Most arguments about unconventional physics theories go nowhere: someone says "it's wrong", someone else says "you just don't get it". I wanted a different format: take the equations literally, turn each claim into a test, and let the numbers decide.
The result is cergneux-audit-demo, an open-source prototype (AGPL-3.0-or-later, Python + NumPy) that audits two documents by Ricardo J. Cergneux:
- Teoría del Tiempo Cuántico (Quantum Time Theory), 2025 version.
- Reinterpretación del efecto túnel cuántico a escala macroscópica (a reinterpretation of Josephson-junction tunneling), December 2025.
The ideas are his. The repo cites them page by page and doesn't reproduce his figures or the PDFs. It is a demo: a v0.x prototype I'll keep working on, not a final verdict on his whole body of work.
The idea: every claim is a test
Each claim from the PDFs is a function with its page number, and each test carries its verdict as metadata:
@pytest.mark.audit(id=10, ref='Túnel §4.1', claim='dn/dt = (E_J sin φ + F)/ħ', verdict='discrepa', note='Contradicts its own dn/dt = −(1/ħ)∂H/∂φ') def test_hamilton_sign_error_in_paper(): ... Running pytest --audit-report docs/AUDITORIA.md regenerates the results table from the tests themselves. CI runs the same command and fails if the committed table drifts. If a test fails, the row shows FALLA, not the author's verdict. The verdict is my thesis; a passing test means the evidence is reproducible.
What came out
Some claims are fine, some are identities in disguise, and some don't survive contact with data.
| Claim | Result |
|---|---|
| Twin paradox: T₁ = γT₂ | Correct. It's standard time dilation with new notation. |
| "c = L_p/T_p confirms discrete motion" | Identity by definition of L_p and T_p. It confirms nothing. |
| "r_s of baryonic mass = radius of the observable universe" | Algebraic identity: r_s of the critical mass inside the Hubble sphere equals c/H₀. |
| Mercury: 28.8 km lag from time dilation | Time dilation gives ~8 km. 28.8 km matches the geometric shift of the GR perihelion precession. |
| Energy conserved band by band while f → f/ρ | Applied to a blackbody it gives an effective emissivity of ρ⁴. The CMB is a blackbody to high precision. |
| Sign of dn/dt in the Josephson Hamiltonian | Wrong sign, checked against a numerical −∂H/∂φ. |
| Test 1: T₂ ~ τ_p/f₀₁ | ~10⁻⁵³ s, versus ~10⁻⁴ s measured in real transmons: about 49 orders of magnitude. |
| The tunneling plateau at low T proves discrete time | Standard dissipative tunneling already explains it. |
The one with real data
The Law of the "Fila" (TC pp. 34, 47) predicts that distant supernovae look shorter: T_obs = T_rest/(1+z). Standard cosmology predicts they look longer.
I tested it against Table 3 of Blondin et al. 2008 (arXiv:0804.3595): spectral aging rates of 35 Type Ia supernovae (22 nearby, 13 distant). Light-curve widths are confounded with intrinsic luminosity, and SALT2 parameters from Pantheon+ already assume (1+z) dilation, so fitting them would be circular. Spectral ages avoid that.
| Model | χ² (35 dof) |
|---|---|
| Standard, rate = 1/(1+z) | ≈ 27 |
| No time dilation | ≈ 150 |
| "Fila" law, rate = 1+z | ≈ 796 |
Fitting rate = (1+z)^−b gives b = 0.97 ± 0.10; the "Fila" law needs b = −1, about 19σ away. I transcribed the table by hand from the PDF and validated the transcription by reproducing the paper's own published numbers (χ² = 27.0, 150.3, b = 0.97 ± 0.10) in a test.
Taking the discrete lattice seriously
The theory says space is a grid of Planck-length cells and light moves one cell per Planck tick. Instead of arguing about it, I implemented it:
- A 1D discrete-time quantum walk reproduces the postulate: massless particles move exactly 1 cell/tick, massive ones slower. The deviation from relativity scales as (E/E_P)², about 10⁻³⁹ for a 7 TeV proton. Undetectable.
- On a cubic 3D lattice, taking "one cell per tick" literally makes the speed of light depend on direction (1/√3 to 1, or 1 to √3): a 42 % anisotropy. Laboratory isotropy tests reach ~10⁻¹⁷. The paper also lets particles jump to "any pixel", which avoids the anisotropy but gives up locality.
- A spatial lattice with continuous time gives quadratic dispersion, v ≈ 1 − (ka)²/8, i.e. E_QG,2 ≈ 3.5 E_P ≈ 4×10¹⁹ GeV. Fermi-LAT (Vasileiou et al. 2013) bounds E_QG,2 > 1.3×10¹¹ GeV, so this is compatible and untestable today.
Honest limitations
- It's a prototype: toy models, a 1D walk, a hand-transcribed sample of 35 supernovae.
- Passing tests establish that a computation is reproducible, not that my verdict is the last word.
- Some parts of the theory (a hidden deterministic "software" running the universe) can't be tested this way, so I left them out.
- The numbers come from tests, and the results table is generated, not hand-written.
Lessons I'd reuse
- Make verdicts part of the test. A generated report that can't drift from the code beats a hand-edited table.
- Validate transcribed data against the source. If your copy of a table reproduces the paper's own χ², you can trust the transcription.
- Beware circular data. An LLM-suggested "supernova dataset" turned out to have an observed factor exactly equal to 1+z in every row, with no error bars. That's the formula pretending to be data. I didn't use it.
- Check the algebra with code. My first hand derivation of the lattice dispersion had the wrong sign; the numerical test caught it.
Try it
git clone https://github.com/amurlaniakea/cergneux-audit-demo cd cergneux-audit-demo python3 -m venv .venv && source .venv/bin/activate pip install -e ".[dev]" pytest -q There are three dependency-free HTML visualizers in viz/ (Josephson washboard, redshift tests, lattice anisotropy).
If you're a physics teacher or student, I'd love feedback on whether this format works for teaching critical reading of a preprint. Issues and PRs are welcome.
This was built with AI assistance (Claude for analysis and code, an agent for repo operations). All numbers are produced by the test suite, and cited values were checked against the original papers.
