decision

Documentation

Index

• Constants
• func DefaultAlwaysAsk() []string
• func DefaultAlwaysSkip() []string
• func ExtractImplicitWhy(subject string) (string, string, float64)
• func FormatCalibration(r *CalibrationReport) string
• type CalibrationReport
  • func ComputeCalibration(store domain.LoreStore) (*CalibrationReport, error)
• type DecisionResult
• type Engine
  • func NewEngine(store domain.LoreStore, config EngineConfig) *Engine
  • func (e *Engine) Evaluate(ctx SignalContext) *DecisionResult
  • func (e *Engine) SetClock(fn func() time.Time)
• type EngineConfig
  • func DefaultConfig() EngineConfig
• type SignalContext
• type SignalScore
  • func FileValueSignal(files []string) SignalScore
  • func ScanDiffContent(diffContent string) SignalScore

Constants

const (
	DefaultThresholdFull      = 60
	DefaultThresholdReduced   = 35
	DefaultThresholdSuggest   = 15
	DefaultLearningMinCommits = 20
)

Default values for the Decision Engine. Exported so that config/defaults.go can reference them, avoiding value duplication across two files.

Functions

func DefaultAlwaysAsk

func DefaultAlwaysAsk() []string

DefaultAlwaysAsk returns the default always_ask types (fresh copy each call).

func DefaultAlwaysSkip

func DefaultAlwaysSkip() []string

DefaultAlwaysSkip returns the default always_skip types (fresh copy each call).

func ExtractImplicitWhy

func ExtractImplicitWhy(subject string) (string, string, float64)

ExtractImplicitWhy splits a commit subject into What and Why parts. Returns what, why, confidence. If no separator found, confidence is 0.

func FormatCalibration

func FormatCalibration(r *CalibrationReport) string

FormatCalibration returns a human-readable summary of calibration metrics.

Types

type CalibrationReport

type CalibrationReport struct {
	TotalCommits      int
	AutoSkipped       int
	SuggestSkipped    int
	AskReduced        int
	AskFull           int
	Documented        int
	FalseNegativeRate float64 // auto-skip then manually documented / total auto-skip
	FalsePositiveRate float64 // ask-full then skipped / total ask-full
	AskFullDocRate    float64 // documented / total ask-full
	AutoSkipRate      float64 // auto-skip / total commits
}

CalibrationReport holds quality metrics for the Decision Engine.

func ComputeCalibration

func ComputeCalibration(store domain.LoreStore) (*CalibrationReport, error)

ComputeCalibration calculates Decision Engine quality metrics from stored commits. Uses a proxy for false negatives: commits with decision='documented' and question_mode='none' indicate an auto-skip that was later resolved via `lore new --commit`.

type DecisionResult

type DecisionResult struct {
	Score                  int
	Action                 string // ask-full, ask-reduced, suggest-skip, auto-skip
	Confidence             float64
	Signals                []SignalScore
	PrefilledWhat          string
	PrefilledWhy           string
	PrefilledWhyConfidence float64
}

DecisionResult holds the scoring outcome for a commit.

type Engine

type Engine struct {
	// contains filtered or unexported fields
}

Engine evaluates commits using multi-signal scoring.

func NewEngine

func NewEngine(store domain.LoreStore, config EngineConfig) *Engine

NewEngine creates a Decision Engine.

func (*Engine) Evaluate

func (e *Engine) Evaluate(ctx SignalContext) *DecisionResult

Evaluate scores a commit and determines the documentation action.

func (*Engine) SetClock

func (e *Engine) SetClock(fn func() time.Time)

SetClock overrides the time source (for testing only).

type EngineConfig

type EngineConfig struct {
	ThresholdFull      int
	ThresholdReduced   int
	ThresholdSuggest   int
	AlwaysAsk          []string
	AlwaysSkip         []string
	CriticalScopes     []string
	Learning           bool
	LearningMinCommits int
}

EngineConfig holds Decision Engine configuration from .lorerc.

func DefaultConfig

func DefaultConfig() EngineConfig

DefaultConfig returns sensible defaults for the Decision Engine.

type SignalContext

type SignalContext struct {
	ConvType     string
	Scope        string
	Subject      string
	Message      string
	DiffContent  string
	FilesChanged []string
	LinesAdded   int
	LinesDeleted int
}

SignalContext contains all inputs needed for scoring a commit.

type SignalScore

type SignalScore struct {
	Name   string
	Input  string
	Score  int
	Reason string
}

SignalScore records an individual signal's contribution.

func FileValueSignal

func FileValueSignal(files []string) SignalScore

FileValueSignal scores modified files by their documentation value. High-value files (API, schema, migration, proto, graphql, security) get +5 each, capped at +15. If 100% of files are test/mock/testdata, score is -10.

func ScanDiffContent

func ScanDiffContent(diffContent string) SignalScore

ScanDiffContent analyses diff lines for security, API, infra, deletion, and tech-debt patterns. Each pattern scores at most once per commit (deduplication). Analysis is limited to the first 1000 lines of the diff for performance.