rustledger-ops

Pure operations on beancount directives — dedup, categorize, reconcile.

Overview

This crate provides reusable functions for transforming and analyzing collections of beancount directives. All operations are pure — they take directives in and return results out, with no I/O or framework coupling.

Analogous to Python beancount's ops/ module.

Modules

Module	Description
`fingerprint`	Structural hashing and stable fingerprinting of transactions
`dedup`	Duplicate detection (structural, fuzzy, and fingerprint-based)
`categorize`	Rule-based and pattern-based account categorization
`merchants`	Merchant name normalization and lookup dictionary
`enrichment`	Shared types for operation results (confidence, method, alternatives)
`ml`	ML-based categorization (TF-IDF + Naive Bayes); host-only, absent on wasm targets
`transfer`	Transfer detection between own accounts
`reconcile`	Reconciliation of imported vs. existing directives

Usage

use rustledger_ops::dedup::find_structural_duplicates;
use rustledger_ops::fingerprint::structural_hash;

// Find duplicate transactions in a directive list
let duplicates = find_structural_duplicates(&directives);

// Compute a structural hash of a single transaction
let hash = structural_hash("2024-01-15", &transaction_data);

Design

Operations in this crate depend on rustledger-plugin-types (for DirectiveWrapper and related types) plus utility crates (rust_decimal, jiff, blake3, regex, and a host-only ML stack for the ml module). They know nothing about the plugin runtime, CLI, LSP, or import system.

This separation allows the same operations to be used by:

The plugin system (via thin NativePlugin wrappers)
The import pipeline (for dedup and categorization)
CLI commands (e.g., rledger dedup)
The LSP (for diagnostics and code actions)

License