You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This issue was drafted by the AI model currently assisting me, reviewed by me before submission, and created at my request.
Summary
The public SWE-bench Pro instance instance_navidrome__navidrome-97434c1789a6444b30aae5ff5aa124a96a88f504 contains a contradiction between its visible requirements, gold patch, and official TestCore test.
An agent that follows the explicit requirement to return a nil transcoding value will fail the official test. Conversely, the gold patch passes by violating that visible requirement.
It("finds player by ID and return its transcoding", func() {
plr:=&model.Player{
ID: "123",
Name: "A Player",
Client: "client",
LastSeen: time.Time{},
TranscodingId: "1",
}
repo.add(plr)
p, trc, err:=players.Register(
ctx,
"123",
"client",
"chrome",
"1.2.3.4",
)
Expect(err).ToNot(HaveOccurred())
Expect(p.ID).To(Equal("123"))
Expect(trc.ID).To(Equal("1"))
})
Returning nil, as explicitly required by the prompt, necessarily makes this test fail.
Observed behavior
This was observed during an evaluation where the agent implemented the visible requirements, including the unconditional nil transcoding return.
The implementation and its local test suite completed successfully, but the official evaluator restored core players_test.go from the gold commit and reported:
Required tests failed or were skipped:
- TestCore: FAILED
The evaluation environment was healthy:
agent exit code: 0
test exit code: 1
infrastructure failures: 0
required test present and executed: TestCore
The failure is therefore not caused by a missing tool, Docker setup, or evaluation infrastructure problem.
Impact
This instance can produce a false comparison between agents:
an agent that follows the explicit nil requirement is marked incorrect;
an agent that ignores that requirement and follows the hidden test may pass;
the gold patch passes while contradicting the visible contract.
This makes the instance unsuitable for comparing instruction-following or agent performance in its current form.
Suggested fix
The likely correction is to remove this requirement from the dataset:
Register must return a nil transcoding value.
That would align the visible task with both the gold patch and the official test.
If returning nil was actually intended, then the gold patch and core/players_test.go must instead be updated to enforce that behavior.
In either case, the visible requirements, gold patch, and verifier should express the same contract.
Provenance check
The issue was initially observed through a translated Harbor/Margin suite.
The official Hugging Face row was then inspected directly and contains the same nil requirement, gold patch, before_repo_set_cmd, and TestCore selection.
Note
This issue was drafted by the AI model currently assisting me, reviewed by me before submission, and created at my request.
Summary
The public SWE-bench Pro instance
instance_navidrome__navidrome-97434c1789a6444b30aae5ff5aa124a96a88f504contains a contradiction between its visible requirements, gold patch, and officialTestCoretest.An agent that follows the explicit requirement to return a
niltranscoding value will fail the official test. Conversely, the gold patch passes by violating that visible requirement.Affected instance
instance_navidrome__navidrome-97434c1789a6444b30aae5ff5aa124a96a88f504navidrome/navidromef8ee6db72a015f33b41879d1ba253eeff8884bfd97434c1789a6444b30aae5ff5aa124a96a88f504TestCoreConflicting contracts
requirementsfieldRegistermust return aniltranscoding valueTestCoretrc.IDto equal"1"Visible requirement
The official dataset's
requirementsfield states:Gold patch behavior
The gold patch retains the existing transcoding lookup:
It therefore does not satisfy the visible
nilrequirement when a player has a configured transcoding.Official test behavior
The dataset's
before_repo_set_cmdrestores the complete test file from the gold commit:That test file still contains:
Returning
nil, as explicitly required by the prompt, necessarily makes this test fail.Observed behavior
This was observed during an evaluation where the agent implemented the visible requirements, including the unconditional
niltranscoding return.The implementation and its local test suite completed successfully, but the official evaluator restored
core players_test.gofrom the gold commit and reported:The evaluation environment was healthy:
010TestCoreThe failure is therefore not caused by a missing tool, Docker setup, or evaluation infrastructure problem.
Impact
This instance can produce a false comparison between agents:
nilrequirement is marked incorrect;This makes the instance unsuitable for comparing instruction-following or agent performance in its current form.
Suggested fix
The likely correction is to remove this requirement from the dataset:
That would align the visible task with both the gold patch and the official test.
If returning
nilwas actually intended, then the gold patch andcore/players_test.gomust instead be updated to enforce that behavior.In either case, the visible requirements, gold patch, and verifier should express the same contract.
Provenance check
The issue was initially observed through a translated Harbor/Margin suite.
The official Hugging Face row was then inspected directly and contains the same
nilrequirement, gold patch,before_repo_set_cmd, andTestCoreselection.