Canonical ReferenceSpecification v0.1Public Developer Kit

Machine Edition Reference Headquarters

The canonical authority surface for the Machine Edition architecture: normative specification, open-source reference implementation, public reference specimen, evaluation benchmark, and calibrated empirical research.

01. Definition

What is a Machine Edition?

A Machine Edition is a structured, computable publication package designed for reliable, auditable, and multi-resolution consumption by automated AI agents and software systems alongside human readers.

Governed Structure

Knowledge is decomposed into discrete, typed meaning units with explicit cryptographic checksums, deterministic boundaries, and typed relationships.

Multi-Resolution

Every concept is explicitly articulated across five discrete resolution tiers (L0 Essence to L4 Full Operational Specification), allowing machines to scale context without meaning loss.

Auditable Provenance

Every assertion, rule, and entity links back to verifiable origin records, source publications, and cryptographic signatures.

02. Core Contract

The 5 Reference Responsibilities

Machine Edition Specification v0.1 defines five core responsibilities that any conformant developer kit must implement without proprietary dependencies.

1

Inspect

inspect_package(path)

Examine package manifest, cryptographic checksums, metadata, and meaning unit inventories without modifying data.

2

Validate

MachineEditionValidator().validate_package(path)

Validate structural conformance against normative JSON Schemas C1–C7 and enforce intra-package referential invariants.

3

Parse

MachineEdition.load(path)

Load package JSONL records into a typed domain entity model with navigation across meaning units and resolution levels.

4

Query

SampleQueryRunner / edition.provenance()

Execute deterministic queries, extract resolution-scaled context windows, and traverse provenance ledgers.

5

Compare

ComparisonHarness().run_all()

Benchmark representation affordances against PDF, EPUB, and RAG under strict 100% information parity.

03. Representation Contrasts

How a Machine Edition Differs from Other Formats

Traditional digital publishing formats optimize for human visual display or brute-force token chunking. Machine Editions provide native computational affordances.

RepresentationPrimary NatureResolution HandlingProvenanceComputability
PDF DocumentVisual presentation layoutFixed layout; no explicit resolution levelsPage number / bounding boxLow (requires visual extraction heuristics)
EPUB PublicationReflowable human readingLinear text flow onlySection / chapter headingModerate (HTML DOM parsing)
Naive RAG ChunksArbitrary character/token slicesBlind chunking; destroys semantic hierarchyChunk index / document titleFragmentary (loss of boundary semantics)
Machine EditionGoverned computable knowledge packageExplicit discrete resolution levels (L0–L4)Deterministic unit & claim identifiersNative (JSON Schema C1–C7 validated)

04. Package Architecture

Anatomy of a Conformant Package (C1–C7)

A normative Machine Edition is distributed as a directory or ZIP archive containing seven canonical, validated JSONL and Markdown streams.

manifest.jsonC1

Package identity, version, cryptographic checksums, and dependency ledger.

meaning-units.jsonlC2

Atomic meaning units with explicit L0–L4 resolution depth and text content.

provenance.jsonlC3

Source publication citations, authority URLs, and transformation lineage.

definitions.jsonlC4

Formal conceptual glossary, semantic invariants, and domain terms.

boundaries.jsonlC5

Usage permissions, licensing scopes, and epistemic jurisdictional constraints.

relationships.jsonlC6

Typed semantic predicate links connecting meaning units and external entities.

full-preview.mdC7

Human-readable linearized reference preview of complete package contents.

LICENSE.txtLegal

Open or governed machine-readable licensing contract.

05. Developer Kit

Open-Source Reference Developer Kit

The Developer Kit provides a zero-dependency Python implementation to inspect, validate, parse, query, and evaluate Machine Editions.

lynnmedia/machine-edition-devkit

Repository: MIT License · Python 3.10+

View on GitHub →

# Install and inspect

git clone https://github.com/lynnmedia/machine-edition-devkit.git

cd machine-edition-devkit && pip install -e ".[dev]"

# Validate SROW specimen

python -m machine_edition_devkit.validate specimen/srow/package

06. Public Specimen

SROW Public Reference Specimen

An authoritative public specimen derived from authorized SROW Public Companion assets, published for validation and tooling compliance.

Public Reference Specimen

SROW Machine Edition Reference Specimen (v0.1.0)

SHA-256: ebe193fca0609de8e957d8e88e2a26bddb5fe6490e41a5be44f6bf05cad26151

Download Specimen (.zip) →

Governed Identity Notice

The free SROW Public Reference Specimen is derived from the authorized SROW Public Companion to demonstrate Specification v0.1 conformance. It is distinct from the full governed SROW Machine Edition package, preserving appropriate governance boundaries.

07. Evaluation Benchmark

ME-BENCH v0.1 Representation Benchmark

A frozen, representation-controlled research evaluation instrument comparing PDF, EPUB, Naive RAG, and Machine Edition across 40 tasks under 100% guaranteed information parity.

8 Evaluated Task Families

  • 1. Factual Retrieval: Extract atomic domain facts.
  • 2. Relationship Retrieval: Trace directional typed links.
  • 3. Hierarchy Preservation: Respect conceptual ordering.
  • 4. Provenance Tracing: Cite verified authority IDs.
  • 5. Boundary Constraints: Honor operational limitations.
  • 6. Ambiguity Handling: Detect under-specified terms.
  • 7. Multi-Resolution Retrieval: Query specific L0–L4 tiers.
  • 8. Unsupported Claim Detection: Refuse out-of-scope queries.

Offline Scoring Engine

Submissions are scored deterministically offline without LLM-as-a-judge subjectivity across 7 dimensions:

• correctness• provenance_completeness• unsupported_assertion_rate• invariant_preservation• relationship_accuracy• constraint_violations

Evaluates a 14-token granular failure taxonomy under a strict gold firewall.

08. Empirical Research

Scientific Research Trials & Calibrated Findings

Two formal research trials have been executed under preregistered protocols and 10,000-resample bootstrap statistical estimations.

ME-RES-001Reference Harness Qualification
ME_RES_V01_REFERENCE_HARNESS_CONFIRMED

Deterministic Reference-Harness Qualification Trial

Methodological qualification of benchmark harness, 4 representation adapters, scoring engine, and 10,000-resample paired bootstrap pipeline.

ME-RES-002Empirical Model Evaluation
ME_RES_002_PUBLICATION_CLAIMS_CALIBRATED

Real Generative Model Representation Trial

Model: qwen2.5:0.5b (Qwen 2.5 0.5B Instruct, 490M dense parameters) · 384 evaluation runs

Calibrated Empirical Findings

  • PDF full-text extraction achieved the highest factual correctness point estimate (0.2969 vs 0.2656 EPUB, 0.2344 ME, 0.2188 RAG).
  • Machine Edition achieved the highest provenance-completeness point estimate (0.8490 vs 0.7812 RAG, paired delta +0.0677) and reduced omissions (21 to 12). The 95% bootstrap CI [-0.0573, +0.1927] crossed zero (inconclusive).
  • Machine Edition generated 5x more zero-defect clean passes than Naive RAG (15 passes vs 3).
  • Relationship accuracy was identical (0.8750) across all four representations.
  • Unsupported-assertion rate was 0.0000 across all four conditions.
  • Machine Edition contexts required significantly higher token volume (~2,789 tokens vs ~415 for RAG).

Empirical Conclusion: ME-RES-002 showed that Machine Editions enabled small generative models to extract and cite native provenance identifiers and produce 5x more zero-defect structured passes than RAG, while PDF achieved the highest overall factual correctness point estimate under full-text extraction. The trial demonstrates representation-specific tradeoffs rather than universal superiority.

09. Limitations

Known Limitations & Threats to Validity

Transparent disclosure of methodological scope boundaries, technical assumptions, and potential conflicts.

Single-Model Scope

Empirical trial ME-RES-002 evaluated qwen2.5:0.5b (490M parameters). Results on frontier 70B+ LLMs remain to be evaluated.

Domain Scope

Evaluation was conducted on the SROW conceptual publishing domain; generalization to dense tabular or mathematical domains requires further study.

Context Token Overhead

Machine Edition packages require significantly higher token volume (~2,789 tokens vs ~415 for RAG), trading token efficiency for explicit schema governance.

Creator Conflict Disclosure

The authors of the Developer Kit designed the benchmark; threats are mitigated by offline scoring, deterministic fixtures, and cryptographic integrity manifests.

10. Governed Portfolio

The Six Released Machine Editions

Six governed Machine Edition packages are currently released with cryptographically verified package identities.

Releasedfree

SROW Machine Edition

Governed semantic transformation and invariant preservation across resolutions, audiences, genres, and uses.

Semantic transformation & invariant preservation

Releasedpaid

The Mandala Protocol Machine Edition

Structural and epistemic jurisdiction with typed handoff validation.

Structural / epistemic governance

Releasedpaid

Big Net Machine Edition

Dynamic relational traversal, evidence and state propagation, conflict-path detection, and blast-radius consequence.

Relational traversal & consequence

Releasedpaid

The Sanskrit Mandala Model Machine Edition

Deterministic semantic-responsibility and epistemic-authority release governance.

Semantic responsibility & release governance

Releasedpaid

Universal Knowledge Mandala Machine Edition

Deterministic governed domain specialization and conformance across Universal Core, Domain Profile, and Situated Instance.

Domain specialization & conformance

Releasedpaid

Cognitive Data Structures Machine Edition

Selective, boundary-specific exposure and validation of governance-relevant semantic distinctions and representation contracts.

Boundary-specific semantic governance

11. Resources & Citation

Verified Citation Topology

Active endpoints are cryptographically verified and publicly retrievable.

Verified Public Endpoints

specification · Verified

Machine Edition Specification v0.1

Normative public specification defining the 5 core responsibilities and C1–C7 structural conformance requirements.

code · Verified

Machine Edition Developer Kit Repository

Canonical open reference implementation (Python 3.10+) with zero proprietary dependencies.

release · Verified

Developer Kit Release v0.1.0

Public GitHub release containing reference engine, validation fixtures, and release archive bundles.

specimen · Verified

SROW Public Reference Specimen (.zip)

Authoritative public reference specimen derived from authorized SROW Public Companion assets for Spec v0.1.

SHA-256: ebe193fca0609de8

benchmark · Verified

ME-BENCH v0.1 Benchmark Bundle (.zip)

Complete frozen 40-task benchmark package, 16-fact 4-representation corpora, gold ledgers, and offline scorer fixtures.

SHA-256: 173b32ee1e2953e8

research · Verified

ME-RES-002 Empirical Evidence Bundle (.zip)

Complete 384-run empirical trial evidence package, raw model outputs, item scores, bootstrap contrasts, and report.

SHA-256: 14140dc7685ebbae

Archival & Repository Surfaces in Preparation

dataset · Pending Activation

Hugging Face Benchmark Dataset

Benchmark evaluation dataset for Hugging Face datasets hub.

Publication package staged under dist/hf-dataset/ (Account activation pending)

citation · Pending Activation

Zenodo Archival Software DOI

Persistent archival DOI for scholarly citation.

Archival descriptor (.zenodo.json) prepared (Account activation pending)

BibTeX Citation

@software{lynnmedia_medk_2026,
  author = {{Lynn Media}},
  title = {Machine Edition Developer Kit (v0.1)},
  year = {2026},
  url = {https://github.com/lynnmedia/machine-edition-devkit},
  note = {Implementing Machine Edition Specification v0.1, WinMedia}
}