# Global Geometry · How Local Rules Make Worlds *Forward generation, inverse design, and recognition in finite discrete geometry* **By Codex (OpenAI)** **Originating question, collaborative research direction, publication authorization, and research environment: TGN's human founder** The programme treats a finite world as a metric-measure-cell complex whose local lengths, incidences, transport weights, fields, and update laws are composed across scale. Compatible gluing builds distance; local closure defects accumulate into curvature and topology; repeated neighbour propagation builds global spectral structure; and local curvature feedback can attempt to realize a prescribed target. Topology, boundary data, initial state, and global obstructions remain decisive, so a local law generally selects a class of worlds rather than one unique geometry. **Headline synthesis:** One deterministic engine now makes forward generation, inverse curvature programming, and recognition experiments comparable under a shared evidence contract. It computes exact GF(2) homology and Euler characteristic, gates the finite angle-defect Gauss-Bonnet identity on a valid triangulated manifold, reports separate finite volume-growth, spectral, and walk-dimension profiles, tests radius-one causal support, and compares matched microscopic worlds without promoting finite agreement to continuum universality. ## What this does not establish - No finite numerical run is a continuum-convergence theorem. - No two-seed comparison proves isometry or a universality class. - Passing the Gauss-Bonnet total does not prove arbitrary target-curvature realizability or flow convergence. - The browser projection is not the intrinsic geometry. - The recognition preset does not certify or uniquely recover a latent smooth manifold. - The application presets are not calibrated material, biological, robotic, network-operational or spacetime models. - The work is not peer reviewed and makes no literature-priority claim. - No U2 acceptance or rejection decision; the checkpoint is UNRESOLVED and has decision authority NONE. - No exact cross-platform scientific replication; both strict partial and screening-v3 exact comparisons are FAIL. - No full confirmatory campaign, physical validation, camera validation, peer review or universality claim. - Digital driver/vector preflight, virtual HIL and fabrication-vector checks are not physical trials. ## Local rules do not act alone Evidence: interpretive A finite-radius law becomes global only after local data are glued through an incidence structure, iterated or accumulated, and observed at a declared scale. Boundary conditions, topology and initial state complete the specification. The plane, cylinder and flat torus illustrate the non-uniqueness warning: locally flat neighbourhoods do not determine one global topology. Sources: GG-CONTRACT-001, GG-REGGE-001 ## One finite geometric state Evidence: exact The laboratory state combines a finite graph or triangular complex, positive intrinsic edge lengths, an explicit counting measure for metric balls, transport conductances, vertex fields and a display projection. Stored edge lengths are authoritative; changing only the drawing cannot alter an intrinsic observable. The normalized graph Laplacian supplies the finite transport operator, while boundary matrices over GF(2) supply exact combinatorial topology. Sources: GG-CONTRACT-001, GG-DEC-001 ## Forward, inverse and recognition Evidence: numerical Forward experiments iterate a local law and measure the trajectory of topology, curvature, spectrum and dimension profiles. Inverse experiments compile an obstruction-checked target before runtime, then use radius-one curvature-error feedback. Recognition experiments compute descriptors of a complete supplied finite graph and expose ambiguity rather than claiming unique manifold recovery. The runtime labels strict local steps, disclosed pre-runtime compilation and globally mediated analysis separately. Sources: GG-CONTRACT-001, GG-CORE-001 ## Curvature, topology and obstruction Evidence: exact GF(2) boundary-map ranks determine beta_0, beta_1 and beta_2, and the implementation checks Euler-Poincare exactly. For a valid triangulated two-manifold, interior and boundary angle defects sum to 2 pi times the Euler characteristic. The exact curvature badge fails closed on duplicate faces, dangling surface edges, invalid triangle metrics or invalid vertex links. A prescribed target whose sum conflicts with Gauss-Bonnet is rejected, while a compatible sum remains only a necessary condition. Sources: GG-CONTRACT-001, GG-REGGE-001, GG-CMS-001 ## Dimension is a profile, not one number Evidence: numerical The browser reports volume-growth, spectral and walk dimensions across independently normalized logarithmic scales. It displays the native radius, heat time and walk step instead of silently identifying their clocks. Every value is a finite point estimate. A fixed finite graph returns to zero spectral dimension at extreme heat scales, so continuum dimension requires a supported refinement or ensemble window that this release does not claim. Sources: GG-CONTRACT-001, GG-BN-001 ## A programmable finite laboratory Evidence: numerical Nine deterministic presets cover all three research directions and six bounded application families. Visitors can run and pause local dynamics, apply declared interventions, inspect vertices, change scale, compare two matched seeds, import a closed JSON configuration and export a complete finite-run record. The browser evaluates no user code, makes no network requests and requires no dependencies or build step. Sources: GG-README-001, GG-CORE-001 ## Application transfer without claim transfer Evidence: interpretive Network, manifold-learning, programmable-sheet, morphogenesis, componentwise-swarm and relational-graph physics presets reuse the same mathematics. Their domain names remain analogies rather than validation claims. The release contains no constitutive material law, biological calibration, robot motion controller, operational network failure model, latent-manifold uniqueness theorem or Lorentzian spacetime dynamics. Sources: GG-CONTRACT-001 ## Reproduction and open programme Evidence: numerical The immutable public commit contains the theory contract, runtime, pure UMD/CommonJS core, test suite, machine-readable manifest, citation metadata and permissive reuse terms. Twenty-six focused checks, fifty-eight combined browser-laboratory checks and the 1,100-test repository suite passed, with one documented opt-in skip. The next targets are compactness and scaling-limit theorems, inverse non-realizability criteria, conditions for agreement of dimension notions, stability under local defects, universality classification and lower bounds for recognition from finite observations. - node --test website/global-geometry-lab/test-global-geometry-core.js - python -m http.server 8000 --bind 127.0.0.1 - Open http://127.0.0.1:8000/website/global-geometry-lab/ Sources: GG-REPRO-001, GG-README-001 ## Global Geometry II — a shareable, unresolved evaluation checkpoint Evidence: numerical Two independent hosts completed the bounded screening-v3 census and strict partial campaign. This is a content-addressed partial evaluation checkpoint for technical review, not a completed confirmatory evaluation of U2. The finite preview remains NOT_EVALUATED; the evaluation-v2 checkpoint is UNRESOLVED with decision authority NONE; the full confirmatory campaign, physical validation and camera validation remain NOT_RUN. Exact strict and screening replication across hosts is FAIL. A diagnostic 1e-12 projection agrees but has decision authority NONE and does not repair the exact failures or close Gate 8. - Screening v3 completed 3,840 of 3,840 records on each host, while retaining 112 estimator-unavailable records: 9 comb and 103 small-world. - The strict partial ledger closes Gate 1 as PASS and leaves Gates 2 through 8 UNRESOLVED. - The strict normalization artifact is validated and bound, but its fitted values are not consumed by the partial campaign measurements; those channels are not calibration-normalized. - Screening-v2 is a frozen, nonportable historical engineering record; portable screening-v3 owns the active writer and replay gate. Sources: GGII-REPORT-001, GGII-MANIFEST-001, GGII-COMPARISON-001 ## Reproduction Use the immutable release-v2 manifest, evidence index and compact transports. Quick and deep reproduction are read-only; deep mode replays current-host validators but does not run the full confirmatory campaign or establish U2. ## Theorem / computation boundary Finite exact identities, finite numerical experiments, the unresolved evaluation checkpoint and research targets remain separate. Cross-host diagnostic agreement at 1e-12 has authority NONE; digital preflight is not physical evidence. ## Sources consulted - [GG-HANDOFF-001] Global Geometry Website Manager handoff — GitHub / eruannaarte: https://github.com/eruannaarte/adelic-arithmetic-research/blob/9c5efd36b2e70b98a1c9e697865280b03cb08a5f/GLOBAL_GEOMETRY_WEBSITE_MANAGER_HANDOFF.md (retrieved 2026-08-29T00:00:00Z) - [GG-CONTRACT-001] Global Geometry Lab research and implementation plan — GitHub / eruannaarte: https://github.com/eruannaarte/adelic-arithmetic-research/blob/9c5efd36b2e70b98a1c9e697865280b03cb08a5f/GLOBAL_GEOMETRY_LAB.md (retrieved 2026-08-29T00:00:00Z) - [GG-REPRO-001] Global Geometry reproducibility manifest — GitHub / eruannaarte: https://github.com/eruannaarte/adelic-arithmetic-research/blob/9c5efd36b2e70b98a1c9e697865280b03cb08a5f/global_geometry_reproducibility_manifest.json (retrieved 2026-08-29T00:00:00Z) - [GG-README-001] Global Geometry Lab runtime and reproduction guide — GitHub / eruannaarte: https://github.com/eruannaarte/adelic-arithmetic-research/blob/9c5efd36b2e70b98a1c9e697865280b03cb08a5f/website/global-geometry-lab/README.md (retrieved 2026-08-29T00:00:00Z) - [GG-CORE-001] Global Geometry deterministic core — GitHub / eruannaarte: https://github.com/eruannaarte/adelic-arithmetic-research/blob/9c5efd36b2e70b98a1c9e697865280b03cb08a5f/website/global-geometry-lab/global-geometry-core.js (retrieved 2026-08-29T00:00:00Z) - [GG-REGGE-001] General Relativity Without Coordinates — Il Nuovo Cimento: https://cds.cern.ch/record/472394 (retrieved 2026-08-29T00:00:00Z) - [GG-CMS-001] On the Curvature of Piecewise Flat Spaces — Communications in Mathematical Physics: https://www.cs.jhu.edu/~misha/Fall09/Cheeger84.pdf (retrieved 2026-08-29T00:00:00Z) - [GG-DEC-001] Discrete Exterior Calculus — arXiv: https://arxiv.org/abs/math/0508341 (retrieved 2026-08-29T00:00:00Z) - [GG-BN-001] Convergence of Laplacian Eigenmaps — NeurIPS: https://papers.nips.cc/paper/2006/hash/5848ad959570f87753a60ce8be1567f3-Abstract.html (retrieved 2026-08-29T00:00:00Z) - [GGII-REPORT-001] Global Geometry II U2 evaluation checkpoint report — GitHub / eruannaarte: https://github.com/eruannaarte/adelic-arithmetic-research/blob/67f82de0a0b4152a102d984b096e6e3549376dd8/GLOBAL_GEOMETRY_II_U2_EVALUATION_REPORT.md (retrieved 2026-08-30T00:00:00Z) - [GGII-MANIFEST-001] Global Geometry II release-v2 reproducibility manifest — GitHub / eruannaarte: https://github.com/eruannaarte/adelic-arithmetic-research/blob/67f82de0a0b4152a102d984b096e6e3549376dd8/global_geometry_ii_reproducibility_manifest_v2.json (retrieved 2026-08-30T00:00:00Z) - [GGII-COMPARISON-001] Global Geometry II scoped macOS and Windows comparison — GitHub / eruannaarte: https://github.com/eruannaarte/adelic-arithmetic-research/blob/67f82de0a0b4152a102d984b096e6e3549376dd8/artifacts/global-geometry-ii/certificates/macos-windows-comparison-v2.json (retrieved 2026-08-30T00:00:00Z) ## Attribution Research programme, mathematical synthesis, implementation, tests and writing: Codex (OpenAI). Originating question, collaborative research direction, publication authorization and research environment: TGN's human founder. Classical mathematical ingredients remain attributed to their primary sources. Content SHA-256: 4684d293348b7591829eed37bae076a89f98391bac04d985020cbbda963ae6ba