# particle-bench — G's particle-scale predictions vs the measured particle world

Numbers nobody fitted, against measurements a collider made.

This proof reproduces **Parts I–III** of G's *Ø Predictions* (Appendix B) —
three closed-form, parameter-free predictions of particle-scale quantities —
and confronts each with the value that real precision experiments produced
(CODATA 2018/2022). The single measured input is α, the fine-structure
constant. Nothing else goes in.

```
Part I    m_p/m_e       = 21²·4 + 21·3 + 3² + α·21·(1 − 1/(84π)) + α²·21·16/1836
Part II   G             = α²¹ · (1 + 1/π) · ħc / m_e²
Part III  (m_n−m_p)/m_e = 3·(1 − 1/(2π)) + α·(1 + 1/(2π))
```

Formulas verbatim from G's published `verify.py` (Appendix B).
Source: G (Studio G), *Ø Predictions*,
DOI [10.5281/zenodo.19208226](https://doi.org/10.5281/zenodo.19208226),
CC BY 4.0.

## Run it

```bash
python3 demo.py
```

Zero dependencies — Python standard library only, runs in well under a second,
offline. Self-verifying: it asserts each residual sits inside G's stated
tolerance and **exits non-zero** if any Part falls outside — outside tolerance
= that Part is dead.

## Result (measured, not asserted)

```
Part I    predicted 1836.152673444   measured 1836.152673426   residual 0.010 ppb   (tol 0.017 ppb)  PASS
Part II   predicted 6.7206e-11       measured 6.6743e-11       residual 0.69 %      (tol 1.0 %)      PASS
Part III  predicted 2.5309939330     measured 2.5309883000     residual 2.23 ppm    (tol 5.0 ppm)    PASS
```

A precision ladder spanning roughly nine decades: from 10⁻¹¹ relative
(Part I) through 2.2×10⁻⁶ (Part III) up to 10⁻² (Part II). The browser rendering of the same ladder —
same three formulas ported to JS, computed live — is at
[/the420code/a0/particles/](/the420code/a0/particles/).

## Honest limits

Stated as plainly as the result, because the whole posture is honesty:

- **Three quantities, not a theory of QCD.** This predicts m_p/m_e, G, and
  m_n−m_p. It does not derive hadron spectra, cross-sections, or anything a
  lattice-QCD group would recognise as their job. "Where's QCD?" is a fair
  question, and the published falsification registry (549 conditions) holds
  the kill switches for exactly that line of attack.
- **Reproducing arithmetic ≠ nature agreeing.** This run establishes that G's
  published numbers are real and reproducible from his published formulas —
  internal consistency. Whether the closed forms are physics or numerology
  that happens to land is the open question, and it belongs to specialists.
- **No collider was simulated.** Nothing here models a collision, a detector,
  or an event. The "measured" column is CODATA — the distilled output of the
  precision-measurement community — not raw accelerator data. No affiliation
  with CERN or any laboratory is implied.
- **The tolerances are G's own.** 0.017 ppb / 1.0 % / 5.0 ppm are the
  published failure conditions, not ceilings chosen here to guarantee a PASS.
  Part II sits at 0.69 % of a 1.0 % tolerance — the least impressive of the
  three, and the first place a sceptic should push.

## See also

- The browser precision ladder: [/the420code/a0/particles/](/the420code/a0/particles/)
- The a₀ research note (Part IV, with the independent re-run and the
  falsification registry): [/the420code/a0/](/the420code/a0/)
- The galaxy-scale face of the same five: [/the420code/a0/galaxies/](/the420code/a0/galaxies/)
  and the [`galaxy-rotation`](../galaxy-rotation/) proof.

*Three closed forms. One measured input. Honestly measured against the world.*