CITATION.cff

cff-version: 1.2.0
message: >-
  If you use Lesscode, its research records or its experiment apparatus,
  please cite it using the metadata below. The Zenodo concept DOI resolves
  to the latest archived release.
type: software
title: "Lesscode: measured deconvolution of AI-generated codebases"
authors:
  - family-names: Ciepluch
    given-names: Marius
version: 1.0.2
date-released: 2026-11-10
doi: 10.5281/zenodo.23958363
identifiers:
  - type: doi
    value: 10.5281/zenodo.23958363
    description: Zenodo concept DOI, resolves to the latest archived release
  - type: doi
    value: 10.5281/zenodo.23958364
    description: Zenodo DOI of release v0.1.0
license: MIT
repository-code: https://github.com/norandom/lesscode
url: https://github.com/norandom/lesscode
abstract: >-
  Lesscode is a research project on making AI-generated codebases smaller
  and less entangled while keeping their behaviour at the public surface.
  It records behaviour from the original under an effect-capturing sandbox,
  scores that recording by mutation, lets a language model regenerate one
  scope behind its derived public surface under a structural contract,
  accepts only on zero divergence, measures size and internal call edges,
  reviews independently, and searches the pure core for differences the
  recording cannot see. The repository holds the preregistered experiments,
  their receipts and the dated research log, a Kiro-style specification of
  the loop, and a native measurement core.
keywords:
  - code simplification
  - AI-generated code
  - behaviour preservation
  - differential testing
  - mutation testing
  - concolic execution
  - spec-driven development
  - research reproducibility