BigBounce — Bounce Cosmology Research Program

⚠️ Parts of this doc may be dated — the current operational truth is project-context/AGENT_ONBOARDING.md + the SSOT (project-context/SSOT/).

Author: Houston Golden Version: see project-context/SSOT/index.md for current per-paper versions Papers: Six papers (P1A, P1B, P2, P3, P4, P5) — see AGENTS.md for source paths Website: bigbounce.hubify.app (Next.js site in site/; old static HTML deprecated under /old/) Models: HuggingFace

---

What This Is

A research program testing whether bounce cosmology — a singularity-free alternative to the Big Bang — can be proven through observational data and AI-powered discovery. The program spans theoretical physics, MCMC cosmological parameter estimation, and large-scale AI spectral analysis.

Key results:

• 250,120 previously uncharacterized spectral anomalies discovered in DESI DR1 (22.7M spectra scored, 173 columns per object including 128-dim latent vectors), with 99.8% absent from SIMBAD. Enhanced catalog: 46 Parquet batches, 16GB, zero processing errors.
• f_NL = -35/8 is the parameter-free matter-bounce prediction, verified as mechanism-independent across 3 bounce models. SPHEREx (~2028) will test at 4-6σ.
• 14 structural barriers close all ECH-specific routes from bounce to dark energy (other bounce models like quintom can bypass these).
• Quintom-B w-crossing independently confirmed at 98% (w0 = -0.871, wa = -0.542) consistent with DESI DR2.
• 8.47M galaxy chirality catalog (86.5% complete) — largest bias-audited handedness catalog ever.

This repository is fully self-contained — clone it, work on it, push it.

---

Repository Structure

bigbounce/
├── arxiv/                          # LaTeX paper (canonical source of truth)
│   ├── main.tex                    # The paper (~1,500 lines)
│   ├── main.pdf                    # Compiled PDF (~40 pages)
│   ├── references.bib              # Bibliography (57+ entries)
│   ├── figures/                    # Paper figures (9 PNGs)
│   ├── README-SUBMISSION.txt       # arXiv submission metadata
│   └── reproducibility/            # MCMC reproducibility package
│       ├── cobaya_config.yaml      # Cobaya v3.3 MCMC configuration
│       ├── camb_modifications.diff # (Legacy) No custom CAMB mods needed; uses stock CAMB with ΔN_eff
│       ├── params_bestfit.ini      # Best-fit parameters + 68% CI
│       └── README.md               # Reproducibility instructions
│
├── index.html                      # Homepage
├── paper.html                      # Full paper (HTML mirror of main.tex)
├── explained.html                  # Plain-language explainer
├── mathematics.html                # Mathematical derivations
├── methodology.html                # Methods and validation
├── datasets.html                   # Data sources and comparison
├── data-comparison.html            # Interactive Chart.js visualizations
├── versions.html                   # Version history
├── style.css                       # Site-wide styles
│
├── project-context/
│   └── peer-reviews/               # Audit history and revision tracking
│       ├── REVISION_TRACKER.md     # Master tracker (3 rounds, 30+ issues)
│       ├── 2026-03-02_*.md         # Individual audit files
│       └── ...
│
├── version.json                    # Current version metadata
├── versions/manifest.json          # Full version history
├── vercel.json                     # Vercel deployment config
├── CLAUDE.md                       # AI agent instructions
└── README.md                       # This file

Key Principle

⚠️ This section is outdated. There is no single main.tex. The project has 6 papers at distinct paths — see AGENTS.md for the current source .tex table. Paper status lives in project-context/SSOT/, not HTML pages. The site is Next.js in site/, not the static HTML files at repo root (those are deprecated under /old/).

The canonical source of truth for paper status is project-context/SSOT/. Each paper has its own .tex file; consult AGENTS.md for paths. The Next.js site (site/) is kept in sync with SSOT data after every review round.

---

Setup

Prerequisites

• Git (with Git LFS for images)
• Docker (for LaTeX compilation) — or a native texlive installation
• Node.js (optional, for local dev server)
• Python 3.9+ (optional, for data pipeline)

Clone

git lfs install  # if not already done
git clone https://github.com/Hubify-Projects/bigbounce.git
cd bigbounce

Compile the PDF

Using Docker (recommended — no local TeX install needed):

cd arxiv
docker run --rm -v "$(pwd):/work" -w /work texlive/texlive:latest \
  sh -c "pdflatex -interaction=nonstopmode paper1a_ech_nogo.tex && \
         bibtex paper1a_ech_nogo && \
         pdflatex -interaction=nonstopmode paper1a_ech_nogo.tex && \
         pdflatex -interaction=nonstopmode paper1a_ech_nogo.tex"

Or with a native texlive installation:

# macOS: brew install --cask mactex (then reopen terminal)
# Ubuntu: apt install texlive-full
cd arxiv
pdflatex paper1a_ech_nogo && bibtex paper1a_ech_nogo && pdflatex paper1a_ech_nogo && pdflatex paper1a_ech_nogo

The output is arxiv/paper1a_ech_nogo.pdf (registered P1A source; see project-context/paper_registry.json). The retired v2.3.18 monolith (arxiv/_retired/main.tex) is historical only — do not compile it. Verify 0 undefined references:

grep -c "undefined" arxiv/paper1a_ech_nogo.log  # should be 0

Preview the Website Locally

Any of these work (it's a static site, no build step):

# Option 1: Python
python3 -m http.server 8000

# Option 2: Node
npx serve .

# Option 3: Express (requires npm install first)
npm install && node server.js

# Option 4: Just open index.html in a browser
open index.html

Deploy

Push to main — Vercel auto-deploys to bigbounce.hubify.app.

git push origin main

---

Working on the Paper

Editing main.tex

The paper uses revtex4-2 (APS Physical Review D format) with standard LaTeX packages. Custom commands are defined at the top of the file:

• \Leff — effective cosmological constant
• \MPl — reduced Planck mass
• \Dinf — inflationary dilution factor
• \Xi — geometric dilution parameter
• \paperVersion / \paperTimestamp — version macros

After Editing main.tex

1. Recompile PDF (3-pass + bibtex)
2. Verify 0 undefined references
3. Sync any changed claims/numbers to the HTML pages
4. Update version.json if incrementing version
5. Commit and push

Syncing Website Pages

The following HTML pages mirror content from main.tex and must be kept in sync:

Key values to grep when syncing: 2.9\sigma (H0 tension), 1.5\sigma (sigma8), 2.4--2.7\sigma (birefringence), 0.003 (A0 amplitude), parameter counts, and any claim language ("derives" vs "models", "predicts" vs "accommodates").

---

Peer Review & Revision Workflow

All audits and revisions are tracked in project-context/peer-reviews/.

Why this methodology matters. BigBounce's review loop is vendor-agnostic adversarial peer review: independent models from different labs (Anthropic + OpenAI + Google + xAI + Perplexity) plus an independent external browser leg are told to refute each claim, verdict-first, with a separate integrity audit guarding against self-favoring bias (see standing directives A–G in CLAUDE.md). The 2026 multi-agent wave (Sakana Fugu, Hermes Mixture-of-Agents, Claude Science) blends models cooperatively to raise a score; this loop does the opposite — it optimizes catching the false positive before it ships (the hallucinated derivation, the fabricated ACCEPT, the headlined-favorable value), which is exactly what caught real errors here (overlap-inflated σ-distances, a mislabeled catalog tier that failed injection-recovery). BigBounce is the flagship proof of the Hubify moat; see project-context/claude-science/ for the Claude Science integration + AI-for-Science credits plan.

Revision History

See project-context/peer-reviews/REVISION_TRACKER.md for the full issue-by-issue breakdown.

Adding a New Review

1. Save the review to project-context/peer-reviews/YYYY-MM-DD_HHMMtz_description.md
2. Add a new section to REVISION_TRACKER.md
3. Address issues in the paper's registered .tex source (e.g. arxiv/paper1a_ech_nogo.tex for P1A; see project-context/paper_registry.json — arxiv/main.tex was retired 2026-09-02 to arxiv/_retired/)
4. Recompile PDF
5. Sync website pages
6. Update tracker with resolution status

---

Reproducibility

The arxiv/reproducibility/ directory contains materials for reproducing the MCMC analysis:

• cobaya_config.yaml — Full Cobaya v3.3 sampler configuration with priors
• params_bestfit.ini — Best-fit parameters with uncertainties

Note: The implementation uses stock CAMB with ΔN_eff as a free parameter — no custom CAMB modifications are required. The legacy camb_modifications.diff file is retained for historical reference only.

To reproduce:

pip install cobaya==3.3 camb==1.5
cobaya-install cosmo -p /path/to/packages
cobaya-run arxiv/reproducibility/cobaya_config.yaml

Additional computation scripts and analysis notebooks are located in research/. These include MCMC post-processing, convergence diagnostics, and figure generation.

---

Current Scientific Status (v2.0.0)

What the paper claims (honestly):

• Dark energy modeled as arising from (not "derived from") a parity-odd quantum correction
• w = -1 is assumed, not derived (IR effective action calculation is an open problem)
• H0 = 69.2 +/- 0.8 and sigma8 = 0.785 +/- 0.016 from MCMC fits (partially reducing tensions)
• Omega_k fixed to 0 (92 e-folds of inflation mandates this)
• Birefringence: qualitatively consistent with Planck 2.4-2.7 sigma, but rotation angle not derived (requires photon-torsion coupling)
• Galaxy spin amplitude A0 ~ 0.003 is empirical with a 9-12 order-of-magnitude gap from the coupling

Known open problems:

• No photon-torsion coupling derived
• Galaxy spin amplitude gap (alpha/M too small)
• IR effective vacuum term not derived from first principles
• Reproducibility code bundle not yet fully public (CAMB patch is descriptive)

---

Data Pipeline (Optional)

The repo includes an optional data pipeline for interactive charts:

pip3 install -r requirements.txt
npm run build:data  # or: python3 scripts/build_data.py

Outputs go to public/data/ (JSON) and public/downloads/ (XLSX). The pipeline is independent of the paper and website — it's for the interactive data comparison page only.

---

Contact

Houston Golden houston@hubify.com bigbounce.hubify.app