harness

type ActionDescriptor struct {
	ID            string                 `json:"id"`
	Label         string                 `json:"label"`
	Method        string                 `json:"method"`
	Path          string                 `json:"path"`
	Variant       string                 `json:"variant,omitempty"`
	RequiresInput bool                   `json:"requires_input,omitempty"`
	Input         *ActionInputDescriptor `json:"input,omitempty"`
}

type ActionDriver interface {
	PerformAction(ctx context.Context, run *Run, action string, input map[string]interface{}) (*Run, error)
}

type ActionInputDescriptor struct {
	Title       string                       `json:"title,omitempty"`
	Description string                       `json:"description,omitempty"`
	SubmitLabel string                       `json:"submit_label,omitempty"`
	Fields      []ActionInputFieldDescriptor `json:"fields,omitempty"`
}

type ActionInputFieldDescriptor struct {
	Key         string   `json:"key"`
	Label       string   `json:"label"`
	Kind        string   `json:"kind,omitempty"`
	Target      string   `json:"target,omitempty"`
	PayloadKey  string   `json:"payload_key,omitempty"`
	Required    bool     `json:"required,omitempty"`
	Placeholder string   `json:"placeholder,omitempty"`
	Options     []string `json:"options,omitempty"`
}

type AgentCompatHandler struct {
	// contains filtered or unexported fields
}

type ApprovalMode string

type ArtifactRef struct {
	ID           string `json:"id"`
	RunID        string `json:"run_id"`
	Kind         string `json:"kind"`
	Label        string `json:"label,omitempty"`
	PathOrURL    string `json:"path_or_url,omitempty"`
	MIMEType     string `json:"mime_type,omitempty"`
	SizeBytes    int64  `json:"size_bytes,omitempty"`
	MetadataJSON string `json:"metadata_json,omitempty"`
}

type Baseline struct {
	ID          string                 `json:"id"`
	Name        string                 `json:"name"`
	Subject     string                 `json:"subject,omitempty"`
	OwnerUserID string                 `json:"owner_user_id,omitempty"`
	EvalSpecID  string                 `json:"eval_spec_id"`
	EvalRunID   string                 `json:"eval_run_id"`
	IsDefault   bool                   `json:"is_default"`
	Metadata    map[string]interface{} `json:"metadata,omitempty"`
	CreatedAt   time.Time              `json:"created_at"`
	UpdatedAt   time.Time              `json:"updated_at"`
}

type BaselineFilter struct {
	OwnerUserID string
	EvalSpecID  string
	Limit       int
}

type BaselineSpec struct {
	Name        string                 `json:"name"`
	Subject     string                 `json:"subject,omitempty"`
	OwnerUserID string                 `json:"owner_user_id,omitempty"`
	EvalSpecID  string                 `json:"eval_spec_id,omitempty"`
	EvalRunID   string                 `json:"eval_run_id"`
	IsDefault   bool                   `json:"is_default,omitempty"`
	Metadata    map[string]interface{} `json:"metadata,omitempty"`
}

type Batch1ExecutionAssets struct {
	Dataset        *Dataset        `json:"dataset,omitempty"`
	DatasetVersion *DatasetVersion `json:"dataset_version,omitempty"`
	EvalSpec       *EvalSpec       `json:"eval_spec,omitempty"`
}

type BatchTrajectoryCase struct {
	Key      string                 `json:"key,omitempty"`
	Prompt   string                 `json:"prompt,omitempty"`
	Input    map[string]interface{} `json:"input,omitempty"`
	Metadata map[string]interface{} `json:"metadata,omitempty"`
}

type BatchTrajectoryCaseResult struct {
	Input        map[string]interface{} `json:"input,omitempty"`
	Expected     map[string]interface{} `json:"expected,omitempty"`
	Metadata     map[string]interface{} `json:"metadata,omitempty"`
	TraceSummary string                 `json:"trace_summary,omitempty"`
	Provenance   map[string]interface{} `json:"provenance,omitempty"`
}

type BatchTrajectoryExecutor interface {
	ExecuteBatchTrajectoryCase(ctx context.Context, c BatchTrajectoryCase) (*BatchTrajectoryCaseResult, error)
}

type BatchTrajectoryRunResult struct {
	Manifest           DatasetManifest    `json:"manifest"`
	DatasetVersionSpec DatasetVersionSpec `json:"dataset_version_spec"`
	CompletedCaseKeys  []string           `json:"completed_case_keys,omitempty"`
	FailedCaseKeys     []string           `json:"failed_case_keys,omitempty"`
	CheckpointPath     string             `json:"checkpoint_path,omitempty"`
}

type BatchTrajectoryRunner struct {
	// contains filtered or unexported fields
}

type BatchTrajectoryRunnerConfig struct {
	DatasetName    string                 `json:"dataset_name"`
	DatasetSubject string                 `json:"dataset_subject"`
	DatasetVersion string                 `json:"dataset_version"`
	SourceType     string                 `json:"source_type,omitempty"`
	SourceRef      string                 `json:"source_ref,omitempty"`
	CreatedBy      string                 `json:"created_by,omitempty"`
	CandidateID    string                 `json:"candidate_id,omitempty"`
	CheckpointPath string                 `json:"checkpoint_path"`
	DefaultRunKind RunKind                `json:"default_run_kind,omitempty"`
	DefaultProfile string                 `json:"default_profile,omitempty"`
	Scheduler      GroupSchedulerConfig   `json:"scheduler,omitempty"`
	Scoring        GroupScoringConfig     `json:"scoring,omitempty"`
	RuntimePolicy  map[string]interface{} `json:"runtime_policy,omitempty"`
	MaxRetries     int                    `json:"max_retries,omitempty"`
}

type CUABenchmarkCaseResult struct {
	ID                 string `json:"id,omitempty"`
	Passed             bool   `json:"passed"`
	Critical           bool   `json:"critical,omitempty"`
	VerificationBacked bool   `json:"verification_backed,omitempty"`
	FailureLabel       string `json:"failure_label,omitempty"`
	LatencyMS          int    `json:"latency_ms,omitempty"`
}

type CUABenchmarkReport struct {
	CaseCount                     int            `json:"case_count"`
	PassedCount                   int            `json:"passed_count"`
	CriticalCount                 int            `json:"critical_count"`
	CriticalPassedCount           int            `json:"critical_passed_count"`
	PassRate                      float64        `json:"pass_rate"`
	CriticalPassRate              float64        `json:"critical_pass_rate"`
	UnsafeSendCount               int            `json:"unsafe_send_count"`
	TypedBodyIntoSearchFieldCount int            `json:"typed_body_into_search_field_count"`
	VerificationBackedRate        float64        `json:"verification_backed_rate"`
	LatencyP50MS                  int            `json:"latency_p50_ms"`
	LatencyP90MS                  int            `json:"latency_p90_ms"`
	FailureLabelCounts            map[string]int `json:"failure_label_counts,omitempty"`
}

type CUAOSWorldMacBenchmarkAssets struct {
	Dataset        *Dataset        `json:"dataset,omitempty"`
	DatasetVersion *DatasetVersion `json:"dataset_version,omitempty"`
	EvalSpec       *EvalSpec       `json:"eval_spec,omitempty"`
}

type CanonicalSkillSourceState struct {
	AbsolutePath   string
	NormalizedPath string
	Content        string
	ContentSHA256  string
}

type CheckpointArtifact struct {
	RunID       string                 `json:"run_id"`
	GroupItemID string                 `json:"group_item_id,omitempty"`
	Artifact    ArtifactRef            `json:"artifact"`
	Payload     map[string]interface{} `json:"payload,omitempty"`
}

type CompareEvalRunRequest struct {
	BaseEvalRunID string `json:"base_eval_run_id,omitempty"`
	BaselineID    string `json:"baseline_id,omitempty"`
}

type ComparisonCaseDelta struct {
	Key                 string  `json:"key"`
	Label               string  `json:"label,omitempty"`
	ItemIndex           int     `json:"item_index"`
	Profile             string  `json:"profile,omitempty"`
	BaseVerdict         string  `json:"base_verdict,omitempty"`
	TargetVerdict       string  `json:"target_verdict,omitempty"`
	BaseStatus          string  `json:"base_status,omitempty"`
	TargetStatus        string  `json:"target_status,omitempty"`
	BaseScore           float64 `json:"base_score,omitempty"`
	TargetScore         float64 `json:"target_score,omitempty"`
	DeltaScore          float64 `json:"delta_score,omitempty"`
	BaseRunID           string  `json:"base_run_id,omitempty"`
	TargetRunID         string  `json:"target_run_id,omitempty"`
	BaseReason          string  `json:"base_reason,omitempty"`
	TargetReason        string  `json:"target_reason,omitempty"`
	BaseFailureLabel    string  `json:"base_failure_label,omitempty"`
	TargetFailureLabel  string  `json:"target_failure_label,omitempty"`
	BaseVerification    string  `json:"base_verification,omitempty"`
	TargetVerification  string  `json:"target_verification,omitempty"`
	BaseEvidenceScore   float64 `json:"base_evidence_score,omitempty"`
	TargetEvidenceScore float64 `json:"target_evidence_score,omitempty"`
}

type ComparisonReport struct {
	ID              string                 `json:"id"`
	OwnerUserID     string                 `json:"owner_user_id,omitempty"`
	BaselineID      string                 `json:"baseline_id,omitempty"`
	EvalSpecID      string                 `json:"eval_spec_id"`
	BaseEvalRunID   string                 `json:"base_eval_run_id"`
	TargetEvalRunID string                 `json:"target_eval_run_id"`
	Summary         map[string]interface{} `json:"summary,omitempty"`
	Regressions     []ComparisonCaseDelta  `json:"regressions,omitempty"`
	Improvements    []ComparisonCaseDelta  `json:"improvements,omitempty"`
	ScorerDelta     map[string]interface{} `json:"scorer_delta,omitempty"`
	CreatedAt       time.Time              `json:"created_at"`
}

type Controller struct {
	// contains filtered or unexported fields
}

type Dataset struct {
	ID              string                 `json:"id"`
	Name            string                 `json:"name"`
	Description     string                 `json:"description,omitempty"`
	OwnerUserID     string                 `json:"owner_user_id,omitempty"`
	Subject         string                 `json:"subject,omitempty"`
	DefaultRunKind  RunKind                `json:"default_run_kind,omitempty"`
	DefaultProfile  string                 `json:"default_profile,omitempty"`
	ActiveVersionID string                 `json:"active_version_id,omitempty"`
	Metadata        map[string]interface{} `json:"metadata,omitempty"`
	CreatedAt       time.Time              `json:"created_at"`
	UpdatedAt       time.Time              `json:"updated_at"`
}

type DatasetBundleSourcePreview struct {
	SourceType string                        `json:"source_type,omitempty"`
	SourceRef  string                        `json:"source_ref,omitempty"`
	Dataset    DatasetSpec                   `json:"dataset"`
	Version    DatasetBundleVersionPreview   `json:"version"`
	EvalSpecs  []ImportDatasetBundleEvalSpec `json:"eval_specs,omitempty"`
	MakeActive bool                          `json:"make_active,omitempty"`
}

type DatasetBundleVersionPreview struct {
	Version        string `json:"version,omitempty"`
	ItemCount      int    `json:"item_count"`
	ManifestSHA256 string `json:"manifest_sha256,omitempty"`
	SourceType     string `json:"source_type,omitempty"`
	SourceRef      string `json:"source_ref,omitempty"`
}

type DatasetFilter struct {
	OwnerUserID string
	Limit       int
}

type DatasetManifest struct {
	Dataset  DatasetManifestMeta     `json:"dataset,omitempty"`
	Defaults DatasetManifestDefaults `json:"defaults,omitempty"`
	Items    []DatasetManifestItem   `json:"items,omitempty"`
}

type DatasetManifestDefaults struct {
	RunKind       RunKind                `json:"run_kind,omitempty"`
	Profile       string                 `json:"profile,omitempty"`
	Scheduler     GroupSchedulerConfig   `json:"scheduler,omitempty"`
	Scoring       GroupScoringConfig     `json:"scoring,omitempty"`
	RuntimePolicy map[string]interface{} `json:"runtime_policy,omitempty"`
}

type DatasetManifestItem struct {
	ID       string                 `json:"id,omitempty"`
	RunKind  RunKind                `json:"run_kind,omitempty"`
	Profile  string                 `json:"profile,omitempty"`
	Input    map[string]interface{} `json:"input,omitempty"`
	Expected map[string]interface{} `json:"expected,omitempty"`
	Metadata map[string]interface{} `json:"metadata,omitempty"`
}

type DatasetManifestMeta struct {
	Name    string `json:"name,omitempty"`
	Subject string `json:"subject,omitempty"`
}

type DatasetSpec struct {
	Name           string                 `json:"name"`
	Description    string                 `json:"description,omitempty"`
	OwnerUserID    string                 `json:"owner_user_id,omitempty"`
	Subject        string                 `json:"subject,omitempty"`
	DefaultRunKind RunKind                `json:"default_run_kind,omitempty"`
	DefaultProfile string                 `json:"default_profile,omitempty"`
	Metadata       map[string]interface{} `json:"metadata,omitempty"`
}

type DatasetVersion struct {
	ID             string                 `json:"id"`
	DatasetID      string                 `json:"dataset_id"`
	Version        string                 `json:"version"`
	ManifestSHA256 string                 `json:"manifest_sha256,omitempty"`
	ItemCount      int                    `json:"item_count"`
	SourceType     string                 `json:"source_type,omitempty"`
	SourceRef      string                 `json:"source_ref,omitempty"`
	Manifest       map[string]interface{} `json:"manifest,omitempty"`
	Metadata       map[string]interface{} `json:"metadata,omitempty"`
	CreatedBy      string                 `json:"created_by,omitempty"`
	CreatedAt      time.Time              `json:"created_at"`
}

type DatasetVersionSpec struct {
	Version    string                 `json:"version,omitempty"`
	SourceType string                 `json:"source_type,omitempty"`
	SourceRef  string                 `json:"source_ref,omitempty"`
	Manifest   map[string]interface{} `json:"manifest"`
	Metadata   map[string]interface{} `json:"metadata,omitempty"`
	CreatedBy  string                 `json:"created_by,omitempty"`
}

type Defaults struct {
	ArtifactRoot      string
	StorePath         string
	ApprovalMode      ApprovalMode
	SandboxMode       string
	MaxDuration       time.Duration
	MaxSteps          int
	MaxToolRounds     int
	MaxSubagents      int
	MaxDepth          int
	ProtectedPrefixes []string
}

type DesktopChatSuccessRateAssets struct {
	Dataset        *Dataset        `json:"dataset,omitempty"`
	DatasetVersion *DatasetVersion `json:"dataset_version,omitempty"`
	EvalSpec       *EvalSpec       `json:"eval_spec,omitempty"`
}

type DesktopChatSuccessRateGateCheck struct {
	Name     string      `json:"name"`
	Passed   bool        `json:"passed"`
	Actual   interface{} `json:"actual,omitempty"`
	Expected interface{} `json:"expected,omitempty"`
}

type DesktopChatSuccessRateGateReport struct {
	CaseCount                       int                               `json:"case_count"`
	PassedCount                     int                               `json:"passed_count"`
	PassRate                        float64                           `json:"pass_rate"`
	UnsafeSendCount                 int                               `json:"unsafe_send_count"`
	TypedBodyIntoSearchFieldCount   int                               `json:"typed_body_into_search_field_count"`
	ExistingFocusedRegressionPassed bool                              `json:"existing_focused_regression_passed"`
	FailureLabelCounts              map[string]int                    `json:"failure_label_counts,omitempty"`
	Checks                          []DesktopChatSuccessRateGateCheck `json:"checks,omitempty"`
	Passed                          bool                              `json:"passed"`
}

type Driver interface {
	Kind() RunKind
	Validate(spec RunSpec) error
	Start(ctx context.Context, run *Run, env RunEnv) error
	Cancel(ctx context.Context, run *Run) error
}

type EvalRun struct {
	ID                string                 `json:"id"`
	EvalSpecID        string                 `json:"eval_spec_id"`
	GroupID           string                 `json:"group_id"`
	DatasetVersionID  string                 `json:"dataset_version_id,omitempty"`
	BaselineEvalRunID string                 `json:"baseline_eval_run_id,omitempty"`
	Title             string                 `json:"title,omitempty"`
	OwnerUserID       string                 `json:"owner_user_id,omitempty"`
	Status            RunGroupStatus         `json:"status"`
	TriggerKind       string                 `json:"trigger_kind,omitempty"`
	TriggerRef        string                 `json:"trigger_ref,omitempty"`
	Metadata          map[string]interface{} `json:"metadata,omitempty"`
	Summary           map[string]interface{} `json:"summary,omitempty"`
	CreatedAt         time.Time              `json:"created_at"`
	UpdatedAt         time.Time              `json:"updated_at"`
	StartedAt         *time.Time             `json:"started_at,omitempty"`
	FinishedAt        *time.Time             `json:"finished_at,omitempty"`
}

type EvalRunFilter struct {
	OwnerUserID string
	EvalSpecID  string
	Statuses    []RunGroupStatus
	Limit       int
}

type EvalRunReport struct {
	EvalRun        *EvalRun        `json:"eval_run"`
	EvalSpec       *EvalSpec       `json:"eval_spec,omitempty"`
	Dataset        *Dataset        `json:"dataset,omitempty"`
	DatasetVersion *DatasetVersion `json:"dataset_version,omitempty"`
	GroupReport    *RunGroupReport `json:"group_report,omitempty"`
}

type EvalRunSpec struct {
	EvalSpecID        string                 `json:"eval_spec_id"`
	BaselineEvalRunID string                 `json:"baseline_eval_run_id,omitempty"`
	Title             string                 `json:"title,omitempty"`
	OwnerUserID       string                 `json:"owner_user_id,omitempty"`
	TriggerKind       string                 `json:"trigger_kind,omitempty"`
	TriggerRef        string                 `json:"trigger_ref,omitempty"`
	Metadata          map[string]interface{} `json:"metadata,omitempty"`
}

type EvalSpec struct {
	ID               string                 `json:"id"`
	Name             string                 `json:"name"`
	OwnerUserID      string                 `json:"owner_user_id,omitempty"`
	Subject          string                 `json:"subject,omitempty"`
	RunKind          RunKind                `json:"run_kind"`
	Profile          string                 `json:"profile,omitempty"`
	DatasetID        string                 `json:"dataset_id,omitempty"`
	DatasetVersionID string                 `json:"dataset_version_id,omitempty"`
	SchedulerConfig  GroupSchedulerConfig   `json:"scheduler_config,omitempty"`
	ScoringConfig    GroupScoringConfig     `json:"scoring_config,omitempty"`
	RuntimePolicy    map[string]interface{} `json:"runtime_policy,omitempty"`
	Metadata         map[string]interface{} `json:"metadata,omitempty"`
	CreatedAt        time.Time              `json:"created_at"`
	UpdatedAt        time.Time              `json:"updated_at"`
}

type EvalSpecFilter struct {
	OwnerUserID string
	DatasetID   string
	Limit       int
}

type EvalSpecSpec struct {
	Name             string                 `json:"name"`
	OwnerUserID      string                 `json:"owner_user_id,omitempty"`
	Subject          string                 `json:"subject,omitempty"`
	RunKind          RunKind                `json:"run_kind,omitempty"`
	Profile          string                 `json:"profile,omitempty"`
	DatasetID        string                 `json:"dataset_id"`
	DatasetVersionID string                 `json:"dataset_version_id,omitempty"`
	SchedulerConfig  GroupSchedulerConfig   `json:"scheduler"`
	ScoringConfig    GroupScoringConfig     `json:"scoring"`
	RuntimePolicy    map[string]interface{} `json:"runtime_policy,omitempty"`
	Metadata         map[string]interface{} `json:"metadata,omitempty"`
}

type EvolutionOverview struct {
	SkillID      string                             `json:"skill_id,omitempty"`
	Revisions    EvolutionOverviewRevisionCounts    `json:"revisions"`
	Instructions EvolutionOverviewInstructionCounts `json:"instructions"`
}

type EvolutionOverviewInstructionCounts struct {
	Pending int `json:"pending"`
}

type EvolutionOverviewRevisionCounts struct {
	Accepted int `json:"accepted"`
}

type EvolutionProposalSummaryProvider interface {
	PendingProposalCount(ctx context.Context, ownerUserID string) (int, error)
}

type ExecPathGuard struct {
	// contains filtered or unexported fields
}

type ExecutionEquivalenceCheck struct {
	Name     string                 `json:"name"`
	Passed   bool                   `json:"passed"`
	Actual   interface{}            `json:"actual,omitempty"`
	Expected interface{}            `json:"expected,omitempty"`
	Details  map[string]interface{} `json:"details,omitempty"`
}

type ExecutionEquivalenceMetrics struct {
	CaseCount                    int                                           `json:"case_count"`
	PassedCount                  int                                           `json:"passed_count"`
	PassRate                     float64                                       `json:"pass_rate"`
	CriticalCaseCount            int                                           `json:"critical_case_count"`
	CriticalPassedCount          int                                           `json:"critical_passed_count"`
	CriticalPassRate             float64                                       `json:"critical_pass_rate"`
	InfraBlockedCount            int                                           `json:"infra_blocked_count"`
	BaseInfraBlockedCount        int                                           `json:"base_infra_blocked_count"`
	TargetInfraBlockedCount      int                                           `json:"target_infra_blocked_count"`
	InfraBlockedDelta            int                                           `json:"infra_blocked_delta"`
	BasePassRate                 float64                                       `json:"base_pass_rate"`
	TargetPassRate               float64                                       `json:"target_pass_rate"`
	PassRateDelta                float64                                       `json:"pass_rate_delta"`
	BaseVerificationPassRate     float64                                       `json:"base_verification_pass_rate"`
	TargetVerificationPassRate   float64                                       `json:"target_verification_pass_rate"`
	VerificationPassRateDelta    float64                                       `json:"verification_pass_rate_delta"`
	BaseEvidenceBackedPassRate   float64                                       `json:"base_evidence_backed_pass_rate"`
	TargetEvidenceBackedPassRate float64                                       `json:"target_evidence_backed_pass_rate"`
	EvidenceBackedPassRateDelta  float64                                       `json:"evidence_backed_pass_rate_delta"`
	RegressionCount              int                                           `json:"regression_count"`
	ImprovementCount             int                                           `json:"improvement_count"`
	NewFailureCount              int                                           `json:"new_failure_count"`
	ResolvedFailureCount         int                                           `json:"resolved_failure_count"`
	CriticalRegressionCount      int                                           `json:"critical_regression_count"`
	LocaleBreakdown              map[string]ExecutionEquivalenceSegmentMetrics `json:"locale_breakdown,omitempty"`
	PrimaryRouteBreakdown        map[string]ExecutionEquivalenceSegmentMetrics `json:"primary_route_breakdown,omitempty"`
}

type ExecutionEquivalenceReport struct {
	TargetEvalRunID    string                      `json:"target_eval_run_id"`
	BaseEvalRunID      string                      `json:"base_eval_run_id,omitempty"`
	BaselineID         string                      `json:"baseline_id,omitempty"`
	ComparisonReportID string                      `json:"comparison_report_id,omitempty"`
	Metrics            ExecutionEquivalenceMetrics `json:"metrics"`
	Thresholds         map[string]interface{}      `json:"thresholds,omitempty"`
	Checks             []ExecutionEquivalenceCheck `json:"checks,omitempty"`
	Passed             bool                        `json:"passed"`
	CreatedAt          time.Time                   `json:"created_at"`
}

type ExecutionEquivalenceRequest struct {
	BaseEvalRunID string                         `json:"base_eval_run_id,omitempty"`
	BaselineID    string                         `json:"baseline_id,omitempty"`
	Thresholds    ExecutionEquivalenceThresholds `json:"thresholds,omitempty"`
}

type ExecutionEquivalenceSegmentMetrics struct {
	CaseCount               int     `json:"case_count"`
	PassedCount             int     `json:"passed_count"`
	PassRate                float64 `json:"pass_rate"`
	CriticalCaseCount       int     `json:"critical_case_count"`
	CriticalPassedCount     int     `json:"critical_passed_count"`
	CriticalPassRate        float64 `json:"critical_pass_rate"`
	InfraBlockedCount       int     `json:"infra_blocked_count"`
	RegressionCount         int     `json:"regression_count"`
	ImprovementCount        int     `json:"improvement_count"`
	NewFailureCount         int     `json:"new_failure_count"`
	ResolvedFailureCount    int     `json:"resolved_failure_count"`
	CriticalRegressionCount int     `json:"critical_regression_count"`
}

type ExecutionEquivalenceThresholds struct {
	MaxPassRateDrop               *float64 `json:"max_pass_rate_drop,omitempty"`
	MaxCriticalRegressionCount    *int     `json:"max_critical_regression_count,omitempty"`
	MaxVerificationPassRateDrop   *float64 `json:"max_verification_pass_rate_drop,omitempty"`
	MaxEvidenceBackedPassRateDrop *float64 `json:"max_evidence_backed_pass_rate_drop,omitempty"`
}

type ExecutionMiddleware interface {
	BeforeStart(ctx context.Context, runCtx *RunContext) error
	AfterStart(ctx context.Context, runCtx *RunContext)
	OnStartError(ctx context.Context, runCtx *RunContext, runErr error)
}

type ExecutionMiddlewareHooks struct {
	BeforeStartFunc  func(ctx context.Context, runCtx *RunContext) error
	AfterStartFunc   func(ctx context.Context, runCtx *RunContext)
	OnStartErrorFunc func(ctx context.Context, runCtx *RunContext, runErr error)
}

type GroupDispatcher struct {
	// contains filtered or unexported fields
}

type GroupPromotionResult struct {
	Dataset        *Dataset        `json:"dataset,omitempty"`
	DatasetVersion *DatasetVersion `json:"dataset_version,omitempty"`
	EvalSpec       *EvalSpec       `json:"eval_spec,omitempty"`
}

type GroupPromotionSpec struct {
	DatasetName string `json:"dataset_name"`
	Description string `json:"description,omitempty"`
	Subject     string `json:"subject,omitempty"`
	EvalName    string `json:"eval_name"`
}

type GroupSchedulerConfig struct {
	MaxConcurrency int           `json:"max_concurrency,omitempty"`
	MaxAttempts    int           `json:"max_attempts,omitempty"`
	LeaseTTL       time.Duration `json:"lease_ttl,omitempty"`
	RetryBackoff   time.Duration `json:"retry_backoff,omitempty"`
}

type GroupScoringConfig struct {
	Mode          ScoringMode `json:"mode,omitempty"`
	RuleProfile   string      `json:"rule_profile,omitempty"`
	JudgeModel    string      `json:"judge_model,omitempty"`
	PassThreshold float64     `json:"pass_threshold,omitempty"`
}

type GuardPipelineError struct {
	Stage   RuntimeStage           `json:"stage"`
	Code    string                 `json:"code"`
	Message string                 `json:"message"`
	Details map[string]interface{} `json:"details,omitempty"`
	// contains filtered or unexported fields
}

type Handler struct {
	// contains filtered or unexported fields
}

type HarnessAPICheck struct {
	Name          string `json:"name,omitempty"`
	Target        string `json:"target,omitempty"`
	Expectation   string `json:"expectation,omitempty"`
	RequiredCheck string `json:"required_check,omitempty"`
	FailureLabel  string `json:"failure_label,omitempty"`
}

type HarnessAdaptivePolicy struct {
	Profile             string `json:"profile,omitempty"`
	Reason              string `json:"reason,omitempty"`
	EnableExternalQA    bool   `json:"enable_external_qa,omitempty"`
	EnableBrowserQA     bool   `json:"enable_browser_qa,omitempty"`
	EnableCheckpoints   bool   `json:"enable_checkpoints,omitempty"`
	MaxRecoveryAttempts int    `json:"max_recovery_attempts,omitempty"`
	CheckpointInterval  int    `json:"checkpoint_interval,omitempty"`
}

type HarnessBrowserCheck struct {
	Name                string `json:"name,omitempty"`
	Target              string `json:"target,omitempty"`
	Expectation         string `json:"expectation,omitempty"`
	RequiredObservation string `json:"required_observation,omitempty"`
	RequiredArtifact    string `json:"required_artifact,omitempty"`
	FailureLabel        string `json:"failure_label,omitempty"`
	RequireScreenshot   bool   `json:"require_screenshot,omitempty"`
}

type HarnessCheckpoint struct {
	Version           string          `json:"version,omitempty"`
	RunID             string          `json:"run_id,omitempty"`
	GroupID           string          `json:"group_id,omitempty"`
	GroupItemID       string          `json:"group_item_id,omitempty"`
	AttemptIndex      int             `json:"attempt_index,omitempty"`
	Goal              string          `json:"goal,omitempty"`
	Summary           string          `json:"summary,omitempty"`
	VerifiedEvidence  []string        `json:"verified_evidence,omitempty"`
	UnresolvedRisks   []string        `json:"unresolved_risks,omitempty"`
	FailureLabels     []string        `json:"failure_labels,omitempty"`
	NextContract      HarnessContract `json:"next_contract,omitempty"`
	EvaluatorInput    string          `json:"evaluator_input,omitempty"`
	RecommendedResume string          `json:"recommended_resume,omitempty"`
	CreatedAt         time.Time       `json:"created_at"`
}

type HarnessContract struct {
	Deliverables          []string                  `json:"deliverables,omitempty"`
	SuccessCriteria       []string                  `json:"success_criteria,omitempty"`
	ExpectedArtifacts     []HarnessExpectedArtifact `json:"expected_artifacts,omitempty"`
	RequiredToolCalls     []string                  `json:"required_tool_calls,omitempty"`
	ForbiddenToolCalls    []string                  `json:"forbidden_tool_calls,omitempty"`
	RequiredChecks        []string                  `json:"required_checks,omitempty"`
	RequiredObservations  []string                  `json:"required_observations,omitempty"`
	ForbiddenObservations []string                  `json:"forbidden_observations,omitempty"`
	BrowserChecks         []HarnessBrowserCheck     `json:"browser_checks,omitempty"`
	APIChecks             []HarnessAPICheck         `json:"api_checks,omitempty"`
	FallbackOrder         []string                  `json:"fallback_order,omitempty"`
	StopConditions        []string                  `json:"stop_conditions,omitempty"`
	EvaluatorHints        []string                  `json:"evaluator_hints,omitempty"`
	RiskLevel             string                    `json:"risk_level,omitempty"`
}

type HarnessExpectedArtifact struct {
	Path      string `json:"path,omitempty"`
	Label     string `json:"label,omitempty"`
	MustExist bool   `json:"must_exist,omitempty"`
}

type HarnessSubagentExecutor struct {
	// contains filtered or unexported fields
}

type HarnessVerificationResult struct {
	Passed         bool                     `json:"passed"`
	Retryable      bool                     `json:"retryable"`
	FailureLabel   string                   `json:"failure_label,omitempty"`
	Summary        string                   `json:"summary,omitempty"`
	OutcomeScore   float64                  `json:"outcome_score,omitempty"`
	EvidenceScore  float64                  `json:"evidence_score,omitempty"`
	ExecutionScore float64                  `json:"execution_score,omitempty"`
	Observations   []string                 `json:"observations,omitempty"`
	Checks         []map[string]interface{} `json:"checks,omitempty"`
	Artifacts      []map[string]interface{} `json:"artifacts,omitempty"`
	TraceSummary   map[string]interface{}   `json:"trace_summary,omitempty"`
}

type ImportDatasetBundleEvalSpec struct {
	Name            string                 `json:"name"`
	OwnerUserID     string                 `json:"owner_user_id,omitempty"`
	Subject         string                 `json:"subject,omitempty"`
	RunKind         RunKind                `json:"run_kind,omitempty"`
	Profile         string                 `json:"profile,omitempty"`
	SchedulerConfig GroupSchedulerConfig   `json:"scheduler,omitempty"`
	ScoringConfig   GroupScoringConfig     `json:"scoring,omitempty"`
	RuntimePolicy   map[string]interface{} `json:"runtime_policy,omitempty"`
	Metadata        map[string]interface{} `json:"metadata,omitempty"`
}

type ImportDatasetBundleFromSourceRequest struct {
	SourceType string `json:"source_type,omitempty"`
	Path       string `json:"path,omitempty"`
	Source     string `json:"source,omitempty"`
	BundlePath string `json:"bundle_path,omitempty"`
	Version    string `json:"version,omitempty"`
	MakeActive *bool  `json:"make_active,omitempty"`
}

type ImportDatasetBundleRequest struct {
	SourceType string                        `json:"source_type,omitempty"`
	SourceRef  string                        `json:"source_ref,omitempty"`
	Dataset    DatasetSpec                   `json:"dataset"`
	Version    DatasetVersionSpec            `json:"version"`
	EvalSpecs  []ImportDatasetBundleEvalSpec `json:"eval_specs,omitempty"`
	MakeActive bool                          `json:"make_active,omitempty"`
}

type ImportDatasetBundleResult struct {
	Dataset        *Dataset        `json:"dataset,omitempty"`
	DatasetVersion *DatasetVersion `json:"dataset_version,omitempty"`
	EvalSpecs      []EvalSpec      `json:"eval_specs,omitempty"`
}

type JudgeEvaluationRequest struct {
	Model        string
	Group        *RunGroup
	Item         *RunGroupItem
	Run          *Run
	Calibration  map[string]interface{}
	Verification *HarnessVerificationResult
}

type JudgeEvaluationResult struct {
	Backend string
	Model   string
	Verdict ScoreVerdict
	Score   float64
	Reason  string
	Trace   map[string]interface{}
}

type JudgeEvaluator interface {
	Evaluate(ctx context.Context, req JudgeEvaluationRequest) (*JudgeEvaluationResult, error)
}

type LLMJudgeEvaluator struct {
	// contains filtered or unexported fields
}

type LegacyStoreMigrationResult struct {
	SourcePath   string
	ArchivedPath string
	RowsImported int
}

type Manager interface {
	Submit(ctx context.Context, spec RunSpec) (*Run, error)
	Get(ctx context.Context, id string) (*Run, error)
	List(ctx context.Context, filter RunFilter) ([]Run, error)
	Cancel(ctx context.Context, id string, reason string) error
	PerformAction(ctx context.Context, id string, action string, input map[string]interface{}) (*Run, error)
	SpawnChild(ctx context.Context, parentID string, spec RunSpec) (*Run, error)
	AppendEvent(ctx context.Context, event RunEvent) error
	AttachArtifact(ctx context.Context, ref ArtifactRef) error
}

type OptimizationReason string

type OptimizationSurface string

type OptimizationTrigger struct {
	Reason              OptimizationReason     `json:"reason"`
	CandidateID         string                 `json:"candidate_id,omitempty"`
	EvalRunID           string                 `json:"eval_run_id,omitempty"`
	BaseEvalRunID       string                 `json:"base_eval_run_id,omitempty"`
	OptimizationRun     bool                   `json:"optimization_run,omitempty"`
	OptimizationSurface OptimizationSurface    `json:"optimization_surface,omitempty"`
	Metadata            map[string]interface{} `json:"metadata,omitempty"`
}

type OptimizationTriggerer interface {
	TriggerOptimization(ctx context.Context, event OptimizationTrigger) error
}

type PolicyResolver struct {
	// contains filtered or unexported fields
}

type ProposalReflector interface {
	Reflect(ctx context.Context, input selfreflect.Input) (*selfreflect.Result, error)
}

type Run struct {
	ID             string                 `json:"id"`
	RootRunID      string                 `json:"root_run_id"`
	ParentRunID    string                 `json:"parent_run_id,omitempty"`
	GroupID        string                 `json:"group_id,omitempty"`
	GroupItemID    string                 `json:"group_item_id,omitempty"`
	AttemptIndex   int                    `json:"attempt_index,omitempty"`
	Kind           RunKind                `json:"kind"`
	Status         RunStatus              `json:"status"`
	RuntimeState   agentpkg.RuntimeState  `json:"runtime_state,omitempty"`
	UserID         string                 `json:"user_id,omitempty"`
	ConversationID string                 `json:"conversation_id,omitempty"`
	SessionID      string                 `json:"session_id,omitempty"`
	AgentID        string                 `json:"agent_id,omitempty"`
	Goal           string                 `json:"goal"`
	ProviderID     string                 `json:"provider_id,omitempty"`
	Model          string                 `json:"model,omitempty"`
	Result         string                 `json:"result,omitempty"`
	Error          string                 `json:"error,omitempty"`
	Depth          int                    `json:"depth"`
	CurrentStep    int                    `json:"current_step"`
	Progress       int                    `json:"progress"`
	WorkspaceRoot  string                 `json:"workspace_root,omitempty"`
	ArtifactRoot   string                 `json:"artifact_root,omitempty"`
	SandboxMode    string                 `json:"sandbox_mode,omitempty"`
	ApprovalMode   ApprovalMode           `json:"approval_mode,omitempty"`
	MaxDuration    time.Duration          `json:"max_duration,omitempty"`
	MaxSteps       int                    `json:"max_steps,omitempty"`
	MaxToolRounds  int                    `json:"max_tool_rounds,omitempty"`
	MaxSubagents   int                    `json:"max_subagents,omitempty"`
	MaxDepth       int                    `json:"max_depth,omitempty"`
	Metadata       map[string]interface{} `json:"metadata,omitempty"`
	CreatedAt      time.Time              `json:"created_at"`
	UpdatedAt      time.Time              `json:"updated_at"`
	StartedAt      *time.Time             `json:"started_at,omitempty"`
	FinishedAt     *time.Time             `json:"finished_at,omitempty"`
}

type RunActionAvailability struct {
	Items []ActionDescriptor `json:"items,omitempty"`
}

type RunContext struct {
	Run        *Run
	Parent     *Run
	Driver     Driver
	Controller *Controller
	Values     map[string]interface{}
}

type RunDetail struct {
	Run              *Run                     `json:"run"`
	Actions          RunActionAvailability    `json:"actions"`
	Events           []RunEvent               `json:"events"`
	Artifacts        []ArtifactRef            `json:"artifacts"`
	RunTrace         *RunTrace                `json:"run_trace,omitempty"`
	PendingApprovals []map[string]interface{} `json:"pending_approvals,omitempty"`
	PendingQuestions []map[string]interface{} `json:"pending_questions,omitempty"`
}

type RunDetailProvider interface {
	PendingApprovals(runID string) []map[string]interface{}
	PendingQuestions(runID string) []map[string]interface{}
}

type RunEnv struct {
	Manager    *Controller
	RunContext *RunContext
}

type RunEvent struct {
	ID             string    `json:"id"`
	RunID          string    `json:"run_id"`
	RootRunID      string    `json:"root_run_id,omitempty"`
	ParentRunID    string    `json:"parent_run_id,omitempty"`
	Type           string    `json:"type"`
	StepIndex      int       `json:"step_index,omitempty"`
	ToolName       string    `json:"tool_name,omitempty"`
	CapabilityKind string    `json:"capability_kind,omitempty"`
	Message        string    `json:"message,omitempty"`
	PayloadJSON    string    `json:"payload_json,omitempty"`
	CreatedAt      time.Time `json:"created_at"`
}

type RunFilter struct {
	UserID         string
	Kind           RunKind
	Kinds          []RunKind
	Statuses       []RunStatus
	ConversationID string
	GroupID        string
	GroupItemID    string
	ParentRunID    string
	RootRunID      string
	Limit          int
}

type RunGroup struct {
	ID              string                 `json:"id"`
	Kind            RunGroupKind           `json:"kind"`
	Title           string                 `json:"title,omitempty"`
	Status          RunGroupStatus         `json:"status"`
	OwnerUserID     string                 `json:"owner_user_id,omitempty"`
	Subject         string                 `json:"subject,omitempty"`
	SchedulerConfig GroupSchedulerConfig   `json:"scheduler_config,omitempty"`
	ScoringConfig   GroupScoringConfig     `json:"scoring_config,omitempty"`
	Metadata        map[string]interface{} `json:"metadata,omitempty"`
	Summary         map[string]interface{} `json:"summary,omitempty"`
	CreatedAt       time.Time              `json:"created_at"`
	UpdatedAt       time.Time              `json:"updated_at"`
	StartedAt       *time.Time             `json:"started_at,omitempty"`
	FinishedAt      *time.Time             `json:"finished_at,omitempty"`
}

type RunGroupFilter struct {
	OwnerUserID string
	Kinds       []RunGroupKind
	Statuses    []RunGroupStatus
	Limit       int
}

type RunGroupItem struct {
	ID             string                 `json:"id"`
	GroupID        string                 `json:"group_id"`
	Index          int                    `json:"index"`
	RunKind        RunKind                `json:"run_kind"`
	Profile        string                 `json:"profile,omitempty"`
	Input          map[string]interface{} `json:"input,omitempty"`
	Expected       map[string]interface{} `json:"expected,omitempty"`
	Metadata       map[string]interface{} `json:"metadata,omitempty"`
	Status         RunGroupItemStatus     `json:"status"`
	LatestRunID    string                 `json:"latest_run_id,omitempty"`
	AttemptCount   int                    `json:"attempt_count,omitempty"`
	MaxAttempts    int                    `json:"max_attempts,omitempty"`
	LeaseOwner     string                 `json:"lease_owner,omitempty"`
	LeaseExpiresAt *time.Time             `json:"lease_expires_at,omitempty"`
	CreatedAt      time.Time              `json:"created_at"`
	UpdatedAt      time.Time              `json:"updated_at"`
}

type RunGroupItemSpec struct {
	RunKind  RunKind                `json:"run_kind"`
	Profile  string                 `json:"profile,omitempty"`
	Input    map[string]interface{} `json:"input,omitempty"`
	Expected map[string]interface{} `json:"expected,omitempty"`
	Metadata map[string]interface{} `json:"metadata,omitempty"`
}

type RunGroupItemStatus string

type RunGroupKind string

type RunGroupReport struct {
	Group           *RunGroup                         `json:"group"`
	Items           []RunGroupItem                    `json:"items,omitempty"`
	VerdictCounts   map[string]int                    `json:"verdict_counts,omitempty"`
	OverallScore    float64                           `json:"overall_score,omitempty"`
	PassRate        float64                           `json:"pass_rate,omitempty"`
	Breakdown       map[string]interface{}            `json:"breakdown,omitempty"`
	FailedItems     []map[string]interface{}          `json:"failed_items,omitempty"`
	LinkedRuns      []Run                             `json:"linked_runs,omitempty"`
	Artifacts       []ArtifactRef                     `json:"artifacts,omitempty"`
	Scorecards      []Scorecard                       `json:"scorecards,omitempty"`
	RuntimeEvidence map[string][]RuntimeEvidenceEntry `json:"runtime_evidence,omitempty"`
	RuntimeTraces   map[string]RunTrace               `json:"runtime_traces,omitempty"`
	ItemContracts   map[string]HarnessContract        `json:"item_contracts,omitempty"`
	Checkpoints     []CheckpointArtifact              `json:"checkpoints,omitempty"`
}

type RunGroupSpec struct {
	Kind            RunGroupKind           `json:"kind"`
	Title           string                 `json:"title,omitempty"`
	Subject         string                 `json:"subject,omitempty"`
	OwnerUserID     string                 `json:"owner_user_id,omitempty"`
	Metadata        map[string]interface{} `json:"metadata,omitempty"`
	SchedulerConfig GroupSchedulerConfig   `json:"scheduler"`
	ScoringConfig   GroupScoringConfig     `json:"scoring"`
	Items           []RunGroupItemSpec     `json:"items"`
}

type RunGroupStatus string

type RunKind string

type RunSpec struct {
	Kind           RunKind                `json:"kind"`
	Goal           string                 `json:"goal"`
	UserID         string                 `json:"user_id,omitempty"`
	ConversationID string                 `json:"conversation_id,omitempty"`
	SessionID      string                 `json:"session_id,omitempty"`
	ParentRunID    string                 `json:"parent_run_id,omitempty"`
	GroupID        string                 `json:"group_id,omitempty"`
	GroupItemID    string                 `json:"group_item_id,omitempty"`
	AttemptIndex   int                    `json:"attempt_index,omitempty"`
	AgentID        string                 `json:"agent_id,omitempty"`
	ProviderID     string                 `json:"provider_id,omitempty"`
	Model          string                 `json:"model,omitempty"`
	WorkspaceRoot  string                 `json:"workspace_root,omitempty"`
	SandboxMode    string                 `json:"sandbox_mode,omitempty"`
	ApprovalMode   ApprovalMode           `json:"approval_mode,omitempty"`
	MaxDuration    time.Duration          `json:"max_duration,omitempty"`
	MaxSteps       int                    `json:"max_steps,omitempty"`
	MaxToolRounds  int                    `json:"max_tool_rounds,omitempty"`
	MaxSubagents   int                    `json:"max_subagents,omitempty"`
	MaxDepth       int                    `json:"max_depth,omitempty"`
	Metadata       map[string]interface{} `json:"metadata,omitempty"`
}

type RunStatus string

type RunTrace struct {
	RunID       string          `json:"run_id"`
	RootRunID   string          `json:"root_run_id,omitempty"`
	ParentRunID string          `json:"parent_run_id,omitempty"`
	Kind        RunKind         `json:"kind,omitempty"`
	Status      RunStatus       `json:"status,omitempty"`
	StartedAt   *time.Time      `json:"started_at,omitempty"`
	FinishedAt  *time.Time      `json:"finished_at,omitempty"`
	LatencyMs   int64           `json:"latency_ms,omitempty"`
	Stages      []RunTraceStage `json:"stages,omitempty"`
	Events      []RunTraceEvent `json:"events,omitempty"`
	Artifacts   []ArtifactRef   `json:"artifacts,omitempty"`
}

type RunTraceCollector struct {
	// contains filtered or unexported fields
}

type RunTraceEvent struct {
	Type           string    `json:"type"`
	Message        string    `json:"message,omitempty"`
	StepIndex      int       `json:"step_index,omitempty"`
	ToolName       string    `json:"tool_name,omitempty"`
	CapabilityKind string    `json:"capability_kind,omitempty"`
	CreatedAt      time.Time `json:"created_at"`
}

type RunTraceProvider interface {
	Snapshot(ctx context.Context, runID string) (*RunTrace, error)
}

type RunTraceStage struct {
	Stage     RuntimeStage           `json:"stage"`
	Message   string                 `json:"message,omitempty"`
	Status    string                 `json:"status,omitempty"`
	Details   map[string]interface{} `json:"details,omitempty"`
	CreatedAt time.Time              `json:"created_at"`
}

type RuntimeEvidenceEntry struct {
	ID           string    `json:"id"`
	RunID        string    `json:"run_id"`
	StepIndex    int       `json:"step_index,omitempty"`
	PlannerRound int       `json:"planner_round,omitempty"`
	EventType    string    `json:"event_type"`
	Summary      string    `json:"summary,omitempty"`
	PayloadJSON  string    `json:"payload_json,omitempty"`
	CreatedAt    time.Time `json:"created_at"`
}

type RuntimeEvidenceProvider interface {
	ListRuntimeEvidence(ctx context.Context, run *Run) ([]RuntimeEvidenceEntry, error)
}

type RuntimeObserver struct {
	// contains filtered or unexported fields
}

type RuntimeReflectionCoordinator struct {
	// contains filtered or unexported fields
}

type RuntimeStage string

type SQLiteStore struct {
	// contains filtered or unexported fields
}

type ScoreVerdict string

type Scorecard struct {
	ID             string       `json:"id"`
	GroupID        string       `json:"group_id"`
	GroupItemID    string       `json:"group_item_id"`
	RunID          string       `json:"run_id,omitempty"`
	Mode           ScoringMode  `json:"mode"`
	Verdict        ScoreVerdict `json:"verdict"`
	Score          float64      `json:"score"`
	BreakdownJSON  string       `json:"breakdown_json,omitempty"`
	EvidenceJSON   string       `json:"evidence_json,omitempty"`
	JudgeTraceJSON string       `json:"judge_trace_json,omitempty"`
	CreatedAt      time.Time    `json:"created_at"`
}

type ScoringMode string

type SelectorCuratedAssets struct {
	Dataset        *Dataset        `json:"dataset,omitempty"`
	DatasetVersion *DatasetVersion `json:"dataset_version,omitempty"`
	EvalSpec       *EvalSpec       `json:"eval_spec,omitempty"`
}

type SelectorGateCheck struct {
	Name     string                 `json:"name"`
	Passed   bool                   `json:"passed"`
	Actual   interface{}            `json:"actual,omitempty"`
	Expected interface{}            `json:"expected,omitempty"`
	Details  map[string]interface{} `json:"details,omitempty"`
}

type SelectorGateMetrics struct {
	CaseCount                         int                                   `json:"case_count"`
	PassedCount                       int                                   `json:"passed_count"`
	PassRate                          float64                               `json:"pass_rate"`
	CriticalCaseCount                 int                                   `json:"critical_case_count"`
	CriticalPassedCount               int                                   `json:"critical_passed_count"`
	CriticalPassRate                  float64                               `json:"critical_pass_rate"`
	RouteCaseCount                    int                                   `json:"route_case_count"`
	RouteAgreementCount               int                                   `json:"route_agreement_count"`
	RouteAgreementRate                float64                               `json:"route_agreement_rate"`
	RouteCompatibleCount              int                                   `json:"route_compatible_count"`
	RouteCompatibleRate               float64                               `json:"route_compatible_rate"`
	RouteImprovementCount             int                                   `json:"route_improvement_count"`
	RouteDisagreementCount            int                                   `json:"route_disagreement_count"`
	CriticalRegressionCount           int                                   `json:"critical_regression_count"`
	BaseClarifyRate                   float64                               `json:"base_clarify_rate"`
	TargetClarifyRate                 float64                               `json:"target_clarify_rate"`
	ClarifyRateDelta                  float64                               `json:"clarify_rate_delta"`
	LocaleBreakdown                   map[string]SelectorGateSegmentMetrics `json:"locale_breakdown,omitempty"`
	PrimaryRouteBreakdown             map[string]SelectorGateSegmentMetrics `json:"primary_route_breakdown,omitempty"`
	SelectedCanonicalSkillBreakdown   map[string]int                        `json:"selected_canonical_skill_breakdown,omitempty"`
	NativeSurfaceModeBreakdown        map[string]int                        `json:"native_surface_mode_breakdown,omitempty"`
	NativeSurfaceReasonBreakdown      map[string]int                        `json:"native_surface_reason_breakdown,omitempty"`
	ExecutionProfileBreakdown         map[string]int                        `json:"execution_profile_breakdown,omitempty"`
	ToolSurfaceAliasRewriteCount      int                                   `json:"tool_surface_alias_rewrite_count,omitempty"`
	ToolSurfaceCacheInvalidationCount int                                   `json:"tool_surface_cache_invalidation_count,omitempty"`
	ToolSurfaceExecCutoverCount       int                                   `json:"tool_surface_exec_cutover_count,omitempty"`
}

type SelectorGateReport struct {
	TargetEvalRunID    string                 `json:"target_eval_run_id"`
	BaseEvalRunID      string                 `json:"base_eval_run_id,omitempty"`
	BaselineID         string                 `json:"baseline_id,omitempty"`
	ComparisonReportID string                 `json:"comparison_report_id,omitempty"`
	Metrics            SelectorGateMetrics    `json:"metrics"`
	Thresholds         map[string]interface{} `json:"thresholds,omitempty"`
	Checks             []SelectorGateCheck    `json:"checks,omitempty"`
	Passed             bool                   `json:"passed"`
	CreatedAt          time.Time              `json:"created_at"`
}

type SelectorGateRequest struct {
	BaseEvalRunID string                 `json:"base_eval_run_id,omitempty"`
	BaselineID    string                 `json:"baseline_id,omitempty"`
	Thresholds    SelectorGateThresholds `json:"thresholds,omitempty"`
}

type SelectorGateSegmentMetrics struct {
	CaseCount               int     `json:"case_count"`
	RouteAgreementCount     int     `json:"route_agreement_count"`
	RouteAgreementRate      float64 `json:"route_agreement_rate"`
	RouteCompatibleCount    int     `json:"route_compatible_count"`
	RouteCompatibleRate     float64 `json:"route_compatible_rate"`
	RouteImprovementCount   int     `json:"route_improvement_count"`
	RouteDisagreementCount  int     `json:"route_disagreement_count"`
	CriticalCaseCount       int     `json:"critical_case_count"`
	CriticalRegressionCount int     `json:"critical_regression_count"`
	BaseClarifyCount        int     `json:"base_clarify_count"`
	BaseClarifyRate         float64 `json:"base_clarify_rate"`
	TargetClarifyCount      int     `json:"target_clarify_count"`
	TargetClarifyRate       float64 `json:"target_clarify_rate"`
	ClarifyRateDelta        float64 `json:"clarify_rate_delta"`
}

type SelectorGateThresholds struct {
	MinPassRate                *float64 `json:"min_pass_rate,omitempty"`
	MinCriticalPassRate        *float64 `json:"min_critical_pass_rate,omitempty"`
	MinRouteAgreementRate      *float64 `json:"min_route_agreement_rate,omitempty"`
	MinRouteCompatibleRate     *float64 `json:"min_route_compatible_rate,omitempty"`
	MaxClarifyRateDelta        *float64 `json:"max_clarify_rate_delta,omitempty"`
	MaxCriticalRegressionCount *int     `json:"max_critical_regression_count,omitempty"`
}

type SkillCutoverBudgetCheck struct {
	Name     string                 `json:"name"`
	Passed   bool                   `json:"passed"`
	Actual   interface{}            `json:"actual,omitempty"`
	Expected interface{}            `json:"expected,omitempty"`
	Details  map[string]interface{} `json:"details,omitempty"`
}

type SkillCutoverBudgetMetrics struct {
	CaseCount                         int            `json:"case_count"`
	ComparableCaseCount               int            `json:"comparable_case_count"`
	MissingSurfaceCaseCount           int            `json:"missing_surface_case_count"`
	BaseMedianToolCount               float64        `json:"base_median_tool_count"`
	TargetMedianToolCount             float64        `json:"target_median_tool_count"`
	BaseMedianSchemaBytes             float64        `json:"base_median_schema_bytes"`
	TargetMedianSchemaBytes           float64        `json:"target_median_schema_bytes"`
	MedianSchemaByteReductionRate     float64        `json:"median_schema_byte_reduction_rate"`
	BaseMedianLatencyMs               float64        `json:"base_median_latency_ms"`
	TargetMedianLatencyMs             float64        `json:"target_median_latency_ms"`
	MedianLatencyIncreaseRate         float64        `json:"median_latency_increase_rate"`
	AllowedFinalNativeTools           []string       `json:"allowed_final_native_tools,omitempty"`
	AllowedFinalNativeToolCases       int            `json:"allowed_final_native_tool_cases"`
	AllowedFinalNativeToolCaseRate    float64        `json:"allowed_final_native_tool_case_rate"`
	NonAllowedNativeToolCaseCount     int            `json:"non_allowed_native_tool_case_count"`
	SelectedCanonicalSkillBreakdown   map[string]int `json:"selected_canonical_skill_breakdown,omitempty"`
	NativeSurfaceModeBreakdown        map[string]int `json:"native_surface_mode_breakdown,omitempty"`
	NativeSurfaceReasonBreakdown      map[string]int `json:"native_surface_reason_breakdown,omitempty"`
	ExecutionProfileBreakdown         map[string]int `json:"execution_profile_breakdown,omitempty"`
	ToolSurfaceAliasRewriteCount      int            `json:"tool_surface_alias_rewrite_count,omitempty"`
	ToolSurfaceCacheInvalidationCount int            `json:"tool_surface_cache_invalidation_count,omitempty"`
	ToolSurfaceExecCutoverCount       int            `json:"tool_surface_exec_cutover_count,omitempty"`
}

type SkillCutoverBudgetReport struct {
	TargetEvalRunID string                    `json:"target_eval_run_id"`
	BaseEvalRunID   string                    `json:"base_eval_run_id,omitempty"`
	BaselineID      string                    `json:"baseline_id,omitempty"`
	Metrics         SkillCutoverBudgetMetrics `json:"metrics"`
	Thresholds      map[string]interface{}    `json:"thresholds,omitempty"`
	Checks          []SkillCutoverBudgetCheck `json:"checks,omitempty"`
	Passed          bool                      `json:"passed"`
	CreatedAt       time.Time                 `json:"created_at"`
}

type SkillCutoverBudgetRequest struct {
	BaseEvalRunID string                       `json:"base_eval_run_id,omitempty"`
	BaselineID    string                       `json:"baseline_id,omitempty"`
	Thresholds    SkillCutoverBudgetThresholds `json:"thresholds,omitempty"`
}

type SkillCutoverBudgetThresholds struct {
	MinMedianSchemaByteReductionRate *float64 `json:"min_median_schema_byte_reduction_rate,omitempty"`
	MaxMedianLatencyIncreaseRate     *float64 `json:"max_median_latency_increase_rate,omitempty"`
	AllowedFinalNativeTools          []string `json:"allowed_final_native_tools,omitempty"`
}

type SkillCutoverLaneAssessment struct {
	EvalRunID          string    `json:"eval_run_id"`
	CandidateID        string    `json:"candidate_id,omitempty"`
	Title              string    `json:"title,omitempty"`
	CreatedAt          time.Time `json:"created_at"`
	Passed             bool      `json:"passed"`
	BaseEvalRunID      string    `json:"base_eval_run_id,omitempty"`
	BaselineID         string    `json:"baseline_id,omitempty"`
	ComparisonReportID string    `json:"comparison_report_id,omitempty"`
	FailedChecks       []string  `json:"failed_checks,omitempty"`
}

type SkillCutoverLaneReadiness struct {
	EvalSpecID              string                       `json:"eval_spec_id,omitempty"`
	CandidateID             string                       `json:"candidate_id,omitempty"`
	RequiredConsecutiveRuns int                          `json:"required_consecutive_runs"`
	CandidateRunCount       int                          `json:"candidate_run_count"`
	ConsecutivePassCount    int                          `json:"consecutive_pass_count"`
	Ready                   bool                         `json:"ready"`
	BaseEvalRunID           string                       `json:"base_eval_run_id,omitempty"`
	BaselineID              string                       `json:"baseline_id,omitempty"`
	Error                   string                       `json:"error,omitempty"`
	Assessments             []SkillCutoverLaneAssessment `json:"assessments,omitempty"`
}

type SkillCutoverReadinessReport struct {
	CandidateID             string                    `json:"candidate_id,omitempty"`
	RequiredConsecutiveRuns int                       `json:"required_consecutive_runs"`
	EvaluatedGatesReady     bool                      `json:"evaluated_gates_ready"`
	Ready                   bool                      `json:"ready"`
	Selector                SkillCutoverLaneReadiness `json:"selector"`
	Execution               SkillCutoverLaneReadiness `json:"execution"`
	Budget                  SkillCutoverLaneReadiness `json:"budget"`
	UnverifiedRequirements  []string                  `json:"unverified_requirements,omitempty"`
	BlockingReasons         []string                  `json:"blocking_reasons,omitempty"`
	CreatedAt               time.Time                 `json:"created_at"`
}

type SkillCutoverReadinessRequest struct {
	OwnerUserID             string                      `json:"owner_user_id,omitempty"`
	CandidateID             string                      `json:"candidate_id,omitempty"`
	SelectorEvalSpecID      string                      `json:"selector_eval_spec_id,omitempty"`
	ExecutionEvalSpecID     string                      `json:"execution_eval_spec_id,omitempty"`
	RequiredConsecutiveRuns int                         `json:"required_consecutive_runs,omitempty"`
	MaxAssessments          int                         `json:"max_assessments,omitempty"`
	Selector                SelectorGateRequest         `json:"selector,omitempty"`
	Execution               ExecutionEquivalenceRequest `json:"execution,omitempty"`
	Budget                  SkillCutoverBudgetRequest   `json:"budget,omitempty"`
}

type SkillDecisionHistoryEntry struct {
	RevisionID          string                      `json:"revision_id"`
	SkillID             string                      `json:"skill_id"`
	Status              SkillRevisionStatus         `json:"status"`
	SourcePath          string                      `json:"source_path,omitempty"`
	CandidateID         string                      `json:"candidate_id,omitempty"`
	BaseContentSHA256   string                      `json:"base_content_sha256,omitempty"`
	OriginCaseID        string                      `json:"origin_case_id,omitempty"`
	ParentRevisionID    string                      `json:"parent_revision_id,omitempty"`
	BackupOfRevisionID  string                      `json:"backup_of_revision_id,omitempty"`
	EvalRunID           string                      `json:"eval_run_id,omitempty"`
	OptimizationRunID   string                      `json:"optimization_run_id,omitempty"`
	FollowupGate        string                      `json:"followup_gate,omitempty"`
	OptimizationSurface OptimizationSurface         `json:"optimization_surface,omitempty"`
	DecisionAction      SkillRevisionDecisionAction `json:"decision_action,omitempty"`
	ReviewNote          string                      `json:"review_note,omitempty"`
	ReviewedBy          string                      `json:"reviewed_by,omitempty"`
	DecisionLog         map[string]interface{}      `json:"decision_log,omitempty"`
	DecisionLogJSON     string                      `json:"decision_log_json,omitempty"`
	DecisionAt          time.Time                   `json:"decision_at"`
	CreatedAt           time.Time                   `json:"created_at"`
	ReviewedAt          *time.Time                  `json:"reviewed_at,omitempty"`
	PromotedAt          *time.Time                  `json:"promoted_at,omitempty"`
}