You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

CatQualia Cross-Domain Structural Isomorphisms

153,017 rows of cross-domain structural mappings. Each row takes a mechanism from one domain (anime, biology, physics, economics, games, security, law) and maps its shape onto a buildable software-engineering construct, together with the rule that makes the mapping hold, the artifact it produces, and a failure_class naming how the mapping breaks when it is wrong.

The corpus exists to answer one question mechanically: when a structural analogy between two domains is claimed, what distinguishes a real mapping from a decorative one? Every row carries both the claimed mapping and the failure mode, so the corpus can be used to train or evaluate a discriminator, not just to generate analogies.

What is actually in it

Property Value
Rows 153,017
File catqualia_cross_domain_isomorphisms.jsonl (JSON Lines, one object per line)
Bytes 135,522,530 (135.5 MB)
Canonical path Trees-git/04_isomorphism/catqualia_cross_domain_isomorphisms.jsonl
Encoding UTF-8, one JSON object per line

Row count and byte count above were measured with wc -l and stat -c%s against the exact file staged in this repository, not against an intermediate copy.

Schema

Fields of the first record, read directly from the file:

Field Type Meaning
work string The source work or system the mechanism is taken from (e.g. an anime, a paper, a protocol)
mechanism string The mechanism in the source domain, described at the level of what it does
canon_rule string The domain-free rule the mechanism instantiates — the actual transferable content
shape string Structural description of the mechanism with domain vocabulary stripped out
artifact string The concrete software artifact this mapping builds
why string Why the mapping holds
build_hint string How to construct the artifact
failure_class string How the mapping fails when it is only superficially similar — the discriminator label
grade string Quality grade assigned to the row
provenance string Where the row came from
_corpus string Source corpus tag
_tier string Tier tag

failure_class is the load-bearing field. It is what makes this corpus different from a list of analogies: a row asserts not only that two mechanisms correspond but how the correspondence degrades, which is the signal a model needs in order to reject plausible but wrong transfers.

How it was produced

Generated by the CatQualia isomorphism engine: candidate mappings were produced from source material, then filtered through an IDF-weighted scoring gate (S_IDF over shared terms weighted by inverse document frequency), which separates genuine structural mappings from rows that share only the engine's own template vocabulary. The published method is described at https://catqualia.com/isomorphism/.

Rows here are the engine's own output. They are not human-verified as correct analogies, and a grade field is the engine's own assessment, not an external label.

Intended use

  • Training or evaluating a structural-transfer model (map a mechanism, produce an artifact).
  • Training a discriminator that separates genuine structural mappings from decorative ones.
  • Studying how failure modes distribute across source domains.

Known limitations

  • Synthetic. These are machine-generated mappings. They have not been validated by domain experts, and a plausible-sounding row may be wrong.
  • Self-graded. grade and _tier come from the same pipeline that generated the rows, so they are not independent quality labels. Treat agreement between source and derived data carefully.
  • English only.
  • Coverage is uneven. Some source domains contribute far more rows than others.
  • Duplication. This exact file also exists at 501_mirage_engine/container/external/04_isomorphism/catqualia_cross_domain_isomorphisms.jsonl inside the author's own tree. This repository is the canonical published copy.

Licensing and attribution

Released under CC BY 4.0. The author's own licence for the wider corpus (CSIL v3.0) also permits research use; CC BY 4.0 is applied here so the dataset can be used without ambiguity. Attribution: Christopher Betances, catqualia.com.

The work field names third-party titles (anime series and similar) as source material for analogy. No third-party text is reproduced; each row is the author's own description of a mechanism.

Citation

@misc{betances_catqualia_isomorphisms,
  author       = {Betances, Christopher},
  title        = {CatQualia Cross-Domain Structural Isomorphisms},
  year         = {2026},
  howpublished = {\url{https://catqualia.com/isomorphism/}},
  note         = {153,017 rows; structural mappings with failure classes}
}
Downloads last month
16