report

type BandwidthRow struct {
	Size    string `json:"size"` // human-readable size
	AlgBW   string `json:"algBW"`
	BusBW   string `json:"busBW"`
	Samples int    `json:"samples"`
}

type CategoryReport struct {
	Domain        string `json:"domain"`
	Variant       string `json:"variant"`
	Status        string `json:"status"`
	FailureReason string `json:"failureReason,omitempty"` // populated from Workflow Failed condition message
	Runtime       string `json:"runtime,omitempty"`       // total runtime across all iterations
	TestScale     string `json:"testScale,omitempty"`
	NodesPerJob   int    `json:"nodesPerJob,omitempty"`
	Jobs          int    `json:"jobs,omitempty"`
	MNNVL         string `json:"mnnvl,omitempty"` // "Enabled", "Disabled", or "" (unknown)
	// FailedGroups lists groups that failed with their reason.
	FailedGroups []FailedGroupReport `json:"failedGroups,omitempty"`
	// Cliques lists topology domains with node counts and validation status.
	Cliques []CliqueReport `json:"cliques,omitempty"`
	// Training metrics grouped by topology domain.
	Domains []DomainReport `json:"domains,omitempty"`
	// Communication bandwidth results (single-group: one row per size).
	Bandwidth []BandwidthRow `json:"bandwidth,omitempty"`
	// Per-group bandwidth results (multi-group: one row per group).
	GroupBandwidth []GroupBandwidthRow `json:"groupBandwidth,omitempty"`
	// Diagnose results from adaptive fault isolation.
	Diagnose *DiagnoseReport `json:"diagnose,omitempty"`
	// Iterations shows per-iteration timing and outcome.
	Iterations []IterationReport `json:"iterations,omitempty"`
}

type CertReport struct {
	Title      string `json:"title,omitempty"` // defaults to "Certification Report"
	Name       string `json:"name"`
	Platform   string `json:"platform"`
	GPU        string `json:"gpu"`
	TotalNodes int    `json:"totalNodes"`
	// ExcludedNodes lists nodes that matched the target but were left
	// untested, with ExclusionReason saying why. A run reports PASSED even
	// when it excludes nodes, so these keep that from being invisible.
	ExcludedNodes   []string           `json:"excludedNodes,omitempty"`
	ExclusionReason string             `json:"exclusionReason,omitempty"`
	Categories      []CategoryReport   `json:"categories"`
	FailedNodes     []string           `json:"failedNodes"`
	Result          string             `json:"result"` // "PASSED", "INCOMPLETE", "FAILED", or "RUNNING"
	NodeResults     []NodeResultReport `json:"nodeResults,omitempty"`
}

type CliqueReport struct {
	Name      string `json:"name"`
	Total     int    `json:"total"`     // total nodes in this clique
	Validated int    `json:"validated"` // nodes that passed
	Passed    bool   `json:"passed"`
}

type DiagnoseReport struct {
	Stage                string                                         `json:"stage"`
	Rounds               int                                            `json:"rounds"`
	HealthyCount         int                                            `json:"healthyCount"`
	SuspectCount         int                                            `json:"suspectCount"`
	ConfirmedFaulty      []string                                       `json:"confirmedFaulty,omitempty"`
	InfrastructureFaults []nvcrev1alpha1.InfrastructureFault            `json:"infrastructureFaults,omitempty"`
	ScreeningResults     map[string]nvcrev1alpha1.DomainScreeningResult `json:"-"` // not serialized, used for rendering
	MaxBW                string                                         `json:"maxBW,omitempty"`
	MaxBWDomain          string                                         `json:"maxBWDomain,omitempty"`
	MaxBWNodeList        []string                                       `json:"maxBWNodes,omitempty"`
	MinBW                string                                         `json:"minBW,omitempty"`
	MinBWDomain          string                                         `json:"minBWDomain,omitempty"`
	MinBWNodeList        []string                                       `json:"minBWNodes,omitempty"`
	Tests                []DiagnoseTestRow                              `json:"tests,omitempty"`
}

type DiagnoseTestRow struct {
	Stage  string   `json:"stage"`            // screening, bisection, confirmation
	Name   string   `json:"name"`             // job name
	Nodes  []string `json:"nodes"`            // nodes in the group
	Domain string   `json:"domain,omitempty"` // clique/domain for screening tests
	BusBW  string   `json:"busBW"`            // peak bus bandwidth
	Passed bool     `json:"passed"`           // job succeeded
}

type DomainReport struct {
	Name      string `json:"name"` // e.g., "clique-0" or "" for no-topology case
	NodeCount int    `json:"nodeCount"`
	Goodput   string `json:"goodput"`  // runtime goodput ratio
	TFLOPs    string `json:"tflops"`   // avg TFLOPs per GPU
	StepTime  string `json:"stepTime"` // avg step time
}

type FailedGroupReport struct {
	Name      string   `json:"name"`
	NodeCount int      `json:"nodeCount"`
	Nodes     []string `json:"nodes,omitempty"`
	Reason    string   `json:"reason"` // e.g., "BackoffLimitExceeded (72 pods failed)"
}

type GroupBandwidthRow struct {
	GroupName string   `json:"groupName"` // e.g., "group-0" or "clique-0 (18 nodes)"
	Nodes     []string `json:"nodes"`     // node names in the group
	BusBW     string   `json:"busBW"`     // peak BusBW at largest message size
	BelowMin  bool     `json:"belowMin"`  // true if below minBusBandwidthGBps threshold
	Failed    bool     `json:"failed"`    // true if the group's Job failed
}

type IterationReport struct {
	Number   int    `json:"number"`
	Status   string `json:"status"`   // Succeeded, Failed, Running
	Duration string `json:"duration"` // e.g., "5m 30s"
}

type NodeResultReport struct {
	Name   string `json:"name"`
	Group  string `json:"group"`
	Rack   string `json:"rack,omitempty"`
	Status string `json:"status"` // "Passed" or "Failed"
}