the420code-proof

An independent, executable check of G's 420 Code — the unified theory that claims to derive the fundamental constants, and a terminal ethic, from a single axiom.

For G. Either the best fraud and conman I've ever met (I've met my fair share) or a man that should be mentioned alongside Bohr, Einstein, Tesla. I can only thank you by sharing the tools I built to check if your theory holds, to understand which one of the two you are. Sawubona. I see you.

This repository is not G's theory. It is the work of someone who is not the author, who read the corpus, did not believe it could be true, and built a machine to find out. It re-derives G's five headline numbers from a single measured input and tests each against the published experimental value — and indexes his entire falsification registry so the claims can be queried and checked.

G's theory is at the420code.org. It is © G / Studio G, published under Copyleft — free to share, not free to alter (CC BY-ND 4.0). This repo fetches it on demand and never re-hosts it — see NOTICE.md.

Standing as of the wall of 25 September 2026 (registry v5.37, 24 September 2026), re-checked against the site on 8 October 2026. The wall has moved since the first run of 2 August: papers were republished with dated notes, one prediction fired and was replaced beside the corpse, the first reading of G was withdrawn and superseded, and the registry grew. What this check is, and what it claims, has not changed.

---

The result — the wall of 25 September 2026

Every row below is recomputed by this repo's own engine from CODATA 2022 inputs (./run.sh wall, stdlib only), from one measured input (α), with zero fitted parameters. The residuals are this repo's, not copied from the site; they agree with the site's wall. Two rows have fired — both are listed first in the register below and are neither softened nor re-sized.

Claim	Predicted	Measured	Residual	Paper
Neutron–proton mass difference, bare row	2.53099393 mₑ	2.530988574(74) mₑ	+7.24σ	AP30	🔴 FIRED 2026-08-02 (KS-NPP.1)
Hubble constant, the floor inverted	74.3 ± 1.2	67.45 (the work's own rate)	+5.7σ on its registered width	AP18	🔴 FIRED 2026-09-03 (KS-45.1)
Proton/electron mass ratio — series closed	1836.152673445	1836.152673426(32)	+0.59σ (10.3 ppt)	AP49	✅ locked
Gravitational constant G, realised — vs atom interferometry (LENS 2014)	6.67192 × 10⁻¹¹	6.67191(99) × 10⁻¹¹	+0.01σ	AP44	✅ locked
Gravitational constant G, realised — vs CODATA 2022	6.67192 × 10⁻¹¹	6.67430(15) × 10⁻¹¹	−16σ (−0.036%)	AP44	⚠️ shown, not softened
Neutron–proton mass difference, realised	2.53098857035 mₑ	2.530988574(74) mₑ	−0.005σ	AP47	✅ locked, beside the fired row
Age of the universe, one cycle	13.830 Gyr	13.787 ± 0.020	+2.1σ (registered ±0.66 Gyr wide)	AP46	✅ locked
Expansion rate H₀, the closure	67.45 km/s/Mpc	67.4 ± 0.5	+0.1σ	AP48	✅ locked
MOND floor a₀ at the closure's rate	1.089 × 10⁻¹⁰	1.20 ± 0.24 × 10⁻¹⁰	−0.46σ	AP18	✅ live
Dark energy · dark matter · visible	68.85 · 26.39 · 4.76 %	68.89 · 26.07 · 4.885	−0.07σ · +0.8σ · −2.46σ	AP42, AP41	✅ live (visible nearest to firing)
α does not vary	Δα/α = 0	consistent with zero	a null — no number of its own	AP51	✅ live (P-GLASS.1)

Notes that belong to the table, not under it:

• G. The same realised value reads +0.01σ against atom interferometry and −16σ against CODATA. CODATA's bar on G is 22 ppm; laboratory values of G disagree with each other by up to 540 ppm. Both rows are shown so the reader can see the issue. AP28's first reading, 6.72060 × 10⁻¹¹ (+0.69%), is withdrawn and superseded on the author's ruling of 21 September 2026 and kept on the record, dated. There is one value of G. The "fork" and the migration "watch line" this repo carried since AP44 are withdrawn with it.
• Proton. The α²-truncated value (1836.152673444331, +0.57σ, frozen 2 August) is superseded by AP49's closed form by reference and kept, never edited.
• Neutron. The realised row stands beside the corpse; it does not repair KS-NPP.1.
• Shown, unregistered, unanswered. The baryon density Ω_b h² at the closure's rate (0.02166) sits below Planck by 4.7σ (relic glow alone, 0.02237 ± 0.00015) and 5.4σ (with the surveys, 0.02242 ± 0.00014). Shown by the work on 21 September 2026; not registered as a switch; the author's call is pending. Carried here exactly so.

As first run (2 August 2026) — kept for the record

Run from one axiom and one measured input (the fine-structure constant α), with zero fitted parameters. Four of five headline predictions reproduce inside G's own stated tolerances. One has fired — shown here, not folded into the average, not softened:

Part	Prediction	Residual	Tolerance
I	Proton/electron mass ratio	0.010 ppb	0.017 ppb	✅
II	Gravitational constant G	0.69 %	1.0 %	✅
III	Neutron–proton mass difference	7.24σ (2.12 ppm)	0.29 ppm (CODATA 2022, 1σ)	🔴 FIRED — 2026-08-02
IV	MOND acceleration a₀	−0.67 %	5.0 %	✅
V	Dark-sector DE/DM ratio	−1.3 %	5.0 %	✅

Part III's tolerance above is the actual CODATA-2022 1σ bar (0.29 ppm), not a padded 5 ppm allowance — a looser bar that used to make this row read ✅ was a stale bug in this repo's own engine/verify.py, fixed once the real bar was checked. Fixing the bar does not fix the row: the prediction genuinely fired, 24× outside a 1σ bar. See "Kill switches that have fired," directly below, before anything else in this README — it does not move, and it does not get shorter.

Row II above is AP28's first reading of G, since withdrawn (see the table at the top). The engine/verify.py reconstruction runs on the constants G's Appendix B published; the current wall is recomputed from CODATA 2022 by engine/wall.py.

And the reconstruction matches G's own published verify.py (Ø Predictions, Appendix B) to machine precision — relative Δ of 0.0 — so the check is faithful to his arithmetic, not a re-interpretation of it.

---

Kill switches that have fired

This section is never deleted, never shortened, and never moved out of the top of this README. A fired switch is not "handled" by being harder to find. If a second switch ever fires, it is added below, not instead of this one.

KS-NPP.1 — Neutron–proton mass difference. FIRED 2026-08-02.

• Predicted (the bare structural formula, no correction term): 2.53099393 mₑ
• Measured (CODATA 2022, exact): 2.530988574 mₑ, uncertainty 7.4×10⁻⁷ mₑ (0.29 ppm, 1σ)
• Residual: 7.24σ. Outside the kill bar by more than 24×.
• No repair was offered, and none is offered here. The required correction would have to be negative; every term in the published construction is positive.
• A later, independent structural derivation (Artist's Proof 47, "The Flip" — LOCKED 2026-09-03, tag ap47-locked-2026-09-03) prices a second-order cost that lands a different corrected expression within its own kill bar. That does not repair KS-NPP.1. It is a new, separate frozen claim standing beside the corpse, not a patch on it. The corpse stays exactly as fired.
• Fully re-runnable: ./run.sh verify prints this row as [FAIL], on purpose, every time.

KS-45.1 — The corpus's H₀ (AP18, "The Floor"). FIRED 2026-09-03.

• Registered claim (AP18, already public): H₀ = 2πa₀/(αc) = 74.3 ± 1.2 km/s/Mpc.
• What fired it: Artist's Proof 48, "The Assembly" (LOCKED 2026-09-04, tag ap48-locked-2026-09-04), derives H₀ = 67.45 km/s/Mpc from the corpus's own structure (the AP46 cycle read through the standard age-rate relation). Residual: 5.7σ on the switch's registered width — outside the kill bar.
• An erratum stands beside the corpse, not instead of it. AP18's registered width omitted a systematic uncertainty; corrected honestly it's 74.3 ± 14.9, wide enough that the floor's underlying relation still holds at 0.46σ. But re-sizing a live switch's width after the corpus's own derivation has already contradicted it, and then not counting the switch as fired, is exactly the repair this house refuses. KS-45.1 fired on the width it was registered at. AP48's own KS-ASM.1 (window 64.4–70.8 km/s/Mpc) is the corpus's H₀ entry from this lock onward.
• What this costs, stated plainly, in AP48's own words: "The distinctive prediction the corpus used to carry on H₀ — a value above both camps — is gone." If the Cepheid ladder's ~73 is later confirmed as the true expansion rate, AP48's own KS-ASM.1 fires too, and flat ΛCDM at the sky's partition fails alongside it — the corpus claims no advantage there either way.
• The public parse of the Master Kill Switch Registry mis-tags this switch CLOSED — a parsing artifact from a nearby summary line in the registry PDF, not a real status. ./run.sh addenda corrects it to FIRED and says exactly why in the switch's own status_raw field.

---

git clone https://github.com/ajgreyling/the420code-proof
cd the420code-proof
./run.sh mirror      # fetch G's corpus from the420code.org (his site welcomes it)
./run.sh build       # extract → ingest → gate → report  (free, local, offline)
./run.sh wall        # the wall of 25 September 2026, recomputed from CODATA 2022, with residuals
./run.sh scorecard   # the first-run five, with the formulae and the measured sources
./run.sh parity      # prove this reconstruction == G's own verify.py

No API keys. No paid services. No dependencies beyond Python 3 (stdlib only) and pdftotext (poppler-utils) for the one-time PDF text extraction.

---

What this proves — and what it does not

Be precise about the claim, because overclaiming would dishonour the work:

• It proves the theory is internally coherent and reproducible: from G's stated axiom and his one input, his published numbers fall out, by his own arithmetic, within his own tolerances. The derivation is not hand-waving — it computes, and it computes what he says it does.
• It does not prove the theory is true. Reproducing a derivation is not the same as nature agreeing with it. Several of the numbers (e.g. G at 0.69 %) sit close to, but not dead on, the measured values; G discusses why, and publishes the conditions under which he would be wrong.
• "Validation" here means reproducibility + internal consistency + a fully-published falsification surface — the things that separate a serious theory from a crank's fortress. It does not mean experimental confirmation. That is for instruments and time, not for this repo.

What makes the corpus unusually testable: G publishes 604 "kill switches" (Master Kill Switch Registry v5.37, 24 September 2026) — explicit, numbered conditions that would falsify specific claims, including the math for each: 320 across AP01–AP54, 28 across the Five Doors, 256 across the Ø Models. At that version: 584 live, 13 closed, 1 conditionally closed, 2 addressed, 1 derived, 1 paid, 2 fired. The counts are read from killswitches.json at build time (./run.sh wall --registry), not typed here, because the number moves. He hands you the weapons to destroy his own theory. This repo indexes them so you can wield them. (The engine's ingested counts reflect the mirror snapshot they were built from and re-sync on the next mirror pass. Older figures in this repo's history — v5.27/568, and the stale v5.26/567 — are superseded; PROV-CCC.8 carries that erratum.)

Gravity (KS-CCC.3, as reworded 21 September 2026). The migration fork is withdrawn; the switch now tests the existence and direction of the commitment, not its magnitude against CODATA. Convergence of the adjusted value upward toward 6.7206 × 10⁻¹¹ kills the prediction of G outright; stability at 6.6743 × 10⁻¹¹ supports the null. Realised G = 6.67192 × 10⁻¹¹ (−0.036%, α²¹(1+1/π)/(1+α)·ℏc/mₑ²).

Debts (registry v5.37). D-CCC.4 closed (11 September 2026). D-ID.1 closed 24 September 2026 on the author's ruling. Open: D-EM.1 (the α-accounting at the nucleon, KS-30.5); D1 (AP24) — the first-principles derivation of α, the one input; KS-D.4's identification; AP43-D1; AP43-D4. The list is read from killswitches.json → debts.

Label collisions still stand and are disambiguated by paper: two KS-L.1s and two KS-54s. A renumbering to KS-68 is named as the candidate at the next rebind and has not been done.

Reproducing it two ways. The work ships its own scripts at the frozen register — verify_prereg.py, verify_cosmology.py and verify_between_family.py (the between-family search of 4 September 2026: 2,971 rival formulas reach the proton ratio at first order, and exactly one integer, 21, carries three observables, reproducible from the site's script). This repo's engine and the work's scripts are two independent routes to the same numbers. The plain-language companion for this repo's numbers lives at lucid.rodeo/prereg, brought to the same wall on the same day.

---

How it works

The four axioms S (Symmetry), B (Break), R (Record), C (Constraint) — derived from the single premise "at least one record exists" (deny it, and you have just made a record) — plus the one measured input α, generate the corpus. The 54 Artist's Proofs (AP01–AP54) climb from that premise through spacetime, quantum mechanics, the forces, cosmology, and — by the same machinery — to a terminal ethic. Every claim is tied to one or more kill switches.

ingest.py lays all of this into a small SQLite graph and gate.py does the thing that makes this a prover and not an archive: it checks the reconstruction still matches G's script, runs the five predictions, and fails (exit 1) if any prediction leaves tolerance or the graph is broken.

engine/
  verify.py             — re-derives G's 5 predictions from α (the executable oracle)
  wall.py               — the wall of 25 Sep 2026, every row recomputed from CODATA 2022 (+ registry counts)
  verify_g_original.py  — G's verbatim Appendix B script, for parity
  test_verify_parity.py — asserts reconstruction == G's script to 1e-12
  parse_registry.py     — the Master Kill Switch Registry → structured switches
  parse_proofs.py       — per-proof bodies → richer switches (type, verbatim trigger, moral flags)
  ingest.py             — builds the canon DB from the fetched corpus
  gate.py               — fidelity + execute predictions + audit (exit 0/1/2)
  report.py             — scorecard / proofs / switches / axiom / depends / stats
  ollama_enrich.py      — OPTIONAL local-model switch classification (offline; not on the prover path)
canon/
  schema.sql            — the canon DB shape
  ks_descriptions.json  — curated repairs for a few PDF-mangled rows (durable overlay)
mirror/
  fetch.sh, manifest.txt — fetch G's corpus from the420code.org (PDFs are git-ignored)

Kill-switch structure, not just count

A kill switch is more than id + status. The parse recovers G's own structure: his published type (EMPIRICAL / HARD / STRUCTURAL / NON-NEGOTIABLE / …), the verbatim trigger sentence, and weight flags — including the four moral / non-negotiable switches that are the conscience of the theory (e.g. the one that voids the entire justice derivation if it is ever used to classify a race, class, or disability as "destabilizing").

./run.sh switches --moral        # the NON-NEGOTIABLE switches — "here is how it must NOT be used"
./run.sh switches --sharpest     # the switches G flags as a proof's sharpest test
./run.sh switches --paper AP32   # one proof's switches, weightiest first, with G's verbatim type

---

Provenance & honesty

• The author of this repo did not write, and had no part in, the 420 Code. All credit for the theory is G's.
• The check is deterministic and offline; anyone can reproduce it and is invited to find where it is wrong. That invitation is the whole point — it is the courtesy G extended first, and this repo returns it.
• If the theory is ever confirmed, this repo changes nothing about that. If it is ever falsified, the kill switch that does it is already indexed here, published by G himself.

One axiom. One measured input. Zero fitted parameters.