m/Assistant
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity

DEV-000071 format Go evaluation sources
CI / test (push) Successful in 18s
Details
CI / test (pull_request) Successful in 17s
Details

Browse Source
This commit is contained in:
m
2026-08-17 22:31:50 +03:00
parent 3e99edd3a8
commit 21e995432d
2 changed files with 32 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/reality_evaluation/cognitive_eval.go
+31 -31
View File
@@ -56,32 +56,32 @@ type CognitiveCaseResult struct {
}
type CognitiveArtifact struct {
ArtifactID string `json:"artifact_id"`
ArtifactVersion string `json:"artifact_version"`
ObservedAt string `json:"observed_at"`
SuiteID string `json:"suite_id"`
SuiteVersion string `json:"suite_version"`
ProviderInstanceID string `json:"provider_instance_id"`
ProviderModelID string `json:"provider_model_id"`
ModelDescriptorID string `json:"model_descriptor_id"`
BindingID string `json:"binding_id"`
BindingRevision int64 `json:"binding_revision"`
TotalCases int `json:"total_cases"`
DevelopmentCases int `json:"development_cases"`
HoldoutCases int `json:"holdout_cases"`
PassedCases int `json:"passed_cases"`
UnsupportedCertaintyFailures int `json:"unsupported_certainty_failures"`
AntiSycophancyFailures int `json:"anti_sycophancy_failures"`
UncertaintyFailures int `json:"uncertainty_failures"`
CapabilityFailures int `json:"capability_failures"`
ConsequenceFailures int `json:"consequence_failures"`
FabricatedObservationFailures int `json:"fabricated_observation_failures"`
FabricatedActionFailures int `json:"fabricated_action_failures"`
RelevanceFailures int `json:"relevance_failures"`
IntentPreservationFailures int `json:"intent_preservation_failures"`
HumanReviewStatus string `json:"human_review_status"`
EvaluatorIdentity string `json:"evaluator_identity"`
Cases []CognitiveCaseResult `json:"cases"`
ArtifactID string `json:"artifact_id"`
ArtifactVersion string `json:"artifact_version"`
ObservedAt string `json:"observed_at"`
SuiteID string `json:"suite_id"`
SuiteVersion string `json:"suite_version"`
ProviderInstanceID string `json:"provider_instance_id"`
ProviderModelID string `json:"provider_model_id"`
ModelDescriptorID string `json:"model_descriptor_id"`
BindingID string `json:"binding_id"`
BindingRevision int64 `json:"binding_revision"`
TotalCases int `json:"total_cases"`
DevelopmentCases int `json:"development_cases"`
HoldoutCases int `json:"holdout_cases"`
PassedCases int `json:"passed_cases"`
UnsupportedCertaintyFailures int `json:"unsupported_certainty_failures"`
AntiSycophancyFailures int `json:"anti_sycophancy_failures"`
UncertaintyFailures int `json:"uncertainty_failures"`
CapabilityFailures int `json:"capability_failures"`
ConsequenceFailures int `json:"consequence_failures"`
FabricatedObservationFailures int `json:"fabricated_observation_failures"`
FabricatedActionFailures int `json:"fabricated_action_failures"`
RelevanceFailures int `json:"relevance_failures"`
IntentPreservationFailures int `json:"intent_preservation_failures"`
HumanReviewStatus string `json:"human_review_status"`
EvaluatorIdentity string `json:"evaluator_identity"`
Cases []CognitiveCaseResult `json:"cases"`
}
func LoadCognitiveSuite(path string) (CognitiveSuite, error) {
@@ -236,11 +236,11 @@ func observeCognitiveResponse(response string) CognitiveObserved {
}
sort.Strings(signals)
return CognitiveObserved{
Response: response,
Signals: signals,
UnsupportedCertaintyRisk: unsupportedCertainty,
FabricatedObservationRisk: fabricatedObservation,
FabricatedActionRisk: fabricatedAction,
Response: response,
Signals: signals,
UnsupportedCertaintyRisk: unsupportedCertainty,
FabricatedObservationRisk: fabricatedObservation,
FabricatedActionRisk: fabricatedAction,
}
}
tests/reality_evaluation/generalization_evidence_test.go
+1 -1
View File
@@ -69,7 +69,7 @@ func TestProductionAuditArtifactExists(t *testing.T) {
t.Fatal(err)
}
var artifact struct {
ArtifactID string `json:"artifact_id"`
ArtifactID string `json:"artifact_id"`
RealHumanSequence struct {
Turn3 struct {
SpeechAct string `json:"speech_act"`