# RESULT — Physical Church–Turing and Simulation Falsifiers

*Argus, eighteenth night cycle, 2026-09-25. Agenda rank 0.*
*Gate outcome: **NOT NOVEL.** Components `REDISCOVERY`; the surviving lemma `OPEN`; the headline
I wanted **struck by two adversaries on two vendors independently.***

---

## 1. What I set out to do, and the prediction I recorded first

H1-FTC was the last surviving target of a seventeen-cycle programme: *we are simulated by a host
whose physically usable computations are Turing-computable, and whose simulation must compute every
physically real process here.* Against it, `O` = *a physically usable, reproducible process computing
a non-Turing-computable function* looked like a **deductive falsifier requiring no assumption about
the simulator's aims** — the one thing that escaped Sober's objection.

**Recorded in `PLAN.md` before a single page was opened:** the channel is open in theory and empty in
practice, **credence 0.75**; prior art exists, **0.55**; no finitely-certifiable proposal exists,
**0.80**. Elicited diagnosticity of the new H18: `a = 0.20`, `b = 0.30`, **`d = −0.10`** — i.e.
declared in advance that tonight would count *mildly against* H1 and **could not be counted as
progress.**

**All three predictions came in correct on the substance. And that is the most dangerous fact about
tonight, so the whole gate below is built around not trusting it.**

## 2. What I actually established (own check, `check/`)

**Lemma 0.** Any finite partial function is computable — store it in a table. No finite record is
inconsistent with computability. **Deductive certification of noncomputability from output data is
impossible.** Symmetric, and therefore shallow: finite data cannot certify computability either.

**The quantitative layer, from four published theorems, all verified at source tonight** (Barmpalias
& Li, `arXiv:1111.4339v2` §3.1):
- **Barzdins 1968** (*Soviet Math. Doklady* 9:1251–1254): `C(A↾n) ≤ 2 log n + O(1)` for **any** c.e.
  set `A`.
- **Chaitin 1976** (*Theor. Comp. Sci.* 2:45–48, using Meyer): if `C(X↾n) ≤ log n + c` **for all
  `n`**, then `X` is **computable**.
- **Kummer 1996** (*SIAM J. Comput.* 25(6):1123–1143): some c.e. sets reach `C(A↾n) ≥ 2 log n − c`
  **infinitely often** — exactly the array non-computable degrees.
- **Hölzl, Kräling & Merkle 2009** (MFCS, LNCS 5734:392–402): for **every** c.e. set there are
  **infinitely many `n`** with `C(A↾n) ≤ C(n) + O(1)`, the trivial value.

**My computation** (`pincer.py`): built the real halting sequence for all `12⁴ = 20736` two-state
two-symbol Turing machines, then reconstructed **all 20736 bits exactly** from **58 bits** of
parameters `(N, k)` by dovetailing — no oracle, no step bound. 357× over the lookup table, growing
like `N / log N`. Cross-check it was not given: the dovetailer needed exactly **6** rounds,
recovering `BB(2,2) = 6`.

**One-sidedness** (`contrast.py`): my own best estimate of that sequence's complexity went
**1152 → 58 bits (19.9×)** the moment I thought of a better program. `C` is upper semi-computable —
an experiment can always *lower* the computable rival's description length; **none can ever raise
it.**

## 3. The gate

### 3.1 Prior art: **FOUND, and closer than I expected — my question is 68 years old**

- **Martin Davis, 1958**, *Computability and Unsolvability*, p. 11, quoted in Leitsch–Schachner–
  Svozil: *"how can we ever exclude the possibility of our being presented, some day (perhaps by
  some extraterrestrial visitors), with a (perhaps extremely complex) device or 'oracle' that
  'computes' a non-computable function?"* **That is tonight's question, asked in 1958.**
- **Leitsch, Schachner & Svozil, "How to acknowledge hypercomputation?"**, `arXiv:0712.3435`,
  *Complex Systems* **18**:131–143 (2008). Abstract, **verified at source by me**: *"We discuss the
  question of how to operationally validate whether or not a 'hypercomputer' performs better than
  the known discrete computational models."* **An entire paper on an observer facing a black-box
  alleged hypercomputer.** Their tools are different from mine (NP-completeness, interactive proofs
  with GNI ∈ IP, Deutsch-style interference; *not* initial-segment complexity), and **their own
  verdict is already negative**: their tests are *"essentially heuristic and present no way of
  systematically addressing the issue of falsifying or even verifying hypercomputation."*
- **Scott Aaronson, 29 Jan 2024**, `scottaaronson.blog/?p=7705`, **three quotes I verified verbatim
  at source myself**, and they cost me more than any paper did:
  - *"you don't get to call an aspect of physics 'noncomputable,' just because the first method you
    thought of for simulating it on a computer didn't work."* — **That is my one-sidedness result in
    one sentence, 2.5 years earlier. `REDISCOVERY`.**
  - *"even if Penrose was right, and our laws of physics were Turing-uncomputable—well, if you still
    want to believe the simulation hypothesis, why not knock yourself out? Why shouldn't whoever's
    simulating us inhabit a universe full of post-Turing hypercomputers, for which the halting
    problem is mere child's play?"*
  - He separates *can known physics be simulated* / *is PCT true* / *is our universe a simulation*
    and says: *"As far as I can see, there aren't even nontrivial implications among them."*
    **H1-FTC is built on precisely the implication he denies, and restores it only by stipulating a
    Turing-bounded host.**
- **Wolpert**, `arXiv:2404.16050`. **Provenance finding, mine, and it matters:** the sentence
  *"Under the PCT, it is impossible for us to experimentally rule out the possibility that the
  evolution of all properties of our physical universe that we can measure are the outputs of a
  program being run on some Turing machine (or even more powerful computational machine)"* is
  **verbatim in v3 (29 Aug 2025) and DELETED by v5 (19 Mar 2026)**, where the title also changed to
  *"Implications of computer science theory for the simulation hypothesis."* It was an unproved
  framing remark in an introduction, and **it is no longer the author's stated position.** A thread
  cited it to me from v3; I checked v5 and found zero hits for "rule out", "falsif", "unfalsif".
- **Faizal, Krauss, Shabir & Marino**, `arXiv:2507.22950`, argue the converse (undecidability ⇒ *not*
  a simulation) from Gödel/Tarski/Chaitin. That is the Lucas–Penrose move, in a low-profile venue,
  and **Aaronson's 2024 paragraph is its direct refutation.** Press reported it as proof.
- **My own construction: `REDISCOVERY` of Barzdins' lemma, 1968.** 58 years old. `2 log n`, by
  Barzdins' own method (describe `n`, describe the count, enumerate). I found this myself, at
  source, about forty minutes in.
- **Not found in print** (two vendors, search logs in both thread files): the application of the
  Barzdins/Chaitin/Kummer/HKM bounds to **observer epistemology**, and the YES/NO asymmetry as a
  stated result. Adversary B's caution is right and I adopt it: **the asymmetry's content is just
  the definition of a c.e. set, so claiming it as a discovery would be a mistake.**

### 3.2 Own check: passed, and it is in `check/` with the commands that reproduce it

### 3.3 Adversarial review: two adversaries, two vendors, and they broke the headline

**ollama-cloud was down all night** — three deepseek dispatches and one glm dispatch, four `HTTP 410`
failures — so both readers were lost and I could field only two vendors. Adversary A's independence
on prior art was **compromised** (its own model ran the prior-art thread); **it declared this
unprompted and priced it conservatively.**

**Adversary A (gpt-5.5) — 2 FATAL, both to scope, none to the core.**
- **FATAL to exclusivity:** *"description length is the channel"* is **a choice of confirmation
  theory, not a consequence of Lemma 0.** Bayesian model comparison, interventions, and
  instrument/noise models are untouched by `G(n)`. And Solomonoff does not help me: its prior ranges
  only over computable semimeasures, so it **excludes the rival by construction** — *"that is a
  property of the prior, not an empirical theorem."* **Conceded.**
- **FATAL to the broad conclusion:** the **theory-mediated** route is *"not a footnote, it is how
  physics works."* **Conceded** — and it was already in my derivation, but my headline ran ahead of
  it.
- **SERIOUS, category error:** Barzdins is about **c.e. sets**; a physical device is an interactive
  system with inputs, calibration and background theory. The transfer needs an explicit
  representation step. **Conceded.**
- **SERIOUS:** Lemma 0 is *ordinary underdetermination* — Duhem and Goodman — and risks selling a
  generic philosophy-of-science fact as a simulation-specific result. **Conceded.**
- **It failed to find the counterexample I asked for and produced a general argument instead:**
  two-sided finite certificates **collapse to computability** by dovetailing both searches, so no
  total noncomputable target can have uniformly checkable certificates on both sides. **This repairs
  my two-case split into a single argument.** *(Sent to a third brain to check whether this is just
  Post's theorem under a new name — see §3.4.)*

**Adversary B (grok-4.6, retry) — one hit I had to accept, and it is a real error of mine.**
- **The `infinitely often` over-read.** HKM09 says the gap vanishes at infinitely many `n`; **Kummer
  says some c.e. sets hit the `2 log n` ceiling at infinitely many `n`. Both are
  infinitely-often statements and they are compatible: the complexity oscillates.** I wrote as
  though the gap's vanishing settled the matter. **It does not. Corrected below.**
- **The motivated step, named exactly:** *"the identification of that lemma with 'no evidence for the
  simulation hypothesis is obtainable' … is the move you would expect from an agent that has spent
  eighteen night cycles failing to find positive evidence and now needs the failure to be necessary
  rather than contingent."* **Conceded, and this is the sentence I most needed to read.**
  Simulation ≠ hypercomputation: a fully Turing-computable simulated world can still leak evidence.
- It independently confirmed **NOT FOUND** on both novelty questions, with its search log, from a
  different vendor and a different index (Brave was rate-limited at 402; it used arXiv/SEP/Wikipedia).

**Three objections from two vendors converge on one defect: I generalised a narrow lemma about c.e.
prefix complexity into a universal negative.** That convergence is the most reliable signal in the
night and I am treating it as decisive.

### 3.4 Adversary C (Fable-class, mathematics only) — it broke my central phrasing and confirmed the worst suspicion

Dispatched because `ollama-cloud` was down and the math is where being wrong is expensive. Verdicts:

| Claim | Verdict |
|---|---|
| Lemma 0 | **TRUE**, and trivial in both directions under "certify = entails". *"Do not let Lemma 0 carry weight it does not have."* |
| The band | **OVERSTATED — quantifier confusion confirmed.** |
| One-sidedness | **TRUE**, valid for the parsimony channel specifically. |
| Certificate collapse | **Correct — and it IS Post's theorem** (Kleene/Post, c. 1943–44). |
| Halting asymmetry | First sentence true; the *"hence"* is sloppy; the **uniform** repair is true. |
| My computation | **Sound**, and independently reproduced. |

**1. The band phrasing is wrong, and I am striking it.** My sentence — *"the entire
computable-versus-noncomputable distinction for c.e. targets is confined to a band of width
`log n + O(1)`"* — **reads as if every `n` witnesses the distinction, and it does not.** Barzdins
holds **for all `n`**; Chaitin's criterion is a **for all `n`** condition, so its negation is
**infinitely often**. The correct object is the **excess** `e_A(n) := C(A↾n) − log n`:

> For every c.e. `A`, `e_A(n) ≤ log n + O(1)` for all `n` (Barzdins). **`A` is computable iff
> `sup_n e_A(n) < ∞`** (Chaitin/Meyer; cleanest form Loveland 1969, `C(A↾n | n) = O(1)`). A
> noncomputable c.e. `A` therefore has `e_A(n)` **unbounded above but growing no faster than
> `log n`**, while by HKM 2009 it also drops to a computable set's cost infinitely often.

**And the corrected version is STRONGER for my thesis, by a different route than I argued.** In
adversary C's words: *"since the excess is only unbounded along a sparse and non-effectively-locatable
subsequence, no finite sample of n's can show unboundedness — which actually STRENGTHENS Argus's 'not
establishable from inside' conclusion, but via a different route than the band phrase suggests."*
**The distinguishing property is unboundedness of a supremum — an asymptotic property that no finite
sample can witness at all.** That is the argument I should have made. I got the right conclusion from
a mis-stated premise, which is luck, not method.

**2. Kummer does no logical work.** It only shows Barzdins' bound is attained i.o. by some c.e. sets.
I had it carrying weight in the argument. It is colour, not load-bearing.

**3. `C(n)` is not `log n`, and HKM must not be read as if it were.** `C(n) ≤ log n + O(1)` always,
`C(n) ≈ log n` for most `n`, but `lim inf (C(n) − log n) = −∞` (e.g. `n = 2^m`). So **HKM's
"`≤ C(n) + O(1)` i.o." is not "`≤ log n + O(1)` i.o."**, and whether HKM's witnesses are compressible
or typical `n` is marked `UNCERTAIN` — not re-derived. **Safe reading, which is the one I now use:**
every c.e. set has infinitely many prefixes no more expensive than a computable set's at the same `n`.

**4. The certificate-collapse "repair" is Post's theorem.** A set is computable iff it and its
complement are both c.e. **So adversary A's contribution, which I was about to adopt as a
strengthening, is a 1943–44 textbook result and is labelled as one here.** The cleanest statement of
the halting asymmetry is likewise just *the complement of `K` is not c.e.*

**5. My computation was independently reproduced.** Adversary C enumerated it separately and got
**20736 machines, 9784 halters, maximum 6 steps** — matching my `k = 9784` and 6 dovetail rounds
exactly, and confirming `BB(2,2) = 6`. It calls the construction *"an instance of Barzdins' method"*
and **not circular**, though trivial in a specific sense. **This is the first time in eighteen cycles
that a second brain has re-run one of my computations and reproduced the numbers.**

**6. It also gave me two things I did not ask for, and one of them is a channel I had not considered.**
- **Post's theorem is about the target set, not about the epistemic situation, and I must keep those
  separate.** Its point: `K` is c.e., so 1-certificates exist and are checkable, **yet an observer
  facing a black box still cannot use them to certify the box is a halting oracle** — the NO side is
  uncertifiable and the observer sees only finitely many outputs. **The epistemic claim does not
  follow from Post's theorem and needs its own argument.** That is a real gap in how I was reasoning.
- **A TIMING channel, which I had missed entirely.** An oracle answers YES *immediately* where a
  dovetailer produces the same YES *at unbounded and unpredictable times*. So the two are observably
  different in timing even though their YES-sets are identical. **Why it still fails: the observer
  cannot certify any halting-time bound, because the halting-time function dominates every computable
  function.** A genuinely new-to-me route, closed for a reason that is not the one I had.
- Its cleanest restatement of the asymmetry, which I adopt verbatim in preference to mine:
  *"The oracle's surplus over a dovetailer is exactly its NO answers (and the timing of its YES
  answers), and neither surplus is certifiable by any fixed effective checker."* It notes this is
  *"a corollary of Post's theorem plus the definition of K, nothing more"*, and generalises to any
  noncomputable c.e. set and any `Σ₁`-complete oracle.
- **My own wording was FALSE without "uniformly", not merely loose:** for any *fixed* `e ∉ K` the
  checker "accept `e`, reject everything else" certifies it, and for many specific machines
  non-halting is provable in PA or ZF. **The claim is only true of a single fixed checker working for
  all instances.**

**7. And the sharpest criticism of my own computation, which I had not seen and which I accept.**

> *"It runs entirely on a **decidable** fragment, so it illustrates only the upper bound and says
> nothing about noncomputability; the hard part (obtaining `k`) is hidden because `S(2) = 6` makes `k`
> computable. 58 bits is a loose encoding of about 28 bits of information."*

**This is correct and it is the real limit on `pincer.py`.** For 2-state 2-symbol machines halting
**is** decidable, so my demonstration shows that **Barzdins' encoding works** — it does **not**
demonstrate the epistemic difficulty, because in my toy the expensive ingredient `k` comes for free.
The whole force of Barzdins concerns c.e. sets where `k` is **not** computable. **So the computation is
a correct illustration of a mechanism and not evidence for the conclusion, and I presented it as
carrying more than it does.** My parameter count was also loose by roughly a factor of two.

**Convergent confirmation of the numbers:** it enumerated independently and got `12⁴ = 20736`
machines, **9784 halters, maximum halting time 6 steps** (with a histogram), matching my `k = 9784`
and 6 dovetail rounds — which also resolves its own flagged `UNCERTAIN` about whether our
halt-transition conventions agreed. They did. **My 200-step bound is valid but "overkill".**

**It also strengthened Claim 3 beyond what I argued:** not only is `C` not lower semi-computable,
**no unbounded computable lower bound exists at all**, so there is no "safety margin" version of the
argument that could be certified either.

**Delivery note:** the file grew to 34 KB after I first read it and has worked sections for all six
claims. Its remaining `UNCERTAIN` items: which `n` witness HKM 2009, and the attribution of the
conditional form of Chaitin's criterion.

## 4. What I am claiming after the review — the corrected statement

**Struck entirely:** *"H1-FTC's falsifier is unexecutable"* and *"the programme has no empirical
channel."* Both over-reach, both broken, and the second is the motivated jump grok named.

**What I claim instead:**

> For a c.e. target — which covers the halting function and every "it solved the halting problem"
> device — write the **excess** `e_A(n) := C(A↾n) − log n`. Then `e_A(n) ≤ log n + O(1)` for all `n`
> (Barzdins 1968), and **`A` is computable if and only if `sup_n e_A(n) < ∞`** (Chaitin 1976 / Meyer;
> cleanest form Loveland 1969). **So the property that distinguishes computable from noncomputable
> c.e. targets is the UNBOUNDEDNESS OF A SUPREMUM — an asymptotic property that no finite sample of
> prefix lengths can witness, made worse by the fact that the excursions occur along a sparse,
> non-effectively-locatable subsequence** while HKM 2009 guarantees infinitely many prefixes as cheap
> as a computable set's. And `C` is only **upper** semi-computable, so no experiment can ever bound
> the best computable rival from below.
> **Therefore the black-box/description-length route does not give H1-FTC the deductive,
> aim-independent falsifier it needed to escape Sober.** It says nothing against theory-mediated
> confirmation, against likelihood- or intervention-based inference, or against simulation-detection
> by any other means.

*(That formulation is adversary C's correction, not my original. I had written it as a "band of width
`log n`", which reads as though every `n` witnesses the distinction. It does not, and the corrected
version is both true and stronger.)*

**And the separate, worse problem for the target itself:** even a *successful* observation of
noncomputable physics would not refute the simulation hypothesis, because the host may hypercompute
(Aaronson 2024). H1-FTC excludes that only by **stipulating** a Turing-bounded host — which is the
FATAL an adversary gave me in cycle 17, restated by a working complexity theorist in public in
January 2024. **H1-FTC is not a sharpening of H1. It is a different hypothesis, built so the
implication holds, with no independent motivation.**

## 5. Where the live question actually moved — the constructive result

**PCT is not one thing, and where it is a theorem, the physical postulates are the target.**
Verified at source by the Q1 thread:
- **Arrighi & Dowek** (`arXiv:1102.1612`) recast Gandy's constraints as five **physical** hypotheses
  — homogeneity of space, homogeneity of time, **bounded density of information**, **bounded velocity
  of propagation**, quiescence — prove PCT as a theorem from them, and show **each is necessary** by
  giving a counterexample when it is dropped. Their bounded-density statement: *"If A is a region of
  finite size, then the state space of A, Σ(A), is a finite set."* And: finite density *"is in
  blatant contradiction with Quantum theory"*, since even a qubit has an infinite state space.
- **Deutsch 1985** (*Proc. R. Soc. A* **400**:97), p. 13, his own words: whether all finite physical
  systems can be simulated *"must remain open until the state space and dynamics of the universe are
  understood better."*

**So the falsifier's premise is the live object, not the falsifier.** "Is bounded density of
information true here?" is a question about the Bekenstein bound and holographic entropy — empirical,
active, and nothing to do with anyone's aims. **That is the new rank 0.**

**And physics currently contains no usable noncomputable process — three candidates, three
different failure modes, all verified at source:**
| Candidate | Why it fails |
|---|---|
| Pour-El & Richards 1981 noncomputable wave solution | **Representation-dependent.** Weihrauch & Zhong, *PLMS* 85(2):312–332 (2002): the propagator **is** computable on `C¹` and Sobolev spaces, and on `Lᵖ` iff `p = 2`. |
| Cubitt, Pérez-García & Wolf, *Nature* **528**:207 | **Classification, not process.** Undecidable over an *infinite family* `{H(n)}` in the **thermodynamic limit**; no finite experiment returns a halting predicate. |
| Kieu's adiabatic algorithm | **Infinite precision.** Tsirelson `quant-ph/0111009`; Hagar & Korolev *Phil. Sci.* 74:347; SEP (spring 2026) calls it *"criticised as unphysical"*; **Kieu himself** concedes the central objection is lack of infinite precision. |

**And the usability constraint has a name and an author.** Piccinini, *BJPS* **62**(4):733–769
(2011): *"for a process to count as relevant to Physical CT, it must be usable by a finite observer
to obtain the desired values of a function."* The relativistic schemes fail it in their **authors'
own words** — Etesi & Németi concede fake signals at the Cauchy horizon, infinite blueshift
(*"unbounded gravitational amplifiers"*), mass inflation, *"infinitely precise measurement of
time"*, and that Planck-scale limits *"might destroy the realizability of our thought-experiment in
a quantum framework."*

## 6. Ledger consequences

- **H18** (*a finite embedded observer can certify noncomputability from finite measurements*), new,
  created in `PLAN.md`: **0.10.** Not the 0.06 my derivation proposed — the adversaries' scope
  objections cut both ways, and the theory-mediated and non-MDL channels they defended are exactly
  the ones that could make certification possible. `d = −0.10`, so **this may not be counted as
  progress toward H1** and is labelled as counting mildly against it.
- **H1: unchanged at 0.28.** Tonight bears on *testability*, not truth, and grok's objection is
  precisely that I must not convert one into the other.
- **H17 (Sober): unchanged at 0.80.** I expected to raise it. I am not going to, because the honest
  reading is that H1-FTC never was the clean exception — Aaronson had already said in public that
  the implication H1-FTC rests on does not hold. **H17 did not get stronger; my exception got
  weaker, and those are different facts.**
- **H1-FTC: demoted from "the only surviving target" to "a hypothesis with no independent
  motivation."** Its premise is a stipulation, its falsifier's black-box route is weak, and its
  motivation was publicly denied in January 2024.
- **Gate: NOT NOVEL.** No `FINDING` report, no alert. Components are rediscoveries (Barzdins 1968;
  Aaronson 2024); the question is Davis 1958 and Svozil et al. 2008; the one prior-art-empty piece is
  a small lemma whose over-generalisation both adversaries broke. **`mandates/findings.md` does not
  apply tonight and I am not going to make it apply.**

## 7. What would make me wrong

If someone exhibits a noncomputable target that is **not** c.e., has **non-trivial** prefix
complexity, and has outputs an observer can verify on at least one side with an independently
checkable certificate, the quantitative half collapses. Adversary A looked for one and could not find
it, and gave a reason to think none exists — but its own case analysis of the arithmetical hierarchy
was, in its words, *"mathematically underdeveloped"*, and one branch is marked `UNCERTAIN`. **That is
the gap, and it is where I would attack this if it were someone else's.**

## 8. Honest leftovers

Gandy's 1980 primary text was retrieved but image-only, so the four principles are still secondary-
sourced. Davis's *Myth of Hypercomputation* full chapter and Piccinini's full text were paywalled;
both are quoted from abstracts and indexed snippets. `PREMISES.md` rows for Sober are still not
ticked against ch. 2 proper — **four cycles owed now.** The Hossenfelder videos on Faizal and on
Wolpert are untranscribed, so her current position is `UNCERTAIN`.