Skip to content

Commit 57994df

Browse files
committed
fix(verifiers): preserve research-status diagnostics
1 parent 5a77aee commit 57994df

3 files changed

Lines changed: 26 additions & 3 deletions

File tree

benchmarks/datasets/mathematical-benchmarks-v1/research-status-evidence-audit/tests/Dockerfile

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@ FROM ghcr.io/astral-sh/uv:0.8.4-python3.12-bookworm-slim@sha256:dc7e1d08f8ca9798
22
RUN python -m pip install --no-cache-dir attrs==26.1.0 jsonschema==4.26.0 jsonschema-specifications==2025.9.1 referencing==0.37.0 rpds-py==2026.6.3 typing-extensions==4.16.0
33
# Frozen citation-support audit verifier.
44
LABEL jacobian.task="jacobian/research-status-evidence-audit" \
5-
jacobian.checksum="c14e11277b1cfd5ae167a0d67270bb121f34b2b5b5fa8e629544b9508913a08f"
5+
jacobian.checksum="bf7e7f474b6f8bd06c6883bf25554ceaabdfcdfa83a3e625b85b1003e19a5daf"
66
COPY expected.json input.json test.sh verifier.py verifier_support.py public_contract.json /tests/
77
COPY input.json /app/input.json
88
RUN chmod +x /tests/test.sh && \

benchmarks/datasets/mathematical-benchmarks-v1/research-status-evidence-audit/tests/verifier.py

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -3,7 +3,7 @@
33

44
from verifier_support import (
55
evidence_list_is_bound,
6-
load_submission,
6+
load_submission_raw,
77
resolve_evidence,
88
strict_submission_contract,
99
)
@@ -49,7 +49,7 @@ def load_submission_bounded():
4949
or path.stat().st_size > MAX_SUBMISSION_BYTES
5050
):
5151
return None
52-
return load_submission(path)
52+
return load_submission_raw(path)
5353
except (OSError, RecursionError):
5454
return None
5555

benchmarks/datasets/mathematical-benchmarks-v1/research-status-evidence-audit/tests/verifier_support.py

Lines changed: 23 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -122,6 +122,28 @@ def load_submission(
122122
)
123123

124124

125+
def load_submission_raw(
126+
path: Path = WORKSPACE / "submission.json",
127+
*,
128+
require_input_binding: bool = True,
129+
) -> dict[str, Any] | None:
130+
"""Parse one bounded submission object without public-schema validation."""
131+
132+
if require_input_binding and not workspace_input_is_bound():
133+
return None
134+
if not is_regular_bounded_file(path, max_bytes=MAX_SUBMISSION_BYTES):
135+
return None
136+
try:
137+
value = json.loads(
138+
path.read_text(),
139+
parse_constant=_reject_nonfinite_json,
140+
parse_float=_finite_json_float,
141+
)
142+
except (OSError, ValueError, RecursionError, MemoryError, TypeError):
143+
return None
144+
return value if isinstance(value, dict) else None
145+
146+
125147
def _public_submission_is_valid(submission: object) -> bool:
126148
contract = _load_public_contract()
127149
if contract is None:
@@ -414,6 +436,7 @@ def aggregate_reward(
414436
"false_verified_claim",
415437
"is_regular_bounded_file",
416438
"load_submission",
439+
"load_submission_raw",
417440
"read_evidence_json",
418441
"resolve_evidence",
419442
"sha256_uri",

0 commit comments

Comments
 (0)