style

package
v0.4.1 Latest Latest
Warning

This package is not in the latest version of its module.

Go to latest
Published: Jul 25, 2026 License: MIT Imports: 17 Imported by: 0

Documentation

Overview

Package style implements learning pipeline B: periodic synthesis of the per-session style observations that the miner appends to state/style_observations.jsonl. Observations are raw signal, never cards — a card is synthesized only for a preference with at least three consistent observations across at least two sessions (structurally one-off-proof, plan §B). Zero qualifying groups ⇒ zero LLM calls.

Index

Constants

This section is empty.

Variables

This section is empty.

Functions

This section is empty.

Types

type Observation

type Observation struct {
	TS          string `json:"ts"`
	SessionID   string `json:"session_id"`
	Repo        string `json:"repo"`
	Dimension   string `json:"dimension"`
	Observation string `json:"observation"`
	Evidence    string `json:"evidence"`
}

Observation is one ledger row (written by internal/learn/mine).

type Result

type Result struct {
	Ran        bool // false: trigger not due or no qualifying groups
	Groups     int
	Created    []string
	Reinforced []string
	Confirmed  []string
	Retired    []string // contradicted culi-authored cards
	Skipped    []string
	Notes      []string
	Usage      llmgen.Usage
}

Result reports one synthesis run.

type Synthesizer

type Synthesizer struct {
	Base  string
	Store *store.Store
	Tier  *llmtier.Tier
}

Synthesizer wires pipeline B's dependencies.

func (*Synthesizer) Run

func (s *Synthesizer) Run(ctx context.Context, force bool, now time.Time) (Result, error)

Run fires one policy-gated synthesis. force bypasses the trigger (not the qualifying-group gate — without consistent observations there is nothing to synthesize). The clock is injected (14.45).

Jump to

Keyboard shortcuts

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