quickbench

type AgentOverlay struct {
	Model  string // optional model selector string
	System string // optional prompt body (markdown body after frontmatter)
}

type BAR struct {
	Path       string
	Meta       Meta
	Matrix     []MatrixEntry     // subject runs from bench.yaml; empty means one "default" cell
	Agents     map[string][]byte // agent name → full .md bytes (verbatim)
	Variations []Variation
	Criteria   string
	Judge      string
	// contains filtered or unexported fields
}

type BashDouble struct {
	Command  string `json:"command,omitempty"`
	Pattern  string `json:"pattern,omitempty"`
	Output   string `json:"output"`
	ExitCode int    `json:"exitCode,omitempty"`
}

type CaptureOptions struct {
	DatabaseURL    string
	ConversationID string
	AgentsDir      string // workspace agents/ directory (required for full tree)
	Root           string // root agent name; default main
	Out            string
	Variation      string
	Pack           bool
}

type CellResult struct {
	Label         string             `json:"label"`
	Variation     string             `json:"variation"`
	Matrix        string             `json:"matrix"`
	Model         string             `json:"model"` // same as Matrix; kept for report compatibility
	Text          string             `json:"text,omitempty"`
	ToolCalls     []observedToolCall `json:"toolCalls,omitempty"`
	Error         string             `json:"error,omitempty"`
	Latency       string             `json:"latency"`
	LatencyMillis int64              `json:"latencyMillis"`
	Metrics       cellMetrics        `json:"metrics"`
}

type EloRating struct {
	Label  string  `json:"label"`
	Rating float64 `json:"rating"`
}

type MatrixAgent struct {
	Model  modelSelector // empty Raw means no model override
	System string        // empty means no system/prompt override
}

type MatrixEntry struct {
	ID     string
	Agents map[string]MatrixAgent // agent name → optional model/system overrides
}

type Message struct {
	Role string `json:"role"`
	Text string `json:"text"`
}

type Meta struct {
	Name        string
	Description string
	Tags        []string
	Root        string // default agent name; empty means "main"
}

type PairResult struct {
	A      string `json:"a"`
	B      string `json:"b"`
	Winner string `json:"winner"` // A, B, or TIE
	Raw    string `json:"raw,omitempty"`
	Error  string `json:"error,omitempty"`
}

type Report struct {
	Path    string       `json:"path"`
	Name    string       `json:"name,omitempty"`
	Cells   []CellResult `json:"cells"`
	Ladder  []EloRating  `json:"ladder,omitempty"`
	Pairs   []PairResult `json:"pairs,omitempty"`
	Skipped string       `json:"eloSkipped,omitempty"`
}

type ToolMock struct {
	Name        string         `json:"name"`
	Description string         `json:"description"`
	Parameters  map[string]any `json:"parameters"`
	Response    string         `json:"response"`
}

type Variation struct {
	ID            string
	System        string // legacy: root prompt overlay when AgentOverlays[root].System empty
	Transcript    []Message
	AgentOverlays map[string]AgentOverlay
	Tools         []ToolMock
	BashDoubles   []BashDouble
}