Machine Edition Developer Kit (v0.1)

Reference developer kit, validation engine, parser, query suite, evaluation benchmark (ME-BENCH-001), and research trial execution packages implementing the 5 core responsibilities defined in Machine Edition Specification v0.1:

inspect → validate → parse → query → compare

---

1. What is this?

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.

This Developer Kit serves as the canonical open reference implementation of the public Machine Edition Specification v0.1, demonstrating that Machine Edition packages can be inspected, validated, queried, and evaluated without proprietary dependencies.

• Conceptual Authority: WinMedia (https://winmedia.com/machine-editions/specification/v0.1)
• Executable Implementation: GitHub (lynnmedia/machine-edition-devkit)
• Evaluation Benchmark: ME-BENCH-001 (winmedia.machine-edition-representation-benchmark.v0.1)
• Research Trials:
  • ME-RES-001: Deterministic reference-harness qualification trial
  • ME-RES-002: Controlled real generative model trial (qwen2.5:0.5b)

---

2. Quick Start

Installation

git clone https://github.com/lynnmedia/machine-edition-devkit.git
cd machine-edition-devkit
pip install -e ".[dev]"

Reference Specimen

An authoritative public reference specimen derived from authorized public companion material is located at:

specimen/
  srow/
    package/                     <-- Normative Machine Edition Package
      manifest.json
      meaning-units.jsonl
      provenance.jsonl
      definitions.jsonl
      boundaries.jsonl
      relationships.jsonl
      full-preview.md
      LICENSE.txt
    SOURCE.json                  <-- Authority provenance & archive SHA-256
    DERIVATION.json              <-- Normalization log
    CONFORMANCE-CROSSWALK.md     <-- Audit crosswalk against Spec v0.1 C1-C7

---

3. Validate Specimen

Inspect and validate package compliance against Machine Edition Specification v0.1 JSON schemas (C1-C7) and structural invariants:

from pathlib import Path
from machine_edition_devkit.inspect import inspect_package
from machine_edition_devkit.validate import MachineEditionValidator

specimen_dir = Path("specimen/srow/package")

# 1. Inspect
summary = inspect_package(specimen_dir)
print(f"INSPECT: {summary.package_id} v{summary.version} ({summary.meaning_units_count} units)")

# 2. Validate against C1-C7 schemas and invariants
validator = MachineEditionValidator()
report = validator.validate_package(specimen_dir)
print(f"VALIDATE: {report.outcome} (Errors: {len(report.errors)})")

---

4. Parse & Query Specimen

Load the edition into a structured entity model and perform deterministic queries with provenance tracking:

from machine_edition_devkit.parse import MachineEdition

edition = MachineEdition.load("specimen/srow/package", validate=True)
unit = edition.get_unit("srow.ref.mu.003")
print(f"PARSE: Loaded '{unit.title}' [Resolution Level L{unit.resolution_level}]")

# Provenance tracing
prov = edition.provenance(unit)
print(f"PROVENANCE: {prov.source_title} ({prov.source_url})")

# Typed relationships
rels = edition.relationships_for(unit)
for r in rels:
    print(f"RELATIONSHIP: {r.subject} --[{r.predicate}]--> {r.object}")

CLI Interface:

# Run the 20-query reference pack
python -m machine_edition_devkit.queries run-all

---

5. Run Representation Comparison

Compare Machine Edition against PDF, EPUB, and Naive RAG representation formats across a 16-task representation matrix:

# Run comparison trial
python -m machine_edition_devkit.comparison run

# Display representation property matrix
python -m machine_edition_devkit.comparison matrix

---

6. Reproduce Benchmark (ME-BENCH v0.1)

ME-BENCH-001 is a frozen research benchmark comparing PDF, EPUB, Naive RAG, and Machine Edition across 40 tasks (8 calibration, 32 evaluation across 8 task families) over a 16-fact source corpus under guaranteed 100% information parity.

# Verify artifact integrity and 16/16 information parity
python -m machine_edition_devkit.benchmark verify

# Run synthetic offline scorer test fixtures
python -m machine_edition_devkit.benchmark test-scorer

---

7. Read Research

ME-RES-001: Deterministic Reference-Harness Qualification Trial

• Purpose: Methodological qualification of the benchmark harness, 4 representation adapters, offline scoring engine, and 10,000-resample paired bootstrap pipeline.
• Classification: deterministic reference-harness trial (ME_RES_V01_REFERENCE_HARNESS_CONFIRMED).
• Report: research/me-res-001/report/ME-RES-001-REPORT.md

ME-RES-002: Real Generative Model Representation Trial

• Purpose: Controlled empirical evaluation using a genuine pretrained neural language model (qwen2.5:0.5b, Qwen 2.5 0.5B Instruct, 490M parameters via local Ollama) across 384 evaluation calls.
• Key Findings:
  • Machine Edition produced the highest provenance-completeness point estimate (0.8490 vs 0.7812 for RAG, paired delta +0.0677, 95% CI [-0.0573, +0.1927]) and reduced provenance omissions from 21 to 12.
  • Machine Edition produced 5x more clean error-free responses than RAG (15 passes vs 3).
  • 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 packages required higher token volume (~2,789 tokens vs ~415 for RAG).
• Report: research/me-res-002/report/ME-RES-002-REPORT.md

# Verify ME-RES-002 research integrity
python -m machine_edition_devkit.research.me_res_002 verify

# Display statistical contrast tables and calibrated hypothesis conclusions
python -m machine_edition_devkit.research.me_res_002 analyze

---

8. 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}
}

---

9. License

MIT License (c) 2026 Lynn Media.