CITATION.cff

cff-version: 1.2.0
message: "If you use this format, please cite it."
title: "Bundle v1: A Multi-Modal Capture Format for Adversarial Research on Blockchain Validator Infrastructure"
version: 0.1.0
date-released: 2026-05-06
license: MIT
type: software
authors:
  - name: NullRabbit
    alias: NullRabbitLabs
    email: simon@nullrabbit.ai
repository-code: "https://github.com/NullRabbitLabs/nr-bundle-spec"
url: "https://github.com/NullRabbitLabs/nr-bundle-spec"
keywords:
  - bundle
  - multi-modal capture
  - blockchain
  - validator
  - adversarial research
  - security
  - dataset format
  - response amplification
  - compute amplification
  - cross-chain holdout
abstract: >-
  Bundle v1 specifies the on-disk shape of a single attack (or benign)
  run against a validator: a manifest, a packet capture, and four
  time-series Parquet modalities (host, app, protocol, responses) all
  keyed off a monotonic t_ns referenced to the manifest's started_at.
  The format is designed for adversarial research on blockchain
  validator infrastructure, with explicit support for cross-chain
  holdout via a chain-agnostic family_id taxonomy paired with
  chain-specific primitive_id labels. Provenance fidelity tiers
  (stub / proxy / lab / lab-tls-fronted / production-captured /
  production-derived) and target authorisation scope (self-owned /
  customer-authorised / public-mainnet-passive / synthetic) are
  required structured fields, gating downstream publication and
  dataset curation. Reference parsers in Python and Rust ship under
  MIT licence with byte-equivalent wire-format and genome_id
  guarantees verified by cross-language test vectors.