Fix run-env, which failed on every clean install - #3
Merged
Conversation
`aimpoint run-env` raised a pydantic ValidationError before running a single sample. Inspect's `eval()` has no `config` parameter and forwards unknown keywords into `GenerateConfig(**kwargs)`, so passing `config=GenerateConfig(seed=...)` became `GenerateConfig(config=...)` and failed immediately. Generation settings go in as keyword arguments. The whole suite was green throughout. Every test drove `inspect_eval` directly, so nothing covered the one function the CLI calls, and a suite that reaches past its own entry point can be entirely green on a product nobody can start. tests/test_runner.py now exercises `run` the way the CLI does, checks the run card serialises, and pins the rule that a single replicate suppresses the headline. Also raises the inspect-ai floor to the version this is verified against. The keyword and `config` object forms are not compatible, so the old floor of 0.3.130 advertised support that could not have worked.
aaygan29
added a commit
that referenced
this pull request
Aug 10, 2026
`aimpoint run-env` raised a pydantic ValidationError before running a single sample. Inspect's `eval()` has no `config` parameter and forwards unknown keywords into `GenerateConfig(**kwargs)`, so passing `config=GenerateConfig(seed=...)` became `GenerateConfig(config=...)` and failed immediately. Generation settings go in as keyword arguments. The whole suite was green throughout. Every test drove `inspect_eval` directly, so nothing covered the one function the CLI calls, and a suite that reaches past its own entry point can be entirely green on a product nobody can start. tests/test_runner.py now exercises `run` the way the CLI does, checks the run card serialises, and pins the rule that a single replicate suppresses the headline. Also raises the inspect-ai floor to the version this is verified against. The keyword and `config` object forms are not compatible, so the old floor of 0.3.130 advertised support that could not have worked. Co-authored-by: Aayush Gandhi <aayushgandhi@Aayushs-MacBook-Air.local>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The bug
aimpoint run-envraised a pydanticValidationErrorbefore running a single sample, onevery clean install. It is the repository's main entry point, so a fresh clone could validate
environments and pass its tests but could not actually run an eval.
Inspect's
eval()has noconfigparameter. It forwards unknown keywords intoGenerateConfig(**kwargs), soconfig=GenerateConfig(seed=...)becameGenerateConfig(config=...)and failed immediately. Generation settings go in as keywordarguments.
Why nothing caught it
The suite was green the whole time. Every test drove
inspect_evaldirectly, so nothingcovered
run, the one function the CLI calls. A suite that reaches past its own entry pointcan be entirely green on a product nobody can start.
tests/test_runner.pynow exercisesrunthe way the CLI does, checks the run cardserialises, and pins the rule that a single replicate suppresses the headline.
Dependency floor
Raised to the version this is verified against. The keyword form and the
configobject formare not compatible, so the previous floor of
0.3.130advertised support that could not haveworked with this call.
Verification
Cloned fresh from this branch,
uv sync, then:Completes, computes all four baselines, writes a run card, and correctly suppresses the
headline with
only 1 replicate(s); headline suppressed.