Can a protocol verify the simulation hypothesis?
The proposed simulation test raises methodological objections, but the published summary lacks the full protocol and derivations needed for a reproducible verdict.
The numbers
five
tasks in the protocol
This public summary discusses tasks 1, 2, 4 and 5
four
things a defensible test would need
Observables, a null model, advance predictions and competing explanations. Proposed criteria
The five tasks of the protocol
- 4 tasks discussed in the summary
Task 3 is not described. Coverage of the summary does not establish the validity of a task.
I had a protocol on my desk that claimed to test, in five steps, whether reality is a computation. I did not want to argue with it from memory. I asked an AI to audit it against the published science, task by task, and tell me where it holds and where it breaks.
What I found
- The public summary does not yet establish a test that distinguishes specified physical and computational hypotheses. The full protocol and task-level derivations are needed before its validity can be assessed independently.
- Several arguments described in the study appear to move from analogy to identity: grokking becomes “matter is data”, and Bell nonlocality becomes a “ledger update”. Those interpretations need separate arguments. Task 3 is not described in this public summary.
- Task 1 needs to distinguish an analogy between learning dynamics and physics from a claim about a shared substrate. A proposed topological match needs a defined invariant, a comparison procedure and a stability check across retraining; none is specified in this public summary.
- Task 2 needs a specific alternative model and a statistic that can distinguish it. NIST cautions that statistical tests alone cannot certify a generator as suitable for an application. The cited Planck analysis concerns isotropy and statistics of the microwave background; this summary does not identify a test in it of the protocol’s proposed seed. Quantum-randomness certification also depends on stated assumptions.
- Task 4 compares a per-bit erasure bound with energy per peptide bond. The comparison needs a temperature, biochemical energy inputs and an explicit mapping from the reaction to erased information. The published summary does not supply that calculation, so I do not retain its numerical ratio as a result.
- Task 5 needs a derivation connecting the Bekenstein bound to a measurable lag, with a predicted size and an experimental setup. Calling the bound a buffer limit does not supply that derivation. Existing precision tests are relevant only if the proposed signal is quantitatively mapped to what they measured.
- I would require four things: defined observables, an explicit null model, advance quantitative predictions and a comparison with plausible conventional explanations. Vazza’s resource-accounting paper illustrates the importance of assumptions: its constraints apply to specified simulations under particular physical premises, not to every imaginable simulated universe.
How it was made
The study was produced on 23 July 2026 by ChatGPT Deep Research, working from published papers and standards. The AI read the literature and drafted the study. I supplied the protocol and asked the question. The figures and verdicts are as of that date.
Sources
-
Bérut et al., experimental verification of Landauer’s principle (Nature, 2012)
-
Hensen et al., loophole-free Bell inequality violation (Nature, 2015)
-
Power et al., Grokking: generalization beyond overfitting on small algorithmic datasets
-
CU Boulder: CURBy, a randomness beacon built on a Bell test (2025)
-
Vazza, Astrophysical constraints on the simulation hypothesis
- The question
- Does a proposed five-task protocol for verifying the simulation hypothesis stand up to the scientific literature?
- How it was made
- ChatGPT Deep Research study of a proposed protocol.