Skip to content

Repository files navigation

FastDiff V2

Exact reconciliation for large datasets. Bounded-memory processing. Results you can replay.

FastDiff V2 is a deterministic reconciliation system for comparing two data states by a declared key and selected typed fields. It turns large, unsorted exports into exact CHANGED, ONLY_A and ONLY_B results, with schema differences, fixed-point financial totals and a retained evidence bundle for offline verification.

External merge sorting and spill-to-disk let the Extended workflow process inputs larger than its configured memory budget. A local, binary-first deployment model keeps sensitive datasets and reconciliation artifacts under the operator's control.

One workflow, from source to verified result

Declared sources + schema + key
              |
      Acquire / normalize
              |
  External sort / spill to disk
              |
       Reconcile A and B
              |
   Delta + manifest + snapshots
              |
       Offline verification
Capability What it delivers
Exact row reconciliation Keyed CHANGED, ONLY_A, ONLY_B and IDENTICAL results over selected normalized fields
Large, unsorted inputs Bounded chunks, external merge sorting and disk spill, with an explicit process memory budget
Acquisition and normalization Configured CSV/JSON sources, optional HTTP GET / read-only GraphQL GET, explicit field mappings and retained snapshots
Schema comparison Field, type, nullability and key differences between explicitly declared schemas
Financial reconciliation Exact fixed-point amounts and grouped totals by account, currency or other declared dimensions
CI fast-fail Full input validation and sorting, then row comparison up to the configured difference limit
Deterministic evidence Result artifacts, manifests and SHA-256 hashes that bind the retained bundle for integrity checks and replay
Offline verification Local integrity checking and result reconstruction from retained inputs

Precise answers to operational questions

Reconcile inventory snapshots, validate migration exports, compare ledger states or track changes in explicitly mapped blockchain-derived data. Define the entity identity once, select the fields that matter, and obtain an exact delta within that comparison contract.

Result Meaning
IDENTICAL The same key has equal selected, normalized fields
CHANGED The same key has different selected, normalized fields
ONLY_A The key exists only in A
ONLY_B The key exists only in B

Extended rejects duplicate, empty and null canonical keys, including collisions introduced by normalization. Row results, schema differences and financial totals remain distinct so each answers its own question.

A focused product architecture

Component Role
Core Streaming comparison of already sorted files
Connectors Source acquisition, explicit normalization and retained snapshots
Extended CSV, TSV, JSON, JSONL, canonical connector CSV and read-only SQLite exports; external sort, schema comparison, financial reconciliation and CI

FastDiff is designed for proprietary binary delivery and local operation. Local-file comparison and offline replay keep datasets on the operator's machine; optional acquisition contacts the endpoints the operator configures. See data handling for bundle retention and access controls.

Explore the documentation

Scope / Current release status

This release provides documentation and synthetic examples. No production binary is currently published in FastDiff-2. Binary delivery awaits final executable qualification, approved signing/trust, production licensing and commercial terms. V2 identifies the product generation; component versions are listed in the release notes.

  • Formats: the documented version supports the inputs listed above. Native Parquet/Avro/ORC and PostgreSQL/MySQL adapters are outside this version's scope.
  • Scale: external sorting supports data beyond the configured memory budget. The current Windows CLI has a 64–1024 MiB process budget and a ten-minute operation deadline; achievable input size depends on records, I/O and available disk. See resource limits.
  • Evidence: hashes establish byte integrity; offline replay is product verification, not third-party certification. Retain a trusted manifest hash separately from the bundle.
  • Source consistency: network pagination is not an atomic snapshot. Use immutable exports when snapshot consistency matters; network acquisition has separate limits.

The contract defines the complete interpretation of results. Earlier releases in FastDiff retain their own versions and terms.

Copyright (c) 2026 cxclrfx. See LICENSE.txt.

About

Deterministic reconciliation for large datasets: exact row differences, schema comparison, financial totals, bounded-memory external sorting and offline replay. Local, binary-first architecture.

Topics

Resources

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors