outcomes

package

v0.12.0 Latest Latest Go to latest Published: Oct 23, 2026 License: MIT Imports: 16 Imported by: 0

Details

Valid go.mod file
Redistributable license
Tagged version
Stable version
Learn more about best practices

Repository

github.com/spfuzzylink/questlock

Links

Open Source Insights

Documentation ¶

Overview ¶

Package outcomes records human acceptance separately from deterministic inference checks. Records contain hashes and accounting metadata, never prompts, generated answers, reviewer names, or free-form feedback.

Constants ¶

View Source

const (
	MaxRuns        = 256
	MaxEvents      = 256
	MaxCohortTasks = 256
)

Variables ¶

View Source

var (
	ErrInvalid    = errors.New("invalid outcome record")
	ErrConflict   = errors.New("outcome version conflict")
	ErrIDConflict = errors.New("outcome ID already has a different payload")
	ErrLimit      = errors.New("outcome ledger limit reached")
)

Functions ¶

func Key ¶

func Key(task Task) (string, error)

Key uses an unambiguous JSON tuple, so no delimiter in an ID can collide with a different task or revision. Broker authentication supplies the scope.

Types ¶

type Attempt ¶

type Attempt struct {
	TierID           string          `json:"tier_id,omitempty"`
	Fingerprint      string          `json:"fingerprint,omitempty"`
	OutputSHA256     string          `json:"output_sha256,omitempty"`
	CheckPassed      bool            `json:"check_passed"` // Observed checks, independently of run admission/durability.
	Complete         bool            `json:"complete"`
	Durable          bool            `json:"durable"`
	Reused           bool            `json:"reused"`
	GenerationCalled bool            `json:"generation_called"`
	UsageKnown       bool            `json:"usage_known"`
	Usage            inference.Usage `json:"usage"`
	EstimatedCostUSD *float64        `json:"estimated_cost_usd"`
	CostBasis        string          `json:"cost_basis"`
}

Attempt describes only the current invocation's usage and charge. A reused result has zero new usage/charge; its original cost is never added again. EstimatedCostUSD is a caller estimate, not a verified provider invoice.

type Feedback ¶

type Feedback struct {
	ID                  string   `json:"id"`
	RunID               string   `json:"run_id"`
	Kind                string   `json:"kind"`             // accepted, rewritten, rejected, reopened
	Reason              string   `json:"reason,omitempty"` // rewrite: correctness, style, requirements, other
	FinalArtifactSHA256 string   `json:"final_artifact_sha256,omitempty"`
	HumanMinutes        *float64 `json:"human_minutes"`
}

Feedback is explicitly human-asserted. A final accepted artifact can differ from the generated output after human correction. HumanMinutes is incremental effort for this event: nil means unknown, zero means explicitly reported zero.

type Receipt ¶

type Receipt struct {
	Task      Task      `json:"task"`
	Key       string    `json:"key"`
	ID        string    `json:"id"`
	Version   int64     `json:"version"`
	CreatedAt time.Time `json:"created_at"`
}

type RecordedFeedback ¶

type RecordedFeedback struct {
	Feedback
	ExpectedVersion int64     `json:"expected_version"`
	Version         int64     `json:"version"`
	CreatedAt       time.Time `json:"created_at"`
}

type RecordedRun ¶

type RecordedRun struct {
	Run
	Version   int64     `json:"version"`
	CreatedAt time.Time `json:"created_at"`
}

type Report ¶

type Report struct {
	Tasks                           []TaskSummary `json:"tasks"`
	TaskCount                       int           `json:"task_count"`
	AcceptedTasks                   int           `json:"accepted_tasks"`
	RunCount                        int           `json:"run_count"`
	AttemptCount                    int           `json:"attempt_count"`
	KnownInferenceSubtotalUSD       float64       `json:"known_inference_subtotal_usd"`
	MissingInferenceCosts           int           `json:"missing_inference_costs"`
	MissingUsage                    int           `json:"missing_usage"`
	RecordedHumanMinutes            float64       `json:"recorded_human_minutes"`
	MissingHumanEffort              int           `json:"missing_human_effort"`
	HourlyRateUSD                   *float64      `json:"hourly_rate_usd"`
	InferenceCostUSD                *float64      `json:"inference_cost_usd"`
	HumanCostUSD                    *float64      `json:"human_cost_usd"`
	TotalCostUSD                    *float64      `json:"total_cost_usd"`
	CostPerAcceptedTaskUSD          *float64      `json:"cost_per_accepted_task_usd"`
	InferenceCostPerAcceptedTaskUSD *float64      `json:"inference_cost_per_accepted_task_usd"`
	GeneratedAt                     time.Time     `json:"generated_at"`
	Basis                           string        `json:"basis"`
	Coverage                        string        `json:"coverage"`
}

Report covers only the explicit cohort's recorded attempts and feedback. Nullable metrics remain unknown when a contributing value is unknown; no external work, hardware cost, or surrounding assistant usage is inferred. Cohort reads are individually versioned, not one atomic multi-task snapshot.

type Run ¶

type Run struct {
	ID            string    `json:"id"`
	RequestDigest string    `json:"request_digest"`
	Status        string    `json:"status"` // completed, quality_failed, stopped
	Attempts      []Attempt `json:"attempts"`
}

type Snapshot ¶

type Snapshot struct {
	Task      Task               `json:"task"`
	Key       string             `json:"key"`
	Version   int64              `json:"version"`
	CreatedAt time.Time          `json:"created_at"`
	Runs      []RecordedRun      `json:"runs"`
	Events    []RecordedFeedback `json:"events"`
	State     State              `json:"state"`
}

type State ¶

type State struct {
	Status              string `json:"status"` // unreviewed, accepted, rewritten, rejected, reopened
	AcceptedRunID       string `json:"accepted_run_id,omitempty"`
	FinalArtifactSHA256 string `json:"final_artifact_sha256,omitempty"`
}

type Store ¶

type Store struct {
	Broker inference.Broker
}

func (Store) Feedback ¶

func (s Store) Feedback(ctx context.Context, task Task, expectedVersion int64, feedback Feedback) (Receipt, error)

Feedback requires the version the operator actually reviewed. It never rebases a new event onto a changed ledger. Exact event retries remain valid after later events and preserve the original receipt and state history.

func (Store) Get ¶

func (s Store) Get(ctx context.Context, task Task) (Snapshot, error)

func (Store) Record ¶

func (s Store) Record(ctx context.Context, task Task, run Run) (Receipt, error)

Record appends a run once. Concurrent independent runs retry only broker version conflicts; an uncertain transport/publication outcome is returned to the caller without retrying. Repeating the same run ID and payload later, including from another token in the same scope, returns its original receipt.

func (Store) Report ¶

func (s Store) Report(ctx context.Context, tasks []Task, hourlyRateUSD *float64) (Report, error)

Report never scans a possibly truncated artifact listing. The caller chooses a bounded explicit cohort, whose entries must all have registered ledgers. Every recorded attempt contributes to the numerator, including failed, rejected, reopened, and still-unreviewed tasks. Only current explicit human acceptance contributes to the denominator.

type Task ¶

type Task struct {
	ID       string `json:"id"`
	Revision string `json:"revision"`
}

Task revisions are independent cohorts. Changing a revision never carries human acceptance from the earlier task artifact.

type TaskSummary ¶

type TaskSummary struct {
	Task                      Task    `json:"task"`
	Version                   int64   `json:"version"`
	State                     State   `json:"state"`
	RunCount                  int     `json:"run_count"`
	AttemptCount              int     `json:"attempt_count"`
	KnownInferenceSubtotalUSD float64 `json:"known_inference_subtotal_usd"`
	MissingInferenceCosts     int     `json:"missing_inference_costs"`
	MissingUsage              int     `json:"missing_usage"`
	RecordedHumanMinutes      float64 `json:"recorded_human_minutes"`
	MissingHumanEffort        int     `json:"missing_human_effort"`
}

Source Files ¶

View all Source files

?	: This menu
/	: Search site
f or F	: Jump to
y or Y	: Canonical URL