bench

type ConvoMemConversation struct {
	Messages []ConvoMemMessage `json:"messages"`
}

type ConvoMemEvidenceFile struct {
	EvidenceItems []ConvoMemItem `json:"evidence_items"`
}

type ConvoMemItem struct {
	Question         string                    `json:"question"`
	Answer           string                    `json:"answer"`
	Conversations    []ConvoMemConversation    `json:"conversations"`
	MessageEvidences []ConvoMemMessageEvidence `json:"message_evidences"`
	CategoryKey      string                    `json:"-"`
}

type ConvoMemItemResult struct {
	Question       string  `json:"question"`
	Category       string  `json:"category"`
	Recall         float64 `json:"recall"`
	RetrievedCount int     `json:"retrieved_count"`
	EvidenceCount  int     `json:"evidence_count"`
	Found          int     `json:"found"`
}

type ConvoMemMessage struct {
	Speaker string `json:"speaker"`
	Text    string `json:"text"`
}

type ConvoMemMessageEvidence struct {
	Text string `json:"text"`
}

type ConvoMemResult struct {
	Items          int                  `json:"items"`
	AverageRecall  float64              `json:"average_recall"`
	ElapsedSeconds float64              `json:"elapsed_seconds"`
	PerCategory    map[string]float64   `json:"per_category"`
	Distribution   map[string]int       `json:"distribution"`
	Results        []ConvoMemItemResult `json:"results,omitempty"`
}

type LatencyStats struct {
	Count int     `json:"count"`
	AvgMs float64 `json:"avg_ms"`
	P50Ms float64 `json:"p50_ms"`
	P95Ms float64 `json:"p95_ms"`
	P99Ms float64 `json:"p99_ms"`
	MinMs float64 `json:"min_ms"`
	MaxMs float64 `json:"max_ms"`
}

type LoCoMoDialog struct {
	DiaID   string `json:"dia_id"`
	Speaker string `json:"speaker"`
	Text    string `json:"text"`
}

type LoCoMoItemResult struct {
	Question  string   `json:"question"`
	Category  int      `json:"category"`
	Evidence  []string `json:"evidence"`
	Retrieved []string `json:"retrieved"`
	Recall    float64  `json:"recall"`
}

type LoCoMoQA struct {
	Question string   `json:"question"`
	Category int      `json:"category"`
	Evidence []string `json:"evidence"`
}

type LoCoMoResult struct {
	Questions      int                `json:"questions"`
	AverageRecall  float64            `json:"average_recall"`
	ElapsedSeconds float64            `json:"elapsed_seconds"`
	PerCategory    map[int]float64    `json:"per_category"`
	Distribution   map[string]int     `json:"distribution"`
	Results        []LoCoMoItemResult `json:"results,omitempty"`
}

type LongMemEvalEntry struct {
	Question           string   `json:"question"`
	QuestionType       string   `json:"question_type"`
	QuestionDateRaw    string   `json:"question_date"`
	AnswerSessionIDs   []string `json:"answer_session_ids"`
	HaystackSessions   [][]Turn `json:"haystack_sessions"`
	HaystackSessionIDs []string `json:"haystack_session_ids"`
	HaystackDates      []string `json:"haystack_dates"`
}

type LongMemEvalItemResult struct {
	Question          string   `json:"question"`
	QuestionType      string   `json:"question_type"`
	CorrectSessionIDs []string `json:"correct_session_ids"`
	TopResults        []string `json:"top_results"`
	RecallAt1         float64  `json:"recall_at_1"`
	RecallAt5         float64  `json:"recall_at_5"`
	RecallAt10        float64  `json:"recall_at_10"`
	ReciprocalRank    float64  `json:"reciprocal_rank"`
	NDCGAt10          float64  `json:"ndcg_at_10"`
}

type LongMemEvalResult struct {
	Questions       int                     `json:"questions"`
	RecallAt1       float64                 `json:"recall_at_1"`
	RecallAt5       float64                 `json:"recall_at_5"`
	RecallAt10      float64                 `json:"recall_at_10"`
	MRR             float64                 `json:"mrr"`
	NDCGAt10        float64                 `json:"ndcg_at_10"`
	ElapsedSeconds  float64                 `json:"elapsed_seconds"`
	PerQuestionType map[string]float64      `json:"per_question_type"`
	Distribution    map[string]int          `json:"distribution"`
	Items           []LongMemEvalItemResult `json:"items,omitempty"`
}

type MemBenchItem struct {
	Category      string          `json:"category"`
	Topic         string          `json:"topic"`
	TID           int             `json:"tid"`
	Turns         interface{}     `json:"turns"`
	Question      string          `json:"question"`
	GroundTruth   string          `json:"ground_truth"`
	AnswerText    string          `json:"answer_text"`
	TargetStepIDs [][]interface{} `json:"target_step_ids"`
}

type MemBenchItemResult struct {
	Category       string `json:"category"`
	Topic          string `json:"topic"`
	Question       string `json:"question"`
	GroundTruth    string `json:"ground_truth"`
	TargetSteps    []int  `json:"target_steps"`
	RetrievedSteps []int  `json:"retrieved_steps"`
	HitAtK         bool   `json:"hit_at_k"`
}

type MemBenchResult struct {
	Items          int                  `json:"items"`
	RecallAtK      float64              `json:"recall_at_k"`
	TopK           int                  `json:"top_k"`
	ElapsedSeconds float64              `json:"elapsed_seconds"`
	PerCategory    map[string]float64   `json:"per_category"`
	Results        []MemBenchItemResult `json:"results,omitempty"`
}

type PerfResult struct {
	Entries                int          `json:"entries"`
	Searches               int          `json:"searches"`
	StoreSizeBytes         int64        `json:"store_size_bytes"`
	IndexSizeBytes         int64        `json:"index_size_bytes"`
	InitMs                 float64      `json:"init_ms"`
	AddLatency             LatencyStats `json:"add_latency"`
	SearchLatency          LatencyStats `json:"search_latency"`
	AddThroughputPerSec    float64      `json:"add_throughput_per_sec"`
	SearchThroughputPerSec float64      `json:"search_throughput_per_sec"`
}

type Turn struct {
	Role    string `json:"role"`
	Content string `json:"content"`
}