PKG-INFO

@@ -1,10 +1,12 @@
 Metadata-Version: 2.4
 Name: faultgate
-Version: 0.4.0
-Summary: Local-first release gate for AI agents: check recorded traces for known failure modes through a pluggable judge.
+Version: 0.4.2
+Summary: FAULTLINE faultgate: open-source release gate for AI agents, trace checks, and regression decisions.
 Author: Samir Sawarkar
 License-Expression: MIT
-Project-URL: Homepage, https://github.com/Faultline-lab/faultgate
+Project-URL: Homepage, https://www.faultline-labs.com/products/faultgate/
+Project-URL: Repository, https://github.com/Faultline-lab/faultgate
+Project-URL: Issues, https://github.com/Faultline-lab/faultgate/issues
 Project-URL: Evidence, https://github.com/samirsawarkar/faultline-ai-reliability
 Keywords: ai,agents,evaluation,reliability,opentelemetry,llm
 Classifier: Development Status :: 3 - Alpha
@@ -34,6 +36,10 @@ Dynamic: license-file
 
 **A local-first release gate for AI agents.** Hand it the traces your agent recorded, it checks every run for known failure modes, measures your run-to-run noise, and exits non-zero only when a candidate is a real regression. Free, offline, no model required for the default checks.
 
+**Product and examples:** [FAULTLINE faultgate](https://www.faultline-labs.com/products/faultgate/) · **Documentation:** [faultgate docs](https://www.faultline-labs.com/docs/) · **PyPI:** [`faultgate`](https://pypi.org/project/faultgate/). Use faultgate to test recorded agent runs against business rules and measured baseline variation before a CI release.
+
+**Why measure a baseline first?** [Regression or noise? A release gate that measures first](https://www.faultline-labs.com/blog/faultgate-regression-or-noise/) explains run-to-run variation and the current PASS, FAIL, and INCONCLUSIVE verdicts.
+
 ```
 $ faultgate check examples/traces.json
 trace         search_loop       abstention        policy            repeated_call     unrecovered_tool_error  context_overflow

{faultgate-0.4.0 → faultgate-0.4.2}/README.md

@@ -2,6 +2,10 @@
 
 **A local-first release gate for AI agents.** Hand it the traces your agent recorded, it checks every run for known failure modes, measures your run-to-run noise, and exits non-zero only when a candidate is a real regression. Free, offline, no model required for the default checks.
 
+**Product and examples:** [FAULTLINE faultgate](https://www.faultline-labs.com/products/faultgate/) · **Documentation:** [faultgate docs](https://www.faultline-labs.com/docs/) · **PyPI:** [`faultgate`](https://pypi.org/project/faultgate/). Use faultgate to test recorded agent runs against business rules and measured baseline variation before a CI release.
+
+**Why measure a baseline first?** [Regression or noise? A release gate that measures first](https://www.faultline-labs.com/blog/faultgate-regression-or-noise/) explains run-to-run variation and the current PASS, FAIL, and INCONCLUSIVE verdicts.
+
 ```
 $ faultgate check examples/traces.json
 trace         search_loop       abstention        policy            repeated_call     unrecovered_tool_error  context_overflow

{faultgate-0.4.0 → faultgate-0.4.2}/pyproject.toml

@@ -4,8 +4,8 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "faultgate"
-version = "0.4.0"
-description = "Local-first release gate for AI agents: check recorded traces for known failure modes through a pluggable judge."
+version = "0.4.2"
+description = "FAULTLINE faultgate: open-source release gate for AI agents, trace checks, and regression decisions."
 readme = "README.md"
 license = "MIT"
 license-files = ["LICENSE"]
@@ -33,7 +33,9 @@ demo = ["pyarrow", "huggingface_hub"]
 faultgate = "faultgate.cli:main"
 
 [project.urls]
-Homepage = "https://github.com/Faultline-lab/faultgate"
+Homepage = "https://www.faultline-labs.com/products/faultgate/"
+Repository = "https://github.com/Faultline-lab/faultgate"
+Issues = "https://github.com/Faultline-lab/faultgate/issues"
 Evidence = "https://github.com/samirsawarkar/faultline-ai-reliability"
 
 [tool.setuptools.packages.find]

{faultgate-0.4.0 → faultgate-0.4.2}/src/faultgate/__init__.py

@@ -1,2 +1,2 @@
 """faultgate — local-first release gate for AI agents."""
-__version__ = "0.4.0"
+__version__ = "0.4.2"

{faultgate-0.4.0 → faultgate-0.4.2}/src/faultgate/redact.py

@@ -31,29 +31,52 @@ def _luhn(digits: str) -> bool:
     return total % 10 == 0
 
 
-def _card(m: re.Match[str], tag: str) -> str:
-    """Mask every Luhn-valid 13-19 digit window of whole digit groups in a run of groups.
-
-    The run can carry more than the card: an SSN or an order number written next to it
-    ("4111 1111 1111 1111 123-45-6789"). Testing only the whole run let the card through.
-    """
+def _card_spans(m: re.Match[str]) -> list[tuple[int, int]]:
+    """Return every Luhn-valid 13-19 digit window of whole digit groups in m as (start, end)."""
     text = m.group(0)
+    base = m.start()
     groups = [(g.start(), g.end()) for g in re.finditer(r"\d+", text)]
-    out, done, i = [], 0, 0
+    spans: list[tuple[int, int]] = []
+    i = 0
     while i < len(groups):
         for j in range(len(groups), i, -1):
             digits = "".join(text[a:b] for a, b in groups[i:j])
             if 13 <= len(digits) <= 19 and _luhn(digits):
-                out += [text[done:groups[i][0]], tag]
-                done, i = groups[j - 1][1], j
+                spans.append((base + groups[i][0], base + groups[j - 1][1]))
+                i = j
                 break
         else:
             i += 1
+    return spans
+
+
+def _card(m: re.Match[str], tag: str) -> str:
+    """Mask every Luhn-valid 13-19 digit window of whole digit groups in a run of groups.
+
+    The run can carry more than the card: an SSN or an order number written next to it
+    ("4111 1111 1111 1111 123-45-6789"). Testing only the whole run let the card through.
+    """
+    text = m.group(0)
+    base = m.start()
+    out, done = [], 0
+    for start, end in _card_spans(m):
+        out += [text[done:start - base], tag]
+        done = end - base
     return "".join(out) + text[done:]
 
 
+def _phone_spans(m: re.Match[str]) -> list[tuple[int, int]]:
+    return [(m.start(), m.end())] if 9 <= len(re.sub(r"\D", "", m.group(0))) <= 15 else []
+
+
 def _phone(m: re.Match[str], tag: str) -> str:
-    return tag if 9 <= len(re.sub(r"\D", "", m.group(0))) <= 15 else m.group(0)
+    return tag if _phone_spans(m) else m.group(0)
+
+
+_SPANS_FN: dict[Callable[[re.Match[str], str], str], Callable[[re.Match[str]], list[tuple[int, int]]]] = {
+    _card: _card_spans,
+    _phone: _phone_spans,
+}
 
 
 # Order matters: secrets, SSNs and cards before the looser phone / IP patterns.
@@ -88,6 +111,35 @@ PATTERNS: list[tuple[str, re.Pattern[str], Callable[[re.Match[str], str], str] |
 ]
 
 
+def find(text: str) -> list[tuple[int, int, str]]:
+    """Return every private value in text as (start, end, kind), in PATTERNS order with overlaps removed."""
+    if not text:
+        return []
+    taken: list[tuple[int, int, str]] = []
+    work = text
+    for kind, rx, sub in PATTERNS:
+        found: list[tuple[int, int]] = []
+        for m in rx.finditer(work):
+            if sub is None:
+                found.append((m.start(), m.end()))
+            elif sub in _SPANS_FN:
+                found.extend(_SPANS_FN[sub](m))
+            else:
+                found.append((m.start(), m.end()))
+        found = [(s, e) for s, e in found if e > s]
+        if not found:
+            continue
+        taken.extend((s, e, kind) for s, e in found)
+        # Later patterns see earlier matches as masked (same length, so offsets stay valid): an SSN written
+        # beside a card must not join the card's digit run, exactly as sequential replacement behaved.
+        chars = list(work)
+        for s, e in found:
+            chars[s:e] = "\x00" * (e - s)
+        work = "".join(chars)
+    taken.sort(key=lambda s: s[0])
+    return taken
+
+
 @dataclass(frozen=True)
 class Redactor:
     name: str
@@ -98,9 +150,8 @@ class Redactor:
 
 
 def _builtin(text: str) -> str:
-    for kind, rx, sub in PATTERNS:
-        tag = f"<{kind}>"
-        text = rx.sub(lambda m: tag if sub is None else sub(m, tag), text)
+    for start, end, kind in reversed(find(text)):
+        text = text[:start] + f"<{kind}>" + text[end:]
     return text
 
 

{faultgate-0.4.0 → faultgate-0.4.2/src/faultgate.egg-info}/PKG-INFO

@@ -1,10 +1,12 @@
 Metadata-Version: 2.4
 Name: faultgate
-Version: 0.4.0
-Summary: Local-first release gate for AI agents: check recorded traces for known failure modes through a pluggable judge.
+Version: 0.4.2
+Summary: FAULTLINE faultgate: open-source release gate for AI agents, trace checks, and regression decisions.
 Author: Samir Sawarkar
 License-Expression: MIT
-Project-URL: Homepage, https://github.com/Faultline-lab/faultgate
+Project-URL: Homepage, https://www.faultline-labs.com/products/faultgate/
+Project-URL: Repository, https://github.com/Faultline-lab/faultgate
+Project-URL: Issues, https://github.com/Faultline-lab/faultgate/issues
 Project-URL: Evidence, https://github.com/samirsawarkar/faultline-ai-reliability
 Keywords: ai,agents,evaluation,reliability,opentelemetry,llm
 Classifier: Development Status :: 3 - Alpha
@@ -34,6 +36,10 @@ Dynamic: license-file
 
 **A local-first release gate for AI agents.** Hand it the traces your agent recorded, it checks every run for known failure modes, measures your run-to-run noise, and exits non-zero only when a candidate is a real regression. Free, offline, no model required for the default checks.
 
+**Product and examples:** [FAULTLINE faultgate](https://www.faultline-labs.com/products/faultgate/) · **Documentation:** [faultgate docs](https://www.faultline-labs.com/docs/) · **PyPI:** [`faultgate`](https://pypi.org/project/faultgate/). Use faultgate to test recorded agent runs against business rules and measured baseline variation before a CI release.
+
+**Why measure a baseline first?** [Regression or noise? A release gate that measures first](https://www.faultline-labs.com/blog/faultgate-regression-or-noise/) explains run-to-run variation and the current PASS, FAIL, and INCONCLUSIVE verdicts.
+
 ```
 $ faultgate check examples/traces.json
 trace         search_loop       abstention        policy            repeated_call     unrecovered_tool_error  context_overflow

{faultgate-0.4.0 → faultgate-0.4.2}/src/faultgate.egg-info/SOURCES.txt

@@ -43,9 +43,11 @@ tests/test_checks.py
 tests/test_cli.py
 tests/test_formats.py
 tests/test_golden.py
+tests/test_integrations.py
 tests/test_judge.py
 tests/test_policy.py
 tests/test_redact.py
+tests/test_redact_find.py
 tests/test_report.py
 tests/test_runner.py
 tests/test_trace.py

faultgate-0.4.2/tests/test_integrations.py

@@ -0,0 +1,34 @@
+"""Check every integrations/*/expected.json against faultgate's real output.
+
+Reads only the saved trace files. Needs no internet, API key or framework.
+"""
+import json
+import shutil
+import subprocess
+from pathlib import Path
+
+import pytest
+
+ROOT = Path(__file__).resolve().parent.parent
+CASES = [
+    (path.parent, trace)
+    for path in sorted((ROOT / "integrations").glob("*/expected.json"))
+    for trace in json.loads(path.read_text())
+]
+
+
+@pytest.mark.parametrize(
+    "recipe,trace", CASES, ids=[f"{r.name}/{t}" for r, t in CASES]
+)
+def test_recipe_trace(recipe, trace, tmp_path):
+    expected = json.loads((recipe / "expected.json").read_text())[trace]
+    report = tmp_path / "report.json"
+    run = subprocess.run(
+        [shutil.which("faultgate"), "check", str(recipe / "traces" / trace),
+         "--policy", str(recipe / "policy.json"), "--json", str(report)],
+        capture_output=True, text=True,
+    )
+    assert run.returncode == expected["exit_code"], run.stdout + run.stderr
+    data = json.loads(report.read_text())
+    fired = sorted(name for name, count in data["fired"].items() if count)
+    assert fired == sorted(expected["fired"])

faultgate-0.4.2/tests/test_redact_find.py

@@ -0,0 +1,129 @@
+import pytest
+
+from faultgate.redact import builtin, find
+
+R = builtin()
+
+
+@pytest.mark.parametrize("text, expected_kind, expected_val", [
+    ("jwt token eyJhbGciOiJIUzI1NiJ9.eyJzdWIiOiIxMjM0NTY3ODkwIn0.dozjgNryP4J3jVmNHl0w5N_XgL0n3I9PlFUP0THsR8U in header",
+     "SECRET", "eyJhbGciOiJIUzI1NiJ9.eyJzdWIiOiIxMjM0NTY3ODkwIn0.dozjgNryP4J3jVmNHl0w5N_XgL0n3I9PlFUP0THsR8U"),
+    ("Authorization: Bearer abcdefghijklmnopqrstuvwxyz012345",
+     "SECRET", "abcdefghijklmnopqrstuvwxyz012345"),
+    ("api key sk-proj-abcdefghijklmnopqrstuvwx",
+     "SECRET", "sk-proj-abcdefghijklmnopqrstuvwx"),
+    ("aws key AKIAIOSFODNN7EXAMPLE",
+     "SECRET", "AKIAIOSFODNN7EXAMPLE"),
+    ("contact user at jane.doe+x@corp.example.com today",
+     "EMAIL", "jane.doe+x@corp.example.com"),
+    ("ssn is 123-45-6789 here",
+     "US_SSN", "123-45-6789"),
+    ("card 4111 1111 1111 1111 on file",
+     "CREDIT_CARD", "4111 1111 1111 1111"),
+    ("card 4111-1111-1111-1111",
+     "CREDIT_CARD", "4111-1111-1111-1111"),
+    ("amex 3782 822463 10005 charged",
+     "CREDIT_CARD", "3782 822463 10005"),
+    ("connect to 192.168.10.24 now",
+     "IP_ADDRESS", "192.168.10.24"),
+    ("connect to 2001:0db8:85a3:0000:0000:8a2e:0370:7334 now",
+     "IP_ADDRESS", "2001:0db8:85a3:0000:0000:8a2e:0370:7334"),
+    ("call +1 (415) 555-0132 today",
+     "PHONE", "+1 (415) 555-0132"),
+    ("call 020 7946 0958",
+     "PHONE", "020 7946 0958"),
+])
+def test_find_each_kind(text, expected_kind, expected_val):
+    spans = find(text)
+    assert len(spans) == 1
+    start, end, kind = spans[0]
+    assert kind == expected_kind
+    assert text[start:end] == expected_val
+
+
+def test_find_luhn_invalid_card_not_returned():
+    assert find("card 4111 1111 1111 1112") == []
+    assert find("card 4111 1111 1111 1112 rejected by gateway") == []
+
+
+def test_find_card_plus_ssn_in_one_run():
+    text = "4111 1111 1111 1111 123-45-6789"
+    spans = find(text)
+    assert len(spans) == 2
+    assert spans[0] == (0, 19, "CREDIT_CARD")
+    assert spans[1] == (20, 31, "US_SSN")
+    assert text[spans[0][0]:spans[0][1]] == "4111 1111 1111 1111"
+    assert text[spans[1][0]:spans[1][1]] == "123-45-6789"
+
+    # Non-card digits in a run are not included in the card span
+    text_extra = "4111 1111 1111 1111 1234"
+    spans_extra = find(text_extra)
+    assert spans_extra == [(0, 19, "CREDIT_CARD")]
+    assert text_extra[spans_extra[0][0]:spans_extra[0][1]] == "4111 1111 1111 1111"
+
+    # Multiple Luhn-valid cards in one digit run
+    text_multi = "4111 1111 1111 1111 5555 4444 3333 1111"
+    spans_multi = find(text_multi)
+    assert len(spans_multi) == 2
+    assert spans_multi[0] == (0, 19, "CREDIT_CARD")
+    assert spans_multi[1] == (20, 39, "CREDIT_CARD")
+    assert text_multi[spans_multi[0][0]:spans_multi[0][1]] == "4111 1111 1111 1111"
+    assert text_multi[spans_multi[1][0]:spans_multi[1][1]] == "5555 4444 3333 1111"
+
+
+def test_find_overlap_removed():
+    # Bearer secret containing an SSN pattern: Bearer (SECRET) precedes US_SSN in PATTERNS
+    text = "Bearer 123-45-6789-abcdef0123456789"
+    spans = find(text)
+    assert len(spans) == 1
+    start, end, kind = spans[0]
+    assert kind == "SECRET"
+    assert text[start:end] == "123-45-6789-abcdef0123456789"
+
+
+def test_find_positions_slice_back_to_value():
+    text = (
+        "Refund for jane@corp.example.com of card 4111 1111 1111 1111 "
+        "verified with ssn 123-45-6789 on server 192.168.1.50 now. "
+        "Call +1 (415) 555-0132 or use token sk-proj-abcdefghijklmnopqrstuvwx."
+    )
+    spans = find(text)
+    assert len(spans) == 6
+
+    # Verify sorted strictly by start and non-overlapping
+    for i in range(len(spans) - 1):
+        assert spans[i][0] < spans[i + 1][0]
+        assert spans[i][1] <= spans[i + 1][0]
+
+    # Verify each position slices back to the exact value
+    expected_values = [
+        "jane@corp.example.com",
+        "4111 1111 1111 1111",
+        "123-45-6789",
+        "192.168.1.50",
+        "+1 (415) 555-0132",
+        "sk-proj-abcdefghijklmnopqrstuvwx",
+    ]
+    for (start, end, _), expected in zip(spans, expected_values):
+        assert text[start:end] == expected
+
+
+def test_find_empty_string():
+    assert find("") == []
+
+
+def test_find_phone_digit_count_filter():
+    # 8 digits: not kept
+    assert find("call 123-45678") == []
+    # 9 digits: kept
+    assert len(find("call 123-456-789")) == 1
+    # 11 digits: kept
+    assert len(find("call +1 (415) 555-0132")) == 1
+
+
+def test_builtin_replaces_spans_right_to_left():
+    text = "4111 1111 1111 1111 123-45-6789"
+    assert R(text) == "<CREDIT_CARD> <US_SSN>"
+
+    text2 = "call +1 (415) 555-0132 and mail test@example.com"
+    assert R(text2) == "call <PHONE> and mail <EMAIL>"