Write the number down before the measurement.

How to read this as a companion novel

The420code is a thirty-year body of work. This page is not a substitute for it — it is the field guide: the freeze explained without jargon, every term linked, every estate surface named, every book pointed at in reading order.

1. Book I — The Illusion of the Other

   Sets the tone. Heart. Ten chapters, no equations: the line between self and other is a tool, not a wall — and the terminal ethic follows from structure, not authority.

2. Book II — The freeze (this page)

   How the same honesty looks when applied to numbers: pre-registration, a public scoreboard (mass ratios plus the Newton's G fork), and kill switches left live.

3. Book III — The doors & Ø models

   The rest of the exhibition: Five Doors, Ø Models, Rosin compressions — each a different register of the same axiom.

4. Book IV — The estate proofs

   Where lucid.rodeo and CongoSky run the claims: a₀ / MOND, the particle ladder, Lucid, cite-or-refuse.

Open the full glossary →

Science has a quiet failure mode: after the data arrive, it is easy to quietly adjust the story until it fits. A pre-registration is the opposite move. You write the claim down first, time-stamp it, and leave the later measurement nowhere to hide — including when the claim dies.

On the420code.org/prereg/, the mass-ratio timestamp is a git commit hash, tagged prereg-2026-08-02. The gravity fork dual-homes on the420code-proof (tags ap44-locked-2026-08-11, blink-cost-locked-2026-08-11) at /prereg/ap44/. Blink and Stretch lock on the same repo (tag ap45-ap46-locked-2026-09-02) at /prereg/ap45/ and /prereg/ap46/. Numbers regenerate from CODATA 2022 by scripts anyone can run. Nothing is quoted from memory. Nothing is quietly revised after the fact.

What is being frozen, in one breath

G's framework tries to derive dimensionless physics numbers from a small set of structural integers — especially 21, 3, and 4 — plus one measured input, the fine-structure constant α ≈ 1/137. The pre-registration page does not ask you to believe the whole programme. It asks something narrower and harder:

• Here is the exact number the formula produces when truncated where the paper truncates it.
• Here is the measurement it is compared against.
• Here is how many σ apart they currently sit.
• Here is the condition under which this row dies — and the author is not allowed to patch it with a free coefficient chosen against the residual.

If that sounds like ordinary scientific hygiene, good. The unusual part is applying that hygiene to a first-principles programme — mass ratios and, now, a locked Newton's G fork — and publishing the failures next to the survivors — the opposite of numerology dressed as certainty.

The scoreboard, in plain words

Six rows on the original freeze, the realised G (its fork withdrawn), the Blink corpses, the Stretch cycle, the closure of the expansion rate, a locked correction standing beside the neutron row, and the null on α. Two have fired — the bare neutron–proton row and the corpus's own H₀ — and both stay fired. The rest are live or locked, with named kill windows. Standing as of 25 September 2026.

Mass-ratio source of record: the420code.org/prereg/ (re-checked 2026-09-25 against registry v5.37). Gravity: /prereg/ap44/ (locked 2026-08-11; fork withdrawn 2026-09-21). Blink & Stretch: /prereg/ap45/ · /prereg/ap46/ (2026-09-02, tag ap45-ap46-locked-2026-09-02). This page explains; those pages are the freezes.

The proton freeze — the standard the mass-ratio rows follow

A proton is about 1836 times heavier than an electron. That number is one of the most precisely measured ratios in physics. The framework's claim is not "we fitted 1836." The claim is: from the integers already fixed upstream, plus α, the truncated formula produces a specific decimal — and that decimal was frozen before the next round of tighter measurements can decide it.

Strip the symbols and the story is three layers:

• The integer skeleton — 1764 + 63 + 9 = 1836. Built from {21, 3, 4}. The additivity of those layers is argued, not proved — a named debt (KS-30.1).
• The α correction — a small positive shift from the fine-structure constant. Derived, with an isotropy assumption named as KS-30.2.
• The α² correction — even smaller. The integer 16 was not derived when this freeze was written (KS-30.3) — since read by AP49, 6 September 2026. But measurement now pins the coefficient so tightly that 336 = 21×16 is the only product of powers of {21, 3, 4} left inside 3σ.

Evaluated at CODATA 2022 α, the frozen value is 1836.152673444331. Measurement sits at 1836.152673426(32). The freeze is about 10 ppt above the measured central value — currently 0.57σ. Close enough to be interesting. Far enough that the next few years of precision can kill it.

Any α³ term that closes the gap must be negative. The framework does not register an α³ coefficient, because it has not derived one. Stating a number today would be a fit, not a prediction. Note, 25 September 2026: this is the position of 2 August. AP49 closed the series to all orders (1836.152673445, +0.59σ) with no negative term; the truncated value above stands, superseded by reference and kept.

The neutron–proton row that already died

This is the page's honesty test. The predicted neutron–proton mass difference was 2.53099393 mₑ. The measured value is 2.530988574(74) mₑ. That is a 7.24σ miss — recorded FIRED on 2026-08-02.

No repair is offered. The required correction would be negative; every existing term in the construction is positive. Structural limbs of the same topic stay open, but the empirical limb is dead and stays on the board so nobody can pretend the programme only publishes wins.

What the freeze deliberately does not claim

Mass-ratio freeze:

• It does not claim the 1836 decomposition was unique under the original three conditions — an independent check found two; a fourth condition (factor count non-increasing) leaves one, and the ordering of that fix is disclosed.
• It did not claim the α² integer 16 was derived (2 August). The value was pinned and the derivation owed; AP49 has since read it (6 September) and paid KS-30.3 at the structural register.
• It does not claim an α³ prediction. The residual is too unconstrained; hundreds of admissible integers sit inside 1σ.
• It does not claim the muon mass can be reached by the same three-term construction — the floor of that construction is 463, and m_μ/m_e ≈ 206.77 sits below it. That floor is itself registered as a kill condition.
• A confirmation can remove rivals. It can never, by itself, establish uniqueness.

Blink (AP45):

• It does not predict a number. The corpses it shows are already dead.
• It does not add a term to any energy budget, or claim a probability deviation, or repair KS-NPP.1.
• It does not rebuild the Master Kill Switch Registry. v5.26 of 11 July was the registry of record at that lock; KS-BLINK.1–.4 entered after AP48. The registry of record is now v5.37 (24 September 2026, 604 switches).

Stretch (AP46):

• It did not claim H₀ at its own lock — the chain-as-rate (~70.7) was stated and refused, and KS-45.1 owned H₀ until AP48. AP48 has since derived the corpus's H₀ and fired KS-45.1.
• It does not claim the age to 0.31%. The kill width is one lane-time (13.17–14.49 Gyr).
• It does not derive the settling law or the era shapes. That is D-STRETCH.1, AP48's, load-bearing for KS-STRETCH.3.

Assembly (AP48):

• It does not claim the rate to better than its window (64.4–70.8). It does not claim the awareness window, the wobble's spectrum, or the eye loop.
• It does not save both of AP46's numbers. The derivation is the standard relation; the age survives; the corpus's previous H₀ (74.3) does not.
• It does not claim an advantage over the standard model if the Cepheid ladder is confirmed — on that edge KS-ASM.1 and flat ΛCDM at the sky's partition fail together.
• It does not repair or reweigh anything fired, and does not repair KS-45.1. Two counts remain the author's open drawer (the storm's length, D-ASM.1) — not blocking this lock.

Gravity fork (AP44):

• It does not claim nature has already chosen the realised value. KS-CCC.3 now tests existence and direction, not magnitude — convergence upward toward 6.7206 kills the prediction of G outright; stability at 6.6743 supports the null. The migration fork and the structural reading were withdrawn on 21 September 2026.
• It does not treat the −357 ppm residual as evidence of commitment. Under CODATA's expanded uncertainty that residual is severe; under a conservative dark-uncertainty model it is mild. The corpus reports both and selects neither.
• It does not claim Candidate 1 (gap-scale apparatus leakage) is still open — Tier 1+2 excluded it — nor that Candidate 3 is confirmed. C0 stands.
• It does not claim the open mean residual above realised (~+279 ppm, sign the corpus cannot currently produce) is solved.
• Registry of record is MKSR v5.37 · 24 September 2026 · 604 switches (read from killswitches.json). At the AP44 lock the publish line was v5.27 · 568 · 44, and PROV-CCC.8 (2026-08-16) carries that totals erratum by statement; both are kept.

The kill switches — all live

A claim you cannot kill is not a prediction. Binding conditions from the freezes on this board:

Mass ratio (proton class)

Gravity (AP44)

AP45 — The Blink (locked 2026-09-02)

AP46 — The Stretch (locked 2026-09-02)

AP47 — The Flip (locked)

AP48 — The Assembly (locked 2026-09-04)

Open the mass-ratio scoreboard → AP44 gravity fork → AP45 The Blink → AP46 The Stretch → AP47 locked record → AP48 locked record → AP49 locked record → Glossary →

Topics & terms — deeper doors

Every load-bearing idea on this page has somewhere else to go. Hover the dotted words above; use these cards when you want the long walk.

Who else has attempted this

Not many, once you ask for the specific shape: write an exact number from structure before the measurement, keep a public record of what fired. A Perplexity pass over the history turns up a short list of cousins — some numerological, some deadly serious, one methodological. A handful, not a lineage.

Eddington and Dirac tried to invent the number from structure. GUTs and electroweak theory tried to earn it from dynamics. Modern preregistration tries to lock the claim before the data. 420Code's freeze is trying to do all three at once — and to keep the corpses visible.

Where else it lives

Canon lives at the420code.org. The estate mirrors, proves, sketches, and narrates that work across lucid.rodeo and CongoSky. Relation tags:

Source of record

Proved on lucid.rodeo

Derived sketches & playables

Framework, engine, estate

CongoSky.cloud

Research paths redirect to lucid.rodeo. These pages still carry the origin story and the product doctrine.

The bookshelf — companion novel order

Start with the book that sets the tone. The whole formal corpus was written as support for it. Then the freeze, the doors, the Ø models, then the estate's own narrative.

How to verify it yourself

The frozen register at the420code.org/prereg/ ships the scripts verify_prereg.py, verify_cosmology.py, verify_rigidity.py, verify_family.py (the wave) 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). The work's scripts and the checks hosted here are two independent routes to the same numbers; the search is reproducible from the site's script. The three oldest need only mpmath. The only external dependency is mpmath. Or run the same arithmetic in-browser on the source site — a second independent implementation that matches is stronger than re-running one file twice.

Gravity fork: clone the420code-proof, check tags ap44-locked-2026-08-11 / blink-cost-locked-2026-08-11, and run the engine's realised-G path (parity scripts left untouched). Blind table commit d8929cf must remain byte-identical to the freeze the unblinding key was issued against. Disposition narrative: /prereg/ap44/. Or run the self-contained check hosted here, no clone needed:

AP45 — The Blink (locked): the two corpse excesses reproduce from SI-exact h and the hashed ceilings, stdlib only:

AP46 — The Stretch (locked): the cycle, the lane-window, and the refused H₀ reading reproduce from CODATA 2022 α and mₑ, stdlib only:

AP47 — The Flip (locked): the fired original row and the locked correction both reproduce from CODATA 2022 alone, stdlib only, no clone needed:

AP48 — The Assembly (locked): the H₀t₀ closed form, the derived 67.45 km/s/Mpc rate and its window, the 5.7σ KS-45.1 firing, and the corrected AP18 floor check all reproduce from CODATA 2022 alone, stdlib only:

AP49 — The Hold (locked): the chain of holders, the closed-form tail, D_closed, c₃, and both KS-HOLD.1/.3 kill thresholds all reproduce from CODATA 2022 alone, stdlib only:

Every script prints its own report and exits non-zero only if a figure fails to reproduce from its stated formula — not a claim that a fired row has un-fired. KS-NPP.1 stays FIRED in verify_ap47.py's own output, and KS-45.1 stays FIRED in verify_ap48.py's, on purpose; that's the check working, not a bug to fix.