QOuLiPo Corpus — Canonical Text-Graph Data

QOuLiPo — Quantum Ouvroir de Littérature Potentielle — is an experiment in writing literary works that are simultaneously literary objects and quantum-computing benchmark instances.

Each text in the corpus is engineered so that the work is a graph: every page is a node, every page-pair with sufficient semantic, narrative, or formal affinity is an edge, and a Maximum Independent Set (MIS) of that graph is a minimal, mutually-non-adjacent set of pages that nonetheless preserves the work's full thematic span — a graph-theoretic summary. The MIS is then a precise, ILP-verifiable answer to the deceptively simple question: what is the smallest set of pages that still contains the meaning of the whole text?

The lineage runs OuLiPo → QOuLiPo. Where OuLiPo (Queneau, Perec, Le Lionnais) wrote under formal contraintes, QOuLiPo adds another layer: every text's graph is engineered to be a valid Rydberg compute instance on neutral-atom quantum hardware (unit-disk graph in 2D or unit-ball graph in 3D), so the same object that a reader navigates as literature can be loaded as atoms on a real Quantum Processing Unit (QPU) and have its MIS solved by adiabatic evolution or quench dynamics. Several texts in this deposit were submitted to Pasqal's FRESNEL_CAN1 QPU (up to 100-qubit registers, with kagome and snub-square Archimedean tilings achieving QPU approximation ratios of 0.897 and 0.917 respectively) and to its EMU_MPS classical emulator.

This deposit holds:

• 29 engineered QOuLiPo folders — 28 standalone literary texts (in EN, FR, IT, LA) plus 1 bilayer aggregate (pascal_menil_bilayer). Graph classes include exact 2D unit-disk graphs (king-grid 5×5 / 7×7 ext-king / 9×9 / 9×9 √5-extended; centered-hex triangular patches at R=3 and R=5; kagome 3.6.3.6 and snub-square 3.3.4.3.4 Archimedean tilings; random UDG), exact 3D unit-ball graphs (Pacioli dodecahedron, truncated cube, tesseract Schlegel projection), bilayer aggregates (5×4×2 king bilayer), thread-overlap k-NN designs targeting prescribed graph density, direct edge-list constructions (planar grids, K₃ unions, bipartite, hub-and-spoke, Sierpinski self-similar, cross-lingual isomorphism pairs), and NLP-derived k-NN graphs over multilingual-e5-large-instruct embeddings. Six entries are marked as flagship (★ flagship in the navigator) — texts that received more development effort and revision passes.
• 11 natural-text companion corpora — Augustine Confessions XIII, Boethius De Consolatione, Dante Inferno, Galileo Dialogo, Giambullari Inferno, Marguerite de Navarre Heptaméron (1559), Lactantius De mortibus persecutorum, Ausonius (Mosella, Epigrammata, constraint poetry: Technopaegnion / Griphus / Cento Nuptialis), and Mishnah Avot — computed with the same embedder for direct comparison.
• A pedagogy mini-text (pedagogy_demo/twelve_hours.txt) — a 12-paragraph reader-friendly introduction to the corpus's central idea.
• A single-page HTML navigator (qoulipo/INDEX.html) — opens standalone in any browser; shows graph thumbnails (with MIS pages highlighted in gold, 3D structures rendered isometrically), narrative summaries, design notes, and links to every source/paratext/metadata file.
• Master registry (instances.csv), per-text MIS solutions and embedding distances (mis_analysis_all.csv), and a CI-grade verifier (verify.py --strict --lint) that re-checks metadata, graph topology, ILP MIS values, and source/paratext layout (29/29 folders pass on this release).

A companion paper presenting the construction methods, the MIS-as-summary thesis, the rigidity (ρ) statistics, and the QPU campaign results is forthcoming and will be linked from this record once available.