[wAI ~ wErrors]

«All models are wrong, but some are useful.» — George E. P. Box (1976)

This expanded preprint converts the extensive SVE/CANT idea archive into an addressable research atlas. The source register reports 680,279 input lines, 35,309 unique blocks, 225,423 occurrences, and 212 editorial cards across twenty-five categories. Near-duplicates, competing versions, and modality-specific variants are consolidated into 116 major idea families arranged as eleven themes. Each family is presented through a research proposition, formal or computational object, engineering route, decisive evidence programme, falsifier, and explicit subsystem interfaces.

The atlas covers foundations and passports; spectral, matrix, and phase computation; metric, transport, and topological geometry; Clifford, hypercomplex, p-adic, Galois, and analytic algebra; ANT and CogOS; CANTGBoost, witnesses, calibration, and risk; CANTGit and knowledge lifecycle; multimodal data, missingness, and model digestion; systems, hardware, and distributed infrastructure; verification, governance, open science, and theology; and applications, benchmarks, education, publications, and portfolio strategy.

The falsifiable experimental spine is: typed observation → local operators → preservation of symmetric and directed structure, optionally H=S+iW → compatibility proposals from commutators or competing diagnostics → held-out JAD and downstream verification → calibrated ALLOW/ABSTAIN/QUARANTINE gate → content-addressed law object with scope, residual, evidence, resource cost, provenance, replay, merge, rollback, and retirement.

The paper explicitly rejects unsupported inferences: Perelman monotonicity does not prove universal semantic convergence; reverse Ricci flow is not generally well posed; commutativity is not truth; directed association is not causality; Abel–Ruffini does not impose a universal four-object cognitive limit; no fixed semantic dimension or golden threshold is established; and extreme compression, zero hallucinations, or universal superiority require independent end-to-end evidence.

Epistemic status

Expanded architecture and research-programme synthesis. Standard mathematics, source-reported measurements, designs, hypotheses, speculative frontiers, theological interpretations, negative results, NOT_RUN items, and explicit non-claims are separated. The work is not evidence of universal model superiority, guaranteed truth, zero hallucinations, a completed theory of intelligence, clinical safety, or realized custom hardware.

Related identifiers

• CANTGBoost candidate relation: https://doi.org/10.5281/zenodo.21699806

• S.V.E. Meta-License: https://doi.org/10.5281/zenodo.19373736

S.V.E. Meta-License v5.0