MaximvsQMC/qmc-volume-i

Quantum Metaphysics of Connectedness — Volume I Training Dataset (v2.0)

Author and research-program originator: Maximvs Shlygin
Corpus period: 2025–2026
Version: 2.0 · Primary records: 100 · Medium essays: 99 · QRT primary records: 1
Canonical entry point: QMC-V1-067 — The Asthma of Reality’s Paradoxes
Previous published dataset record: Zenodo 23145624, Version 1.0
Hugging Face repository: MaximvsQMC/qmc-volume-i. A new Zenodo Version 2.0 DOI has not been assigned in this package.

License: The author has selected Creative Commons Attribution 4.0 International (CC BY 4.0) for the original author-controlled dataset materials distributed in this Hugging Face release. Attribute Maximvs Shlygin and retain source provenance. Third-party quoted material, external publications, and linked resources may carry separate rights; this license does not override them. See LICENSE.md and License and rights notes. Earlier frozen archival source metadata stating not_specified_in_source_materials reflects its original preparation and has not been silently rewritten.

Dataset summary

This repository provides a machine-readable, provenance-preserving representation of 100 full research records from Quantum Metaphysics of Connectedness (QMC), an open research program. The corpus contains essays spanning methodological and conceptual questions, cosmology and quantum physics, mathematics, biology, AI, social systems, economics, and ecology. It is intended for long-document search, retrieval-augmented generation (RAG), conceptual graph exploration, corpus studies, and testing AI systems' ability to distinguish established science, QMC-internal reasoning, hypotheses, and metaphors.

QMC is not represented as a completed empirical theory, a substitute for scientific proof, or a uniform source of validated scientific facts. Its methodology uses concepts such as invariant Z, comparison ~, unfolding ==>, retained difference, and time-dependent viability checks. A proposed correspondence is not an identity or a proof of causation.

The canonical body of every record has been preserved exactly from the Version 2.0 Markdown training dataset: no summarization, correction, title rewriting, or removal of formulas and ASCII schemes was applied to the full_text field. SHA-256 payload hashes verify the preservation.

Changes since published Version 1.0

The prior dataset contained 95 primary records (94 Medium essays and one QRT primary record). Version 2.0 contains 100 (99 Medium essays and one QRT). Five records were added, with their stable IDs:

Record	Primary work	New line of inquiry
QMC-V1-096	The Indifference Paradox	Addressability, indifference, persistence of connectedness.
QMC-V1-097	The Paradox of the Pauli Principle	Quantum-state admissibility and coexistence.
QMC-V1-098	The First Generation Paradox	Fermion generations and material distinctions.
QMC-V1-099	The Boson Paradox	Collective modes and bosonic interactions.
QMC-V1-100	The Transparency Paradox	Molecular water, optical transparency, chirality, chemical continuation, and cosmic comparison.

The 95 Version 1.0 records retain their source payloads. 14 QRT-related external mathematical Zenodo source records remain nested references, not 14 additional QMC primary essays. Version 2.0 has 12 primary domains and 115 curated conceptual dependency edges (79 in Version 1.0).

File layout

This web-upload-friendly repository uses unique filenames at the repository root, so browser uploads cannot accidentally overwrite two identically named train.jsonl files. The two JSONL exports are distinct and both must be present: qmc_full_text.jsonl and qmc_rag_chunks.jsonl.

Configurations and splits

The Hub will expose two non-overlapping representations of the same underlying 100 studies; they must not be counted as 1,382 independent original research works.

Configuration	Split	Rows	Unit	Recommended use
full_text (default)	train	100	Complete source essay	Citation-grounded reading, structured analysis, graph tasks.
rag_chunks	train	1,282	Lossless contiguous text span of an essay	Semantic indexing, RAG, long-document retrieval.

The name train here means one storage split, not a scientifically constructed training/validation/test partition or a benchmark. Do not report evaluation accuracy based on random splits of strongly cross-referencing studies without controlling document and concept overlap.

Every rag_chunks record has record_id, chunk_index, start_char, end_char, chunk_sha256, and the full record's body_sha256. Concatenating the chunks of a record in chunk_index order yields the exact original full_text value. The chunk boundaries favor blank lines and intact fenced code blocks. Chunks are not human semantic annotations and may exceed the target size for exceptionally long individual blocks.

How to load

After upload to the registered dataset repository:

from datasets import load_dataset

full = load_dataset("MaximvsQMC/qmc-volume-i", "full_text", split="train")
chunks = load_dataset("MaximvsQMC/qmc-volume-i", "rag_chunks", split="train")

assert len(full) == 100
assert len(chunks) == 1282
print(full[0]["record_id"], full[0]["title"])
print(chunks[0]["record_id"], chunks[0]["chunk_index"])

No custom dataset loader or remote code execution is required. Standard Hub JSONL support and the Dataset Card's YAML configs specify which files to load. HF Dataset Viewer may generate convenient Parquet exports from these JSONL configurations. This package also includes an optional local Parquet export script (requires pyarrow).

For offline use, datasets.load_dataset("json", data_files="qmc_full_text.jsonl", split="train") also works when datasets is installed.

Data fields

full_text contains the entire authorial text, and all source metadata fields are retained from the canonical corpus wrapper, including:

• IDs and ordering: record_id, source_sequence, training_order, dataset_version.
• Authorship and publication: title, author, source_kind, source_url, medium_id, source_date, URL and extraction status.
• Editorial grouping: primary_domain_id, primary_domain, domain_tags.
• Graph links: dependency_parents, dependency_children, explicit_title_crossrefs.
• Text-preservation metadata: text_diagrams_preserved, text_formulas_preserved, tables_preserved_as_text, body_char_count, body_word_count, body_line_count, body_sha256.
• Text: full_text (not rewritten, redacted, normalized, or synthesized).

A dependency edge represents a curated conceptual prerequisite, not external scientific proof. A record's primary domain is an editorial grouping, not a claim it belongs exclusively to that discipline. source_sequence preserves inherited corpus provenance and is not publication chronology. training_order indicates a methodological reading route, not a label describing model suitability.

A separate dependency_edges.tsv and domain_counts.json aid graph exploration. hf_manifest.json describes the transform and verification values.

Provenance, extraction and limitations

• QMC_Volume_I_Training_Dataset_Version_2.0.md is the normative canonical full-text Markdown input for this Hugging Face export; qmc_full_text.jsonl is its record-wise view.
• An author-supplied English Markdown file supplied the content of QMC-V1-100 in the canonical source, and its Medium URL documents its published source. Exact identity to the current public Medium HTML text is not asserted.
• Some earlier sources were obtained through web or PDF extraction; source-specific preservation indicators are in each record. This export does not regenerate missing image-only content.
• Language is predominantly English. This export does not perform passage-level language tagging, scientific verification, claim checking, PII filtering, or automated evidence classification.
• The nested 14 QRT external research sources are cited/referenced in the canonical corpus, not duplicated as full-text records.
• Source publication is not independent verification of mathematical proofs, biomedical advice, or QMC hypotheses.

Training considerations: This is long-form, non-instruction-tuned text with potentially philosophical, speculative, contested, or provisional assertions. It is not a curated fact-only corpus. Model developers should preserve article-level attribution, maintain explicit distinctions among scientific claims and speculative interpretations, implement appropriate evaluation and data governance, and follow applicable content rights.

Integrity and reproducibility

The provided exports were checked against every source body_sha256, the canonical concatenated-body SHA-256, row counts, graph reciprocity, and round-trip reconstruction of all chunk texts. The SHA-256 file covers repository artifacts, excluding the checksum file itself.

Rebuild or validate locally:

python build_dataset.py    # requires PyYAML
python validate_dataset.py # Python standard library only

See hf_manifest.json, dependency_edges.tsv, SHA256SUMS.txt, and UPLOAD_GUIDE.md.

Responsible use and citation

Attribution should include Maximvs Shlygin, Quantum Metaphysics of Connectedness — Volume I Training Dataset (Version 2.0), the canonical record ID and source_url of the essay cited, and the eventual deposited dataset version / DOI when available. Do not cite the prior Version 1.0 DOI as though it identified Version 2.0.

The purpose of the dataset is to allow reproducible examination of this open research program, not to elevate every included hypothesis to an established scientific statement.