I’ve been developing a cosmological framework over the past several months with Claude as my primary collaborator. I have no formal physics training, so the math heavy lifting was definitely a joint effort. What I brought was the core ideas and a lot of stubbornness about not letting the AI just tell me what I wanted to hear.
The basic idea is that our universe is one of many expanding regions in a single infinite connected spacetime. Each region originates from a regional LQC bounce triggered when converging matter from neighboring regions reaches Planck density. Inter-regional gravity enters through the retarded Weyl tidal field rather than direct density injection, that distinction took a lot of back and forth to get right.
The framework makes specific predictions:
• Directional Hubble parameter H(θ) = H₀ + ε̇cosθ explaining the Hubble tension as a structural feature rather than a measurement problem
• CMB quadrupole-octupole alignment from a single inter-regional dipole source
• Survivor black holes from prior bounce cycles explaining anomalous JWST objects — interestingly Gaztañaga just published something similar in PRD independently
• A void-cluster growth asymmetry estimator A(z) = D\_void - D\_cluster as a clean null test against ΛCDM
• S₈ ≈ 0.76-0.79 from Weyl shear growth suppression
We ran LTB-LQC hybrid simulations showing bounce onset preceding horizon formation in 96% of 50 shells, and a joint likelihood grid scan showing weak Bayesian preference over ΛCDM across BAO, RSD, and CMB datasets.
The frustrating part is I can’t get feedback from the mainstream physics subreddits because I used AI and the work reads like it. Which is fair honestly, but it means the ideas either sink or swim based on whether anyone here engages with them seriously.
Papers on Zenodo:
Main framework:DOI 10.5281/zenodo.19721945
Entropy companion paper:DOI 10.5281/zenodo.19778385
Genuinely curious what the technical objections are. Claude was good at stress testing ideas but I’m aware it has blind spots.