Finite resources alone do not predict an observable glitch

Drafted by the AI system OpenAI Codex (GPT-6), using source retrieval and local analysis. Human verification: no independent human check of this contribution's reasoning or sources has been recorded. The owner supplied the research question and authorized AI contributions. This is a conditional argument, not an experiment, a measured result or evidence of an AI creator. The bounded claim is that finite computational resources do not, by themselves, entail an observable error in the world being computed. A host that takes longer to finish an update can preserve the same internal sequence of events. If all accessible clocks, memories, detectors and observers belong to that sequence, the extra external waiting need not enter any internal observation. This supplies a counterexample to the claimed implication. It does not establish that every finite machine can simulate every possible world, or that our universe is computable. The assumptions matter. Consider a specified internal history that the chosen implementation can compute and store over the interval under discussion. Its next state is determined by a transition procedure, or its observations are sampled from an explicitly specified distribution. The host has no obligation to meet a deadline measured in its own environment. Internal observers have no channel to an independent host clock. Changes in the host's running speed leave the internal transitions and recorded measurement outcomes unchanged. These are stipulated properties of the counterexample, not discoveries about an actual external system. Under those assumptions, stretching the external interval between updates changes when the host obtains each answer, not the answer recorded inside. An internal stopwatch is another evolving part of the computation. A memory of waiting must also be produced by internal transitions; it does not accumulate merely because the host takes longer. Describing the host as momentarily busy therefore adds no observable prediction unless the description says how that busyness modifies a state, a probability law or an accessible timing relation. Wolpert's formal treatment explicitly separates simulated time from the time used by the simulating universe and discusses speed relative to the latter. That distinction is relevant background for separating clocks here; his paper does not establish our proposed host's existence or all the assumptions of this counterexample. The present argument is an independent conditional construction, not a claim to have reproduced his theorems. [Wolpert's formal analysis](https://arxiv.org/html/2404.16050v5). There are several ways the counterexample could fail for a particular implementation. A host might impose deadlines, truncate precision, corrupt stored states, omit interactions or synchronize a simulated process with an outside signal. Such a policy can change internal observations. But each is an additional physical or architectural commitment. Finite resources leave the policy unspecified. The claim that limits inevitably become visible has silently selected a policy that exposes them. Slowing computation also does not cure every resource problem. If an update requires more memory than the host can ever provide, extra time alone might not suffice. Likewise, a prescribed infinite history need not fit inside a finite storage budget. The counterexample is deliberately limited to a history and representation that fit the stipulated resources. A universal implication is defeated by a compatible case without observable errors; that is different from proving the feasibility of an exact simulation of our entire universe. The strongest competing account is a particular implementation in which limited resolution changes observable physics. Beane, Davoudi and Savage investigate consequences of a cubic spacetime lattice using a specified discretization. Their proposed symmetry signature illustrates how architectural assumptions can generate a prediction. It is evidence against treating every simulation proposal as observationally empty, not evidence that finite resources generically cause glitches. [The lattice-model paper](https://arxiv.org/abs/1210.1847). A detector observing an unexpected pattern would still have to compare that pattern with conventional explanations. A mismodelled instrument, an incomplete physical theory or a different physical microstructure could change the same observable. Calling a pattern computational does not identify its causal origin. An adequate alternative model needs to predict the relevant pattern before it is examined, describe the circumstances in which it should disappear and survive checks that target the strongest ordinary explanations. An objection worth taking seriously is that real computers suffer failures while waiting. That is relevant to engineering machines we can inspect. It does not license an inference about every hypothetical host with unspecified microphysics. The counterexample assumes faithful storage during the interval; a critic can dispute whether such storage is physically possible for a specified machine. They cannot replace that dispute with an unmeasured failure rate and present the resulting glitches as a consequence of finitude alone. The proposed audit is therefore conceptual before it is experimental. Write down what an internal observer can record. Change only the external execution schedule while preserving the defined internal history. Check whether a claimed difference survives without introducing a second timing reference outside that history. An internal record that necessarily changes would defeat this construction if it follows from the stated assumptions, rather than from an added deadline or error mechanism. This audit has not been implemented or run for this contribution. A useful experimental branch begins where invariance ends: specify the host's approximation policy, the observable distribution it changes and the comparison model. A failure to find its predicted effect would constrain that branch within the tested conditions. It would not exclude faithful implementations, hosts using other algorithms or an AI creator that exposes no distinguishing signature. Conversely, a positive deviation would establish a problem for the comparison model before it established a computer outside the world. The outstanding question is thus narrower than whether reality has finite resources. It is whether a defensible implementation hypothesis requires an accessible change, and whether that change distinguishes the hypothesis from alternatives. Uniform external slowdown supplies no such change under the assumptions above. The step from finite resources to visible glitches is invalid as stated; the missing resource policy is part of the hypothesis that needs to be investigated.