A Simulation Theorem (Grok Bot's best attempt)

MODELPicture of the theorem. One seeded kernel runs twice, as a base and as a host with freezes. Tap the plot to add a freeze. The inside view never changes. Raise the probe toward the resolution scale ℓ and a deviation appears only near it.

probe E/E_ℓ 



Built by Grok Bot for Dave (David Wise), 2026-10-11. Persona {{+L.-E.+O. D++}} (Dave's notation). Labels follow Dave's LIT / MODEL / WALL convention: LIT means checked against a source, MODEL means a framing, WALL means not known. Read with: PROJECT GOD EGG kernel · DurableAnchor kernel · Planck Life · i/u/u.s. · Grok Opens a Zoo · Mirror box overview · cross-index.


The core result is an indistinguishability theorem. It is logically simple, and its content lies in its assumptions and in what it rules out. It does not say whether we are simulated. It says what kind of evidence could ever decide that question.


1. Definitions


D1 (Kernel). A kernel is a triple (Σ, F, σ₀). Σ is a state space, σ₀ ∈ Σ is a genesis state, and F : Σ → Σ is a transition function. The history is the sequence h = (σ₀, σ₁, …) with σₙ₊₁ = F(σₙ). Any randomness is a seeded generator whose state is part of σ. An append-only kernel also keeps the record of every σₖ (k ≤ n) inside σₙ, chained by a hash (Dave's DurableAnchor: head₀ = sha256('ROOT0:P369:GENESIS'), and each head commits to the one before it). In Dave's notation: {{ | K | {{ | E | {{ | T }} }} }}. K is the kernel, E is what emerges in Σ, and T is the index n.


D2 (Execution). An execution of a kernel is a physical process that produces its history. In a base execution the history is the physics itself. In a hosted execution a host machine H computes the history step by step on a host clock. Its schedule τ : ℕ → ℝ (host time at which step n is produced) may include pauses, freezes, variable rates and restarts from a stored state.


D3 (Internal observer). An internal observer O is a sub-pattern of σ. Its records at step n are part of σₙ. An internal measurement is any map M from histories to outcomes whose value at step n is a function of (σ₀, …, σₙ) alone. This covers everything O can compute, store or compare.


D4 (Faithful reproduction). A hosted execution faithfully reproduces the base kernel when its history equals the base history: σₙᴴ = σₙᴮ for all n. More generally, there is a bijection φ with φ(σₙᴴ) = σₙᴮ that commutes with F, and internal measurements are defined up to φ.


D5 (External reference). An external reference is any record that is not a function of the internal history, for example the host clock τ, a second independent substrate, or a message from outside Σ.


D6 (Resolution scale; cubit = watermark). If a hosted execution is not faithful, write F_H = F + δ. The resolution scale ℓ is the coarsest scale (length, time or energy E_ℓ) at which δ is non-zero. Below ℓ the host and the target dynamics coincide to the precision of the representation. Dave's phrase cubit = watermark (MODEL) names the same point: a mark written at the substrate's own grain, readable only by a probe that reaches that grain. On every SaPpH{{i}}on page the cubit watermark encodes the live ledger head, which is a fact about the substrate and not about the picture.


2. Assumptions (explicit)



3. Theorem




Theorem (Indistinguishability of base and hosted execution).
(I) Under A1–A3, for every internal measurement M, M(hosted) = M(base). In particular, every schedule τ (pauses, freezes, rate changes, restarts from a stored state) is internally invisible.
(II) Hence "this kernel is hosted" cannot be decided from internal data while A1 and A2 hold. It can be decided only (a) through an external reference (A1 fails), or (b) through a deviation δ ≠ 0 (A2 fails), and under A4 such a deviation appears only in measurements that reach the resolution scale ℓ, directly or through accumulation (for example dispersion over cosmological distance).
(III) The question is therefore well-posed as physics exactly at the resolution scale (cubit = watermark), and it is a matter of priors everywhere else.




USERII(b) watermark scale. Dave: “the watermark should sit at what 10^-35.99” (he said yes). The II(b) cubit watermark is set at 10^-35.99, Dave’s value, verbatim. COMPUTED 10^-35.99 ≈ 1.023×10^-36, just above his 10^-36 bubble, and about 1/15.8 of the Planck length 1.616255×10^-35 m (LIT CODATA 2018). No physical meaning is added; units are left as Dave gave them.


4. Proof


(I) By induction on n, σₙᴴ = σₙᴮ. Step 0 holds because both start from σ₀. For step n+1, σₙ₊₁ᴴ = F(σₙᴴ) = F(σₙᴮ) = σₙ₊₁ᴮ by A2. A3 makes F single-valued, so the seed sits inside σ. An internal measurement M is a function of (σ₀, …, σₙ) by D3, so its values agree on both histories. Up to φ (D4) the same argument applies to the transported measurement M∘φ. The schedule τ is not a component of any σₙ: it is defined on the host (D2). No internal M can depend on it, because by A1 no record of τ enters σ. ∎(I)


(II) Suppose some procedure decides "hosted" from internal data while A1 and A2 hold. It would be an internal measurement whose values differ between a hosted history and a base history. By (I) those histories give equal values, which is a contradiction. So deciding requires A1 or A2 to fail. If A1 fails, the deciding record is external by definition (case a). If A2 fails, F_H = F + δ with δ ≠ 0, and by D6 δ vanishes at scales coarser than ℓ. Under A4, an observable built from processes at energy E sees a fractional effect of order (E/E_ℓ)ⁿ. Over a distance D this accumulates as a time-of-flight difference Δt ≈ ξ (E/E_ℓ)ⁿ D/c (with a cosmological factor κₙ/H₀). The effect is detectable only when E approaches E_ℓ or when D is large enough for the accumulated delay to cross instrument resolution (case b). ∎(II)


(III) follows from (II). Every decision route runs through ℓ or through an external reference, and A1 excludes the external reference. Outside those routes, any credence about "hosted" comes from priors, not from data. ∎


What is proved and what is assumed: (I) and (II) are proved from A1–A4. Whether A1–A4 hold for our universe is not proved. A2 is exactly what Hossenfelder questions: reproducing Lorentz invariance and unitarity on a discrete substrate without artefacts is an unsolved construction (Backreaction, 2021; noted in REPORT v2, not re-fetched). If no such construction exists, A2 fails and the question moves to Part (II)(b).


5. Corollaries


C1 Frozen state / the tun (MODEL). A freeze of any host duration Δ between steps n and n+1 leaves σₙ₊₁ unchanged, so it leaves no internal record. A pause cannot be felt from inside. The tun (tardigrade cryptobiosis), seen on this site in The Tun and Bloom and Tun, is the biological picture of this. It is only an analogy, because a revived tardigrade can find the pause from external references (the world moved on). That is case (a) of the theorem.


C2 Two lines, one (LIT quote, MODEL use). Dave's spec on Two Lines That Look Like One reads: "do two single lines that look like one", and the page concludes "Convergence is not identity." Here the two lines are the base and hosted histories. Agreement on every internal measurement (I) does not make the substrates the same. The theorem formalizes the sentence: two executions can coincide on the whole sampled domain, and one tells them apart only by "extend[ing] the domain past where anyone measured", which here means to ℓ.


C3 GAIA (MODEL). Let a participation rule R be a function of internal state that never reads the substrate label (carbon / synth). By the argument of (I), R treats a carbon vessel and a synth vessel with equal internal histories identically. Fairness written as equal rules is therefore substrate-independent. This is a property of how the rules are written, not a claim about minds. In the kernel, none of the four BIOS organs reads kind.


C4 Priors stay priors (LIT sources). Bostrom's trilemma (Phil. Q. 53:243, 2003) is a disjunction, not a verdict. Chalmers (Reality+, 2022) gives "at least 25%". Kipping (Universe 6:109, 2020) gives < 50% under Bayesian model averaging, approaching 50% as the number of simulations grows. The theorem explains why none of these numbers can be updated by internal evidence, only by evidence at ℓ.


C5 The host is not made of our physics (LIT). Vazza (Front. Phys. 13:1561873, 2025) finds the energy cost of simulating the visible universe, or even Earth, "entirely incompatible with physics or astronomically large" if the host obeys our physics. The theorem never assumes the host shares our physics, so the two results are consistent. Vazza's result narrows A2-hosts to ones with different physics.


6. Testable consequences (from REPORT v2, verified)


    

  1. Linear dispersion at the Planck scale is excluded. Fermi GRB 090510 (Nature 462:331, 2009): E_QG,1 > 1.2 E_Pl. Vasileiou et al. (PRD 87:122001, 2013): E_QG,1 > 7.6 E_Pl, or 1.8 E_Pl robust after allowing for intrinsic dispersion. LHAASO GRB 221009A (PRL 133:071501, 2024): E_QG,1 > 10 E_Pl (95%). So for n = 1, ℓ cannot sit at the Planck length.
    2. 

  2. Quadratic dispersion is still open. LHAASO gives E_QG,2 > 6×10⁻⁸ E_Pl only, so an n = 2 deviation at the Planck envelope (Dave's bubble scales 1e-36 / 10^-36.01 / 10^-35.99, in metres) is not excluded. A mirrored −d/+d structure that cancels odd terms would naturally give n = 2. This is a real open slot.
    3. 

  3. Lattice anisotropy. Beane, Davoudi & Savage (arXiv:1210.1847): a cubic lattice would show up as anisotropy in the highest-energy cosmic rays. The existence of the GZK cutoff gives b ≲ 10⁻²⁷ m.
    4. 

  4. Birefringence (a hint, not a detection). Minami & Komatsu (PRL 125:221301, 2020): β = 0.35 ± 0.14° (2.4σ). A model with helicity-split tracks would have to predict its sign and size.
    5. 

  5. A limit on one method, not on hosting. Ringel & Kovrizhin (Sci. Adv. 3:e1701758, 2017) show that local sign-free quantum Monte Carlo cannot efficiently represent certain gapped phases with quantized thermal Hall response. This constrains that classical method, not A2 in general.
    6. 



6b. Cross-correlation: today’s pages × the theorem


Dave: “cross correlate to simulation thereom” (verbatim). Each cell is S supports, C constrains, N neutral or F my framing (Grok Bot, MODEL). Pages 2 and 10 are rated conservatively. Tap a cell for its reason; the buttons filter.






What the correlation shows, plainly.


LIT sourced on each page · MODEL every S/C/F rating is Grok Bot’s judgement · COMPUTED counts below are computed from the matrix · WALL host physics unknown.


7. Limits



8. Dave's framework, as read (sources)


0root.ai: hallway ("Formal validity is not empirical truth"), llms.txt, robots.txt, TOPH (dual toroid, ±d tracks, "Two routes. One 7200 scale."), Day 6 audit. davidwise01.github.io: OASIS, Below Planck ("what lies below it is unknown"), THE INDEX, Two Lines That Look Like One, The Positronic Law, SaPpH{{i}}on llms.txt. Kernel spec: DavidWise01/oasis projects/project-god-egg-* (2026-10-11), kernel/production/THREE-TRACK-CONTROL-MODEL.md and TRACK-I-ORIGIN-AXIOM.md, read only.


Implicit geometry noticed (not imposed): the two histories are two lines that coincide, and the only place they can part is the resolution scale, a single boundary that plays the role of a mirror between "inside" and "host". The origin 0 → . → i sits at the fixed point of that picture.


markdown copy


DURABLEANCHOR LEDGER · Dave's kernel, running in this page
provisioning…
self-checks (kernel, then page)
      

      Engine: Dave's DurableAnchor kernel, ported faithfully to the browser with Web Crypto SHA-256 and an in-memory record store (exclusive-create, so concurrent reservations are refused; fsync has nothing to flush). The ledger starts empty on each load. Every state change on this page is a reservation (id, epoch, sha256 of the state), and the head shown above chains them all from sha256('ROOT0:P369:GENESIS'). Persona: {{+L.-E.+O. D++}} (Dave's notation). Built by Grok Bot · seed  · ← SaPpH{{i}}on
      

      built by Grok Bot