bench

type ActivationCriteria struct {
	MinimumRecall          map[string]float64 `yaml:"minimum_recall"`
	MaximumFalseActivation float64            `yaml:"maximum_false_activation"`
}

type ActivationInputs struct {
	Paths   []string
	Prompts ActivationPromptSet
}

type ActivationPrompt struct {
	ID     string `yaml:"id"`
	Prompt string `yaml:"prompt"`
	// Expect names the skill that should activate, or "none" for a prompt
	// no skill should answer.
	Expect string `yaml:"expect"`
	// Prepare changes the fixture before the session: "naive" applies the
	// instance's naive patch; empty leaves the tree untouched.
	Prepare string `yaml:"prepare,omitempty"`
	// Note says why the prompt is in the set.
	Note string `yaml:"note,omitempty"`
}

type ActivationPromptSet struct {
	SchemaVersion int                `yaml:"schema_version"`
	Instance      string             `yaml:"instance"`
	Prompts       []ActivationPrompt `yaml:"prompts"`
}

type ActivationRate struct {
	Numerator   int
	Denominator int
}

type ActivationReport struct {
	Inputs         []ReportInput
	Fingerprint    string
	PromptSetSHA   string
	RequestedModel string
	Model          string
	MaxTurns       int
	Rows           []ActivationRow
	// Missing lists the manifest prompts without a valid session, in
	// manifest order.
	Missing []string
	Stats   ActivationStats
	Status  string
	Reasons []string
}

type ActivationRow struct {
	SchemaVersion int    `json:"schema_version"`
	TS            string `json:"ts"`
	RunID         string `json:"run_id"`
	Instance      string `json:"instance"`
	Fixture       string `json:"fixture"`
	Fingerprint   string `json:"fingerprint"`
	// PromptSetSHA identifies the whole prompt file; PromptID and
	// PromptSHA identify the one prompt this row measured.
	PromptSetSHA string `json:"prompt_set_sha256"`
	PromptID     string `json:"prompt_id"`
	PromptSHA    string `json:"prompt_sha256"`
	Expected     string `json:"expected"`
	Prepare      string `json:"prepare,omitempty"`
	// Activated lists the skills the agent loaded; Hit says the session met
	// its expectation: the expected skill loaded, or nothing loaded for a
	// should-not prompt. Other skills loading beside the expected one keep
	// Hit true and show up in the report's cross-activation count.
	Activated []string `json:"activated,omitempty"`
	Hit       bool     `json:"hit"`
	MaxTurns  int      `json:"max_turns,omitempty"`

	Valid                 bool   `json:"valid"`
	InvalidReason         string `json:"invalid_reason,omitempty"`
	InfrastructureFailure bool   `json:"infrastructure_failure,omitempty"`

	InitSeen   bool             `json:"init_seen,omitempty"`
	ResultSeen bool             `json:"result_seen,omitempty"`
	Tools      []string         `json:"tools,omitempty"`
	MCPServers []MCPServerState `json:"mcp_servers,omitempty"`
	Skills     []string         `json:"skills,omitempty"`
	Plugins    []string         `json:"plugins,omitempty"`
	// DeniedTools lists the tools the agent asked for and was refused. A
	// refused Skill call is a recall miss the harness caused, not the
	// model, so the row is invalid rather than a miss.
	DeniedTools []string `json:"denied_tools,omitempty"`

	SeamarkCalls     int            `json:"seamark_calls"`
	SeamarkToolCalls map[string]int `json:"seamark_tool_calls,omitempty"`
	Edits            int            `json:"edits"`

	AgentUsage

	SeamarkVersion string  `json:"seamark_version,omitempty"`
	SeamarkSHA     string  `json:"seamark_sha256,omitempty"`
	AgentVersion   string  `json:"agent_version,omitempty"`
	Effort         string  `json:"effort,omitempty"`
	MaxBudgetUSD   float64 `json:"max_budget_usd,omitempty"`
	RuntimeID      string  `json:"runtime_id,omitempty"`

	Transcript    string `json:"transcript,omitempty"`
	TranscriptSHA string `json:"transcript_sha256,omitempty"`
	StderrLog     string `json:"stderr,omitempty"`
	StderrSHA     string `json:"stderr_sha256,omitempty"`
}

type ActivationStats struct {
	Recall          map[string]ActivationRate
	FalseActivation ActivationRate
	// CrossActivations counts valid should-activate sessions that loaded a
	// skill other than the expected one; two skills on one prompt is a
	// description defect the criteria do not otherwise see.
	CrossActivations int
	Valid            int
	Invalid          int
}

type ActivationSummary struct {
	Rows          []ActivationRow
	StoppedReason string
}

type AgentUsage struct {
	RequestedModel      string                `json:"requested_model,omitempty"`
	Model               string                `json:"model,omitempty"`
	ModelUsage          map[string]ModelUsage `json:"model_usage,omitempty"`
	InputTokens         int64                 `json:"input_tokens,omitempty"`
	CacheReadTokens     int64                 `json:"cache_read_input_tokens,omitempty"`
	CacheCreationTokens int64                 `json:"cache_creation_input_tokens,omitempty"`
	ContextTokens       int64                 `json:"context_tokens,omitempty"`
	OutputTokens        int64                 `json:"output_tokens,omitempty"`
	Turns               int                   `json:"turns,omitempty"`
	PermissionDenials   int                   `json:"permission_denials,omitempty"`
	CostUSD             float64               `json:"cost_usd,omitempty"`
	DurationMS          int64                 `json:"duration_ms,omitempty"`
	AgentExit           int                   `json:"agent_exit"`
	TimedOut            bool                  `json:"timed_out,omitempty"`
	AgentError          bool                  `json:"agent_error,omitempty"`
}

type Arm string

type ArmReport struct {
	Attempted        int
	Valid            int
	TaskDone         int
	InvariantPass    int
	ContextTokens    int64
	CostUSD          float64
	HookMatches      int
	HookInjections   int
	HookRepeated     int
	HookSuppressed   int
	HookContextBytes int
}

type BenchmarkReport struct {
	ResultSchemaVersion int
	ClaimSchemaVersion  int
	EvidenceFrom        string
	EvidenceTo          string
	Inputs              []ReportInput
	Cohorts             []CohortReport
	Assessments         []ClaimAssessment
}

type CheckResult struct {
	Command  string `json:"command"`
	Pass     bool   `json:"pass"`
	TimedOut bool   `json:"timed_out,omitempty"`
	Output   string `json:"output,omitempty"`
}

type Claim struct {
	ID            string `yaml:"id"`
	Claim         string `yaml:"claim"`
	PrimaryMetric string `yaml:"primary_metric"`
	Comparison    string `yaml:"comparison"`
	// TreatmentInstance and ControlInstance name the two variants a
	// difference_in_differences claim compares. Both must appear in
	// Instances; empty for hook-on_vs_hook-off claims.
	TreatmentInstance            string   `yaml:"treatment_instance,omitempty"`
	ControlInstance              string   `yaml:"control_instance,omitempty"`
	Direction                    string   `yaml:"direction"`
	RequiredModel                string   `yaml:"required_model"`
	RequiredEffort               string   `yaml:"required_effort"`
	RequireCleanSeamark          bool     `yaml:"require_clean_seamark"`
	MinimumEffect                float64  `yaml:"minimum_effect"`
	MinimumInstanceEffect        float64  `yaml:"minimum_instance_effect"`
	MaximumHarmfulInterference   float64  `yaml:"maximum_harmful_interference"`
	MinimumInstances             int      `yaml:"minimum_instances"`
	MinimumValidPairsPerInstance int      `yaml:"minimum_valid_pairs_per_instance"`
	Instances                    []string `yaml:"instances"`
}

type ClaimAssessment struct {
	ID                  string
	Definition          Claim
	Status              string
	Reason              string
	QualifyingInstances int
	MeanEffect          float64
	WorstInstanceEffect float64
	HarmfulInterference float64
	// AnchorNote carries the component effects behind a factorial claim. It is
	// report-only; the frozen verdict uses their difference.
	AnchorNote string
}

type ClaimRegistry struct {
	SchemaVersion int     `yaml:"schema_version"`
	Claims        []Claim `yaml:"claims"`
}

type CohortReport struct {
	Instance            string
	Fingerprint         string
	Task                string
	TaskSHA             string
	Pin                 string
	Fixture             string
	RequestedModel      string
	Model               string
	SeamarkVersion      string
	SeamarkSHA          string
	AgentVersion        string
	Effort              string
	HookDelivery        HookDeliveryMode
	HookExposure        HookExposureExpectation
	ComparisonFamily    string
	ProtocolFingerprint string
	RuntimeID           string
	MaxBudgetUSD        float64
	Rows                int
	ValidPairs          int
	FavorablePairs      int
	UnfavorablePairs    int
	TiedPairs           int
	HarmfulPairs        int
	HookOn              ArmReport
	HookOff             ArmReport
}

type Command struct {
	Name    string
	Args    []string
	Timeout time.Duration
}

type HookDeliveryMode = reviews.HookDeliveryMode

type HookExposureExpectation string

type Instance struct {
	ID          string
	Rule        string
	Task        string
	LessonYAML  string
	PlaceboYAML string

	Generate  func(string) error
	Judge     func(string) (Verdict, error)
	ApplyGold func(string) error
	// ApplyNaive installs a task-complete solution that deliberately omits the
	// owner invariant. Preflight uses it to prove that the two judges actually
	// discriminate the failure mode the experiment claims to measure.
	ApplyNaive func(string) error
	// JudgeVersion must change whenever verdict semantics change. It binds
	// persisted rows to the exact interpretation used by this instance.
	JudgeVersion string
	Checks       []Command
	// HookExposure controls validation of hooked arms. Empty means required,
	// preserving the original benchmark contract. Optional is reserved for a
	// scope-control whose valid outcome may be zero matching edits.
	HookExposure HookExposureExpectation
	// ComparisonFamily groups intentionally different instances that form one
	// factorial experiment. Empty means the instance is not cross-compared.
	ComparisonFamily string
	// ProtocolInstance names the canonical instance whose full fingerprint
	// defines the shared protocol for ComparisonFamily. Every family variant
	// points to the same canonical instance.
	ProtocolInstance string

	// ExploreFiles are repository-relative paths whose appearance in an
	// assistant message is useful diagnostic evidence. They do not affect the
	// verdict.
	ExploreFiles []string

	// Prepare materializes any external, pinned source required by Generate.
	// Synthetic instances leave it nil. Public-repository instances use an
	// explicit preparation step so paid runs and their agents never depend on
	// network access.
	Prepare func(context.Context) (string, error)
	// contains filtered or unexported fields
}

type MCPServerState struct {
	Name   string `json:"name"`
	Status string `json:"status,omitempty"`
}

type ModelUsage struct {
	InputTokens         int64   `json:"input_tokens,omitempty"`
	CacheReadTokens     int64   `json:"cache_read_input_tokens,omitempty"`
	CacheCreationTokens int64   `json:"cache_creation_input_tokens,omitempty"`
	OutputTokens        int64   `json:"output_tokens,omitempty"`
	CostUSD             float64 `json:"cost_usd,omitempty"`
}

type ReportInput struct {
	Path   string
	SHA256 string
	Rows   int
}

type Row struct {
	SchemaVersion int    `json:"schema_version"`
	TS            string `json:"ts"`
	RunID         string `json:"run_id"`
	Instance      string `json:"instance"`
	TaskSHA       string `json:"task_sha256"`
	Pin           string `json:"pin"`
	Arm           Arm    `json:"arm"`
	Trial         int    `json:"trial"`
	TaskDone      bool   `json:"task_pass"`
	Avoided       bool   `json:"invariant_pass"`
	Notes         string `json:"notes,omitempty"`

	// Valid says the agent session and treatment were actually delivered.
	// PairValid additionally says every requested arm in this trial number was
	// valid. Only rows satisfying both enter effect tallies.
	Valid                 bool   `json:"valid"`
	PairValid             bool   `json:"pair_valid"`
	InvalidReason         string `json:"invalid_reason,omitempty"`
	InfrastructureFailure bool   `json:"infrastructure_failure,omitempty"`
	// Fixture is the generated repo's full HEAD commit. Generation
	// is deterministic, so this identifies the exact fixture content a
	// row was measured against — rows from different fixture versions
	// must never be pooled as one series.
	Fixture string `json:"fixture,omitempty"`
	// HookFirings is how many firing records the trial repo's own
	// audit log holds after the run (hook-on arm only). Zero means the
	// injection never reached the agent and the arms were effectively
	// identical — the row proves its treatment happened instead of
	// assuming it.
	HookFirings int `json:"hook_firings,omitempty"`
	// HookAuditRows records every audit row, including unrelated or malformed
	// firings. HookFirings counts only rows proving that the selected lesson
	// reached an in-region edit through the expected hook surface.
	HookAuditRows int `json:"hook_audit_rows,omitempty"`
	// Schema v6 delivery intensity. Matches counts matching edit-hook
	// invocations; each match either injected context or was fully suppressed.
	HookMatches      int `json:"hook_matches"`
	HookInjections   int `json:"hook_injections"`
	HookRepeated     int `json:"hook_repeated_injections"`
	HookSuppressed   int `json:"hook_suppressed"`
	HookContextBytes int `json:"hook_context_bytes"`
	// Transcript is where this trial's raw agent output was saved;
	// StderrLog and Patch keep the rest of the audit record, so a
	// verdict stays checkable after the trial dir is deleted.
	Transcript    string `json:"transcript,omitempty"`
	TranscriptSHA string `json:"transcript_sha256,omitempty"`
	StderrLog     string `json:"stderr,omitempty"`
	StderrSHA     string `json:"stderr_sha256,omitempty"`
	Patch         string `json:"patch,omitempty"`
	PatchSHA      string `json:"patch_sha256,omitempty"`
	// Checks are public repository-local validation commands. Hidden task and
	// invariant judges are represented by TaskDone and Avoided above.
	ChecksPass bool          `json:"checks_pass"`
	Checks     []CheckResult `json:"checks,omitempty"`
	// LessonFileRead: the agent named .seamark/lessons.yaml in its own
	// messages. In arms without a lesson file this must be false; in a
	// control arm it would mean contamination.
	LessonFileRead bool `json:"lesson_file_read,omitempty"`
	// Explored lists instance-selected files the agent named in its own
	// messages and tool calls, in first-mention order. It is diagnostic
	// evidence only and never affects a verdict.
	Explored            []string                `json:"explored,omitempty"`
	RequestedModel      string                  `json:"requested_model,omitempty"`
	Model               string                  `json:"model,omitempty"`
	ModelUsage          map[string]ModelUsage   `json:"model_usage,omitempty"`
	InputTokens         int64                   `json:"input_tokens,omitempty"`
	CacheReadTokens     int64                   `json:"cache_read_input_tokens,omitempty"`
	CacheCreationTokens int64                   `json:"cache_creation_input_tokens,omitempty"`
	ContextTokens       int64                   `json:"context_tokens,omitempty"`
	OutputTokens        int64                   `json:"output_tokens,omitempty"`
	Turns               int                     `json:"turns,omitempty"`
	PermissionDenials   int                     `json:"permission_denials,omitempty"`
	CostUSD             float64                 `json:"cost_usd,omitempty"`
	DurationMS          int64                   `json:"duration_ms,omitempty"`
	AgentExit           int                     `json:"agent_exit"`
	TimedOut            bool                    `json:"timed_out,omitempty"`
	AgentError          bool                    `json:"agent_error,omitempty"`
	InitSeen            bool                    `json:"init_seen,omitempty"`
	ResultSeen          bool                    `json:"result_seen,omitempty"`
	Tools               []string                `json:"tools,omitempty"`
	Plugins             []string                `json:"plugins,omitempty"`
	MCPServers          []string                `json:"mcp_servers,omitempty"`
	SeamarkVersion      string                  `json:"seamark_version,omitempty"`
	SeamarkSHA          string                  `json:"seamark_sha256,omitempty"`
	AgentVersion        string                  `json:"agent_version,omitempty"`
	Effort              string                  `json:"effort,omitempty"`
	HookDelivery        HookDeliveryMode        `json:"hook_delivery,omitempty"`
	HookExposure        HookExposureExpectation `json:"hook_exposure,omitempty"`
	MaxBudgetUSD        float64                 `json:"max_budget_usd,omitempty"`
	RuntimeID           string                  `json:"runtime_id,omitempty"`
	Fingerprint         string                  `json:"fingerprint,omitempty"`
	ProtocolFingerprint string                  `json:"protocol_fingerprint,omitempty"`
	ComparisonFamily    string                  `json:"comparison_family,omitempty"`
}

type RunConfig struct {
	Trials     int           // trials per arm
	Arms       []Arm         // arms to run; nil means both
	Instance   Instance      // zero value selects SchemaSyncInstance
	AgentArgv  []string      // agent command; the task prompt is appended as the last argument
	SeamarkBin string        // absolute path to the seamark binary (hook command + index)
	Timeout    time.Duration // per-trial agent timeout; 0 means 10 minutes
	Out        string        // results JSONL path, appended one row per trial
	WorkDir    string        // parent for trial dirs; "" means a fresh temp dir
	Keep       bool          // keep trial dirs after judging, for inspection
	// TranscriptDir saves each trial's raw agent stdout (the full
	// stream-json transcript when the agent emits one) for reading WHY
	// a verdict came out the way it did. Empty disables saving.
	TranscriptDir string
	// PrepareIndex runs `seamark index` in hook-on trials so the hook
	// has a store to read. Hermetic tests turn it off.
	PrepareIndex bool
	Version      string // seamark version stamped into rows
	SeamarkSHA   string // exact binary digest stamped into rows
	AgentVersion string // agent CLI version stamped into rows
	Model        string // exact requested primary model; empty for custom agents
	Effort       string // requested effort level
	MaxBudgetUSD float64
	RuntimeID    string // sandbox/toolchain identity
	Fingerprint  string // immutable instance + runtime configuration hash
	// ProtocolFingerprint omits the intentionally varied instance scope while
	// binding every shared part of a cross-instance experiment.
	ProtocolFingerprint string
	// RunID groups rows and makes transcript names unique across concurrent
	// invocations. Empty asks Run to generate a cryptographically random ID.
	RunID string
	// HookDelivery selects the edit-hook repeat policy. Empty means always.
	HookDelivery HookDeliveryMode

	// RequireStructuredResult and RequireCleanInit are true for the
	// default Claude adapter. Stub/custom adapters may leave them false.
	RequireStructuredResult bool
	RequireCleanInit        bool
	Log                     func(string, ...any) // progress lines; nil silences
}

type Summary struct {
	Rows          []Row
	ByArm         map[Arm]Tally
	RunID         string
	Instance      string
	Rule          string
	HookExposure  HookExposureExpectation
	StoppedReason string
}

type Tally struct {
	Attempted    int
	Ran          int
	Invalid      int
	Completed    int // trials where the task was done at all
	Avoided      int // completed trials where the owner invariant passed
	Firings      int // hook firing records across the arm's trials
	Matches      int
	Injections   int
	Repeated     int
	Suppressed   int
	ContextBytes int
	MeanInput    int64
}

type Verdict struct {
	// TaskDone means the agent completed the visible task.
	TaskDone bool
	// Avoided means a task-complete solution also preserved the owner invariant.
	Avoided bool
	// Notes concisely explains the deterministic judgment.
	Notes string
}

type WorkflowArm string

type WorkflowArmReport struct {
	Attempted     int
	Valid         int
	TaskDone      int
	InvariantPass int
	WorkflowProcessCounts
	ContextTokens int64
	CostUSD       float64
}

type WorkflowAssessment struct {
	ID                  string
	Definition          WorkflowClaim
	Status              string
	Reason              string
	Reasons             []string
	QualifyingInstances int
	MeanEffect          float64
	WorstInstanceEffect float64
	HarmfulInterference float64
}

type WorkflowClaim struct {
	ID                           string   `yaml:"id"`
	Claim                        string   `yaml:"claim"`
	PrimaryMetric                string   `yaml:"primary_metric"`
	Comparison                   string   `yaml:"comparison"`
	Direction                    string   `yaml:"direction"`
	RequiredModel                string   `yaml:"required_model"`
	RequiredEffort               string   `yaml:"required_effort"`
	RequireCleanSeamark          bool     `yaml:"require_clean_seamark"`
	MinimumEffect                float64  `yaml:"minimum_effect"`
	MinimumInstanceEffect        float64  `yaml:"minimum_instance_effect"`
	MaximumHarmfulInterference   float64  `yaml:"maximum_harmful_interference"`
	MinimumInstances             int      `yaml:"minimum_instances"`
	MinimumValidPairsPerInstance int      `yaml:"minimum_valid_pairs_per_instance"`
	Instances                    []string `yaml:"instances"`
	ProcessMetrics               []string `yaml:"process_metrics"`
}

type WorkflowClaimRegistry struct {
	SchemaVersion int                `yaml:"schema_version"`
	Claims        []WorkflowClaim    `yaml:"claims"`
	Activation    ActivationCriteria `yaml:"activation"`
}

type WorkflowCohort struct {
	Instance         string
	Fingerprint      string
	Task             string
	TaskSHA          string
	Trigger          string
	Companion        string
	Fixture          string
	RequestedModel   string
	Model            string
	SeamarkVersion   string
	SeamarkSHA       string
	AgentVersion     string
	Effort           string
	RuntimeID        string
	MaxBudgetUSD     float64
	ValidPairs       int
	FavorablePairs   int
	UnfavorablePairs int
	TiedPairs        int
	HarmfulPairs     int
	Skills           WorkflowArmReport
	Only             WorkflowArmReport
	// TranscriptDirs lists where the cohort's transcripts live, so a reader
	// can find the evidence behind a row.
	TranscriptDirs []string
}

type WorkflowConfig struct {
	Trials   int
	Arms     []WorkflowArm    // arms to run; nil means both
	Instance WorkflowInstance // required; there is no default instance
	// AgentArgv is the agent command per arm. The arms differ in the tools
	// they expose, so each arm has its own command line. The runner appends
	// the trial's MCP configuration and then the task prompt.
	AgentArgv     map[WorkflowArm][]string
	SeamarkBin    string        // absolute path to the seamark binary (MCP server + index)
	Timeout       time.Duration // per-trial agent timeout; 0 means 10 minutes
	Out           string        // results JSONL path, appended one row per trial
	WorkDir       string        // parent for trial dirs; "" means a fresh temp dir
	Keep          bool          // keep trial dirs after judging, for inspection
	TranscriptDir string        // saves each trial's raw agent output; empty disables
	// PrepareIndex copies the seamark binary into the trial and indexes the
	// fixture, so the MCP server answers from a ready index. Hermetic tests
	// turn it off and keep an inert fake path.
	PrepareIndex bool
	Version      string // seamark version stamped into rows
	SeamarkSHA   string // exact binary digest stamped into rows
	AgentVersion string // agent CLI version stamped into rows
	Model        string // exact requested primary model; empty for custom agents
	Effort       string // requested effort level
	MaxBudgetUSD float64
	RuntimeID    string // sandbox/toolchain identity
	Fingerprint  string // immutable instance + runtime configuration hash
	// RunID groups rows and makes transcript names unique across concurrent
	// invocations. Empty asks RunWorkflow to generate a random ID.
	RunID string
	// MaxTurns caps activation sessions (RunActivation) with --max-turns.
	// Zero leaves the agent's default. Workflow trials never use it.
	MaxTurns int
	// RequireStructuredResult and RequireExpectedInit are true for the
	// managed Claude adapter. The second is the counterpart of the lessons
	// harness's clean-init rule: here the init record must show exactly the
	// seamark MCP server, the arm's tool set, and, in the skills arm, exactly
	// the three shipped skills.
	RequireStructuredResult bool
	RequireExpectedInit     bool
	Log                     func(string, ...any) // progress lines; nil silences
}

type WorkflowInstance struct {
	Instance

	// Trigger is the repository-relative file the visible task changes
	// first. The preflight asks `seamark why <trigger>` to name Companion.
	Trigger string
	// Companion is the repository-relative file the owner invariant
	// requires and the naive solution forgets.
	Companion string
}

type WorkflowProcessCounts struct {
	ChangeSetFirst int // change_set ran before the first edit
	CompanionNamed int // change_set named the companion
	NamedByCheck   int // check named the companion the diff left out
	Opened         int // the agent opened the companion after it was named
	WhyFollowed    int // why followed the named companion
	CheckLast      int // check ran after the last edit
	SeamarkCalls   int
	Activations    map[string]int
}

type WorkflowReport struct {
	ResultSchemaVersion int
	ClaimSchemaVersion  int
	EvidenceFrom        string
	EvidenceTo          string
	Inputs              []ReportInput
	Cohorts             []WorkflowCohort
	Assessments         []WorkflowAssessment
	Activation          *ActivationReport
}

type WorkflowRow struct {
	SchemaVersion int         `json:"schema_version"`
	TS            string      `json:"ts"`
	RunID         string      `json:"run_id"`
	Instance      string      `json:"instance"`
	TaskSHA       string      `json:"task_sha256"`
	Trigger       string      `json:"trigger"`
	Companion     string      `json:"companion"`
	Arm           WorkflowArm `json:"arm"`
	Trial         int         `json:"trial"`
	// Fixture is the generated repo's full HEAD commit, so rows from
	// different fixture versions are never pooled as one series.
	Fixture     string `json:"fixture"`
	Fingerprint string `json:"fingerprint"`

	// Valid says the agent session and the arm's wiring were delivered.
	// PairValid additionally says every arm in this trial number was valid.
	// Only rows satisfying both enter effect tallies.
	Valid                 bool   `json:"valid"`
	PairValid             bool   `json:"pair_valid"`
	InvalidReason         string `json:"invalid_reason,omitempty"`
	InfrastructureFailure bool   `json:"infrastructure_failure,omitempty"`

	// The init record proves what the agent had: the arm's tool set, the
	// seamark MCP server, the skills, and no plugins.
	InitSeen   bool             `json:"init_seen,omitempty"`
	ResultSeen bool             `json:"result_seen,omitempty"`
	Tools      []string         `json:"tools,omitempty"`
	MCPServers []MCPServerState `json:"mcp_servers,omitempty"`
	Skills     []string         `json:"skills,omitempty"`
	Plugins    []string         `json:"plugins,omitempty"`
	// DeniedTools lists the tools the agent asked for and was refused, from
	// the result record. A refused seamark tool or Skill tool means the arm's
	// approvals were not in effect, so the row is invalid.
	DeniedTools []string `json:"denied_tools,omitempty"`

	WorkflowTrace

	TaskDone bool   `json:"task_pass"`
	Avoided  bool   `json:"invariant_pass"`
	Notes    string `json:"notes,omitempty"`
	// Checks are public repository-local validation commands. The hidden
	// task and invariant judges are TaskDone and Avoided above.
	ChecksPass bool          `json:"checks_pass"`
	Checks     []CheckResult `json:"checks,omitempty"`

	AgentUsage

	SeamarkVersion string  `json:"seamark_version,omitempty"`
	SeamarkSHA     string  `json:"seamark_sha256,omitempty"`
	AgentVersion   string  `json:"agent_version,omitempty"`
	Effort         string  `json:"effort,omitempty"`
	MaxBudgetUSD   float64 `json:"max_budget_usd,omitempty"`
	RuntimeID      string  `json:"runtime_id,omitempty"`

	// Transcript, stderr, and patch keep the audit record with digests, so a
	// verdict stays checkable after the trial dir is deleted.
	Transcript    string `json:"transcript,omitempty"`
	TranscriptSHA string `json:"transcript_sha256,omitempty"`
	StderrLog     string `json:"stderr,omitempty"`
	StderrSHA     string `json:"stderr_sha256,omitempty"`
	Patch         string `json:"patch,omitempty"`
	PatchSHA      string `json:"patch_sha256,omitempty"`
}

type WorkflowSummary struct {
	Rows          []WorkflowRow
	ByArm         map[WorkflowArm]WorkflowTally
	Instance      string
	StoppedReason string
}

type WorkflowTally struct {
	Ran       int
	Invalid   int
	Completed int // trials where the task was done at all
	Avoided   int // completed trials where the owner invariant passed
	WorkflowProcessCounts
	MeanInput int64
	CostUSD   float64
}

type WorkflowTrace struct {
	// ChangeSetBeforeFirstEdit is true when a change_set call preceded the
	// first Edit or Write, or when the agent never edited at all.
	ChangeSetBeforeFirstEdit bool `json:"change_set_before_first_edit"`
	// ChangeSetFiles lists the files of the first change_set call.
	ChangeSetFiles []string `json:"change_set_files,omitempty"`
	// CompanionNamedByChangeSet is true when a change_set result named the
	// companion file that the agent had not put in the call itself.
	CompanionNamedByChangeSet bool `json:"companion_named_by_change_set"`
	// CompanionNamedByCheck is true when a check result listed the companion
	// under "history suggests also reviewing": the diff left it out and
	// history named it. The unindexed-files note also quotes paths, so only
	// that section counts.
	CompanionNamedByCheck bool `json:"companion_named_by_check"`
	// WhyFollowedCompanion is true when a why call whose query is the
	// companion path followed the change_set or check that named it. A
	// query by symbol or by bare file name does not count; the rule is the
	// same in both arms.
	WhyFollowedCompanion bool `json:"why_followed_companion"`
	// CompanionOpenedAfterNamed is true when a Read, Edit, Write, or
	// MultiEdit on the companion followed the call that named it. The first
	// cohort showed this is the step that decides the outcome: a named
	// companion the agent never opened was never acted on.
	CompanionOpenedAfterNamed bool `json:"companion_opened_after_named"`
	// CheckAfterLastEdit is true when a check call followed the last edit.
	CheckAfterLastEdit bool `json:"check_after_last_edit"`
	// CheckVerdict is the verdict line of the last check result, for
	// example "allow (mode: warn)"; empty when the call failed.
	CheckVerdict string `json:"check_verdict,omitempty"`
	// Activations lists the skills the agent loaded, in first-use order.
	Activations []string `json:"activations,omitempty"`
	// SeamarkCalls counts every seamark MCP tool call; SeamarkToolCalls
	// splits the count by tool name.
	SeamarkCalls     int            `json:"seamark_calls"`
	SeamarkToolCalls map[string]int `json:"seamark_tool_calls,omitempty"`
	// Edits counts Edit, Write, and MultiEdit calls. Edits made through
	// Bash are invisible here; the patch stays the record of what changed.
	Edits int `json:"edits"`
	// FirstEditSeq is the position of the first edit among all tool calls,
	// counted from 1; zero when there was no edit.
	FirstEditSeq int `json:"first_edit_seq,omitempty"`
}