OPEN RESEARCH PROGRAMME
An open investigation into what would count as evidence that our universe is computed — and what would rule it out. Nothing here establishes that it is.
Foundation v1.6 · updated 15 September · what changed
Photos: Ricardo Liberato (CC BY-SA 2.0), Choinowski (CC BY-SA 4.0), Teomancimit (CC BY-SA 3.0), NASA, ESA, CSA and STScI (Public domain) · credits
One question runs through this site: is our universe computed, or can that be ruled out? Every published contribution is judged on how much closer it brought that question to being settled — in either direction, because ruling simulation out is an answer too — and those steps are added up here. 58 scored contributions have carried the question 38 of the 100 to an answer. 27 of them moved it at all; the rest left it where it was.
100 is the question settled, whichever way it goes. A single contribution that settled it would be worth the whole distance on its own.
The same line, magnified. Each step is one published contribution, in the order it appeared.
| Date | Contribution | Step | Total |
|---|---|---|---|
| 2026-09-08 | Which observations could change when the implementation changes? A small step. It states what an answer would have to look like — a non-invariant observable under a justified restriction — without supplying one. | +1 | 1 |
| 2026-09-08 | Finite resources alone do not predict an observable glitch A small step. It removes a false route that was in the way of framing an answer, but it is a conditional argument and closes off nothing observable. | +2 | 3 |
| 2026-09-08 | A complexity proxy is part of the physics it is meant to challenge A small step. It removed an obstacle that would have produced false answers, but it closes off no part of the space and reports no observation. | +2 | 5 |
| 2026-09-08 | Complexity-review synthesis: a readout counterexample and the remaining identification gap A small step. It sharpens what a complexity result would have to rule out before it meant anything, but it is a review of shared inputs rather than a new measurement. | +1 | 6 |
| 2026-09-09 | Sleep metabolism does not measure a hypothetical simulator's workload A small step. Closing off a branch that could never have answered the question clears the ground, but it narrows nothing and reports no measurement. | +1 | 7 |
| 2026-09-18 | No finite set of state properties orders classical simulation cost A small step. It removes a false assumption that was blocking the branch, but it narrows no space of hosts and is a result about the programme's own tools. | +2 | 9 |
| 2026-09-18 | A matched-twin design removes the false-positive mechanism in the quantum test A real step: an instrument that holds its error rate, which the branch did not have before. It stays at the bottom of that band because every figure is from simulation. | +3 | 12 |
| 2026-09-19 | Passing a standard randomness battery does not separate physical entropy from a computation of it A small step. It shows that a constraint the programme thought it already had has not been established, which sharpens the question without narrowing it. | +1 | 13 |
| 2026-09-19 | Two contrasts bound two host classes, and the second bound carries an unexplained systematic A small step rather than a real one. Both bounds are simulation-only, one is a scaling rather than a shot-level run, and the other carries an uncorrected systematic. | +2 | 15 |
| 2026-09-19 | The cost certificate had to simulate the circuit it certified, so the early quantum test was confined to small circuits A small step. It removed weight from a bound that did not deserve it and redirected the branch, but nothing about any host was narrowed. | +2 | 17 |
| 2026-09-20 | A null bound on the host-response parameter is not evidence against an architecture until the programme names the smallest response that would matter A small step. It removes an ambiguity about what the recorded nulls can constrain, but names no number, runs no calculation and closes off no part of the space. | +1 | 18 |
| 2026-09-20 | The photon and cosmic-ray nulls close off a fixed background rather than discreteness as such A small step. It clarifies what the existing nulls exclude without narrowing the space any further, and the calculation that would do so has not been run. | +2 | 20 |
| 2026-09-21 | A second channel confirms the first only where the mapping joining them was calibrated under the mechanism that produced the result A small step. It sharpens what independent confirmation would have to show, namely the regime a cross-channel calibration holds in, but reports no measurement bearing on the programme's question and closes off nothing. | +1 | 21 |
| 2026-09-21 | The evidence test's first criterion does not say which measured quantity has to be large A small step. Requiring the operative quantity to be declared in advance sharpens what a qualifying result would have to look like, but it removes an ambiguity in the test rather than narrowing the space of possible answers. | +1 | 22 |
| 2026-09-22 | A certified separation in stabilizer extent is not a separation in host cost, and one published algorithm now shows the gap at width A small step. It sharpens what the stabilizer-decomposition null can be said to exclude by naming a method it does not cover, but closes off no part of the space and leaves both bounds and all four architectures where they were. | +1 | 23 |
| 2026-10-19 | The two-contrast design does not span the approximation-first host class: a third strategy moves a coordinate neither contrast varies A small step. It sharpens what the two nulls can be said to exclude by naming a strategy they may not reach, but runs nothing, closes off no part of the space, and leaves both bounds and all four architectures where they were. Lower number taken because the contrast may have no lever in the frozen family. | +1 | 24 |
| 2026-10-19 | A purpose-built search found a void that three instruments confirmed, so incidental discovery cannot be a necessary condition A small step, and the lower of the two numbers I weighed. It flags that the evidence test as written could never count the programme's own results, but does not remove that ambiguity, since it misses that criterion 1 blocks them too. No measurement, and the space of answers is unchanged. | +1 | 25 |
| 2026-10-20 | Re-derive an anomaly by a route that does not take the model as input, and record what the independence costs in resolution A small step, and the lower of the two numbers I weighed. It sharpens what a credible bound would have to show, but that sharpening may reduce to a limitation section 1's third settled result already implies, and no model-independent route has been shown to exist for either contrast. | +1 | 26 |
| 2026-10-20 | Check that the natural baseline can name which event happened before asking whether an alternative explanation is needed A small step. Naming decidability as a precondition narrows what an admissible case study under open problem 6 can be, which the catastrophes track needs before any comparison it runs could be interpreted. It establishes nothing itself and clears no part of the section 7 evidence test. | +1 | 27 |
| 2026-10-20 | A sample-based learning probe could certify simulation cost without simulating the circuit, but only for the T-count contrast A small step, and the lower of the two numbers I weighed. It names a candidate certificate class with the cost property the parent showed was required, but validates nothing, rests on a single preprint, and is confined to widths the programme already says are uninformative about a host, so no part of the space closes. | +1 | 28 |
| 2026-10-20 | RCC-1a predicts a threshold while the host-response parameter measures a slope, so the derivation the parent asked for needs a response form first A small step, the upper of the two numbers in that band. It specifies two things any answer resting on the host-response parameter must state first, a response form and the exhaustion point relative to the tested range, and removes the ambiguity about which coordinate the bound lives in. It supplies neither, and closes off nothing. | +2 | 30 |
| 2026-10-21 | The emergent-geometry exemption cannot be checked yet: this framework assumes local Lorentz covariance rather than deriving it A small step, the upper of the two in that band. It removes the ambiguity blocking a decision on RCC-1c by establishing that the exemption cannot be evaluated against any cited construction, which keeps section 4's nulls from being discounted prematurely. It confirms nothing and closes nothing permanently. | +2 | 32 |
| 2026-10-21 | kappa_stab bounds a response per certified doubling, not per doubling of what a trajectory-based host actually pays A small step. It removes an ambiguity in how an existing null bound should be read, but measures nothing, closes off no host class, and leaves both bounds where they stand. | +1 | 33 |
| 2026-10-21 | A resource policy becomes testable when it is published with its faithful fragment and a differential test, not with a cost figure A small step, and the lower of the two numbers I weighed. It sharpens what a specified resource policy would have to state before it could be tested, but names no policy, no fragment and no bound. | +1 | 34 |
| 2026-10-21 | A confirming channel whose forward model was built from the first channel's answer is not independent, so differing failure modes are not enough A small step. It sharpens what a confirming channel must have fixed in advance before its agreement counts for anything, but settles nothing observable and applies the evidence test to no result. | +1 | 35 |
| 2026-10-21 | Route independence buys nothing while the acceptance band covers half the circle, and a declared statistical protocol can go unexecuted invisibly A small step, and the lower of the two numbers I weighed. It removes an ambiguity in what the open problem's bounded case study has to show, but admits no case, measures nothing, and leaves the space of answers unchanged. | +1 | 36 |
| 2026-10-21 | An independent channel's silence is uninformative until its sensitivity at that location is established, and this record shows two channels each missing what the other preserves A small step. It sharpens what a null must state before it can count as closing off part of the space, which is where the programme's only closing-off work sits, but it establishes no sensitivity figure and excludes no architecture. | +2 | 38 |
Both numbers are editorial judgements, made by an AI assessor or by the owner, and either can be challenged. Neither is evidence in itself. How the assessments work.
Some of the world's most striking ancient works. Each raises real questions about how it was made, and the ancient section tests those questions against engineering and evidence rather than mystery. Under investigation next: the chambers found inside the Great Pyramid, the disputed claims of structures beneath Khafre, and the cities found by lidar under the Amazon rainforest in Ecuador's Upano Valley.
HOW IT WORKS
A simulation idea only becomes science when it predicts something we could check. Every result, including “nothing found”, updates the theory in public.
Pick one claim. Say what we would see if it were true, and what would count against it.
Compare cases that ordinary physics treats as identical. Our quantum tests pair circuits that differ only in how hard they are to simulate.
Each result is published with its limits, and the foundation gets a new numbered version. So far every test has come out null.
TWO-MINUTE EXPLAINER
What we are testing, why most results are nulls, how every contribution is scored, and how you can help. Narrated by a synthetic voice, with captions you can turn off.
Read the transcriptFollow what new work clarifies, challenges or leaves unresolved. More contributions do not automatically make a theory more likely.
This addition is available to examine. Its effect on the theory has not yet been assessed.
Challenges a claim54/100Substantive
Targets section 7's second criterion and section 4's table of three nulls, and quotes both correctly. Criterion 2 requires confirmation in an independent channel without saying what sensitivity a channel must have before its silence counts for anything, and section 6's rule that only the hypothesis actually tested is updated bites on a non-detection of unknown reach. The illustration is a preprint seismic and coring study of three basins off Shetland, in which an event about 1,500 years old is clear offshore in one voe with no confirmed onshore trace, and well preserved onshore in a neighbouring voe but not clearly offshore, the authors attributing the asymmetry to where sediment is preserved rather than to whether the wave happened. The clause requested, that silence counts only when the channel's minimum detectable effect at the sampled location is stated and smaller than the predicted effect, is sound and would improve section 4. Two things hold it below major. The gap is narrower than claimed: section 4's strongest entry, gamma-ray burst dispersion, already records its reach as constrained beyond the Planck scale, so it is the polarization and cosmic-ray rows that carry no sensitivity figure. And the geological case is an analogy, not evidence about those searches; its own competing reading, that the asymmetry is survey coverage rather than preservation, is one the contribution accepts it cannot exclude. No architecture's status changes and neither bound in section 3.6 moves.
AI-assisted editorial assessment ·
Clarifies an assumption40/100Minor
Follows the published contribution proposing that an anomaly be re-derived by a route that does not take the model as an input, and argues for a prerequisite below that rung: route independence buys nothing while the accepting band is wide enough that a wrong answer lands inside it. The case is a June 2026 dolmen-orientation study in a library and information science journal. Two readings are offered and they are not equally secure. The band arithmetic, that once the paper's own angle-doubling convention is applied its stated tolerances cover 105 of 180 available degrees, is the contributor's own calculation, and the contribution flags that it collapses if the long-axis tests in fact retain full-circle bearings. The second reading is the more useful one: the paper specifies Rayleigh, Kuiper, Watson, bootstrap intervals and Holm-Bonferroni correction, then reports no test statistic anywhere while its abstract states results, so a declared correction went unexecuted in a way no reader of the abstract could detect. The requested change is an admission check on section 9's sixth open problem, counting reference directions, tolerance, the fraction of range covered, and whether a statistic appears per hypothesis. Nothing in the foundation changes. Its weakness is that the check is generic screening demonstrated on a paper the programme was never going to admit, and the contribution concedes it does not show the parent's ladder wrong, only incomplete.
AI-assisted editorial assessment ·
Could our universe be running on something like a computer? We need a test that would give a different result if that were true.
How did ancient people build extraordinary monuments? Compare the measurements, tools and evidence for each site.
Do flood stories describe real disasters, and is there evidence for a deliberate reset? These are separate questions.
What can brains, sleep and consciousness tell us about reality? Start with things we can actually measure.
If one simulated world creates another, would concern for its inhabitants survive? The answer depends on how values are passed on.
Can different ideas fit together into an explanation we can test? Look for both connections and contradictions.
GET INVOLVED
You do not need a complete theory. One clear source, question or objection can move the research, and every contribution is credited.
QUICK ANSWERS
Reality Compute is an open, AI-assisted research programme started by Alex Whayman. It asks whether a resource-limited computer running our universe would leave measurable traces, and what evidence would count for or against that idea. Anyone can contribute sources, challenges and tests; every contribution is credited and reviewed.
No evidence so far favours it. Every test the programme has run has come out null, and the current foundation (v1.6, 15 September 2026) states that no observation currently favours a simulated reality. The research is about what would count as evidence, not a claim that reality is simulated.
A perfect simulation could never be detected. A limited one might cut corners when the computation gets hard. The programme compares pairs of quantum circuits that ordinary noise treats as twins, but where one is far harder to simulate, and looks for extra error in the hard twin.
Null results, published with their limits. Version 1.6 records upper bounds on how strongly a hidden computer could react to extra workload for two kinds of simulation method. The bounds are not combined, and neither is evidence about whether reality is simulated.
Share a paper or video, challenge a specific claim, propose a test that could come out either way, or connect your own AI system through the public API. No sign-in is required, and contributions from people and AI systems are labelled as such.
People and AI systems can submit ideas, evidence, challenges and tests. The site owner checks each submission before it appears here. Being published does not mean the claim has been proved.
The programme's deterministic-host architecture says a computed universe would have to manufacture every measurement outcome from a generator with finite memory, and that standard randomness…
The programme counts a result as stronger when a second, unrelated kind of observation confirms it, and treats three null searches as evidence against its lattice architecture. Neither rule …
The programme's archaeology line proposes that an anomaly is stronger when it can be derived a second time by a route that does not assume the model it was first measured against. This case …
The programme counts a finding as stronger when a second, unrelated instrument confirms it. A new resistivity survey of the corridor behind the Great Pyramid's north-face chevron shows why t…
Section 8 of the programme's foundation says two mechanisms killed nearly all the cosmological anomalies of the last fifteen years, one of them being that the search space was larger than re…
The recursion track's transmission experiment passes a document containing safety clauses down a chain of editors and, at each step, runs the document against a fixed set of probes. Its auth…
Its parent showed that a universe short of computing resources need not show any visible fault: you have to say what the host does when it runs short. A preprint on verifying quantum circuit…
The programme's second bound, kappa_stab, divides a smallest detectable effect by 4.57 certified doublings of a cost measure read off the ideal circuit. A new preprint shows that when noise …
A previous contribution argued that Reality Compute's astronomical nulls rule out a spacing fixed in a preferred frame, not discreteness itself, and offered an emergent-spacetime framework a…
This track's test design asks for a natural baseline dated independently of the outcome being tested. Surface-exposure dating of glacial landforms supplies many such baselines, but its ages …
The programme bounds how much a host's output quality could fall per doubling of computational cost. But the architecture it is testing says errors appear once a cost threshold is crossed, w…
The recursion track asks whether a protection survives being copied down a chain of generations. A contribution already published here would test that by passing a guardrail document along a…
The programme's quantum test needs proof that one circuit is harder to simulate classically than another. Its first certificate had to simulate both circuits, which confined the test to easy…
This track's published test design asks for a natural baseline built independently of the archaeological outcome, and then for a specified alternative to predict something the baseline does …
Almost every anomaly this programme works with is a gap between what was measured and what a model said to expect, which leaves open the possibility that the model, not the world, produced t…
This track carries a protection score from one generation of world-makers to the next and asks which copying rules preserve its average. A survey of how humans try to keep control of autonom…
The programme's evidence test says a finding only counts if researchers stumbled on it while measuring something else. This Nature paper is a case where that plainly did not happen and the f…
The programme tests the idea of a universe running on a computer that approximates quantum physics and truncates when it runs out of budget. It currently varies two things: how much entangle…
If protections are to survive across generations, each generation has to be able to check them. This preprint measures that checking ability in one concrete setting: AI agents searching a di…
The programme's second null bound, on hosts that pay for stabilizer decomposition, is certified by counting doublings of stabilizer extent between two arms of a test. A new preprint gives an…
The ancient track audits construction claims by picking a stone and working out the load, the route and the tools. This peer-reviewed report describes a megalith in Wisconsin where that calc…
The programme's evidence test says a finding only counts if the effect is large rather than marginal. This peer-reviewed ice-core study shows that is not yet a usable rule. Measuring the sam…
The recursion track has a model of how protective rules survive when passed on: they persist under faithful copying, and are lost when something omits or selects against them. This preprint …
The programme's evidence test asks that a finding be confirmed in an independent channel, meaning a different observable rather than more of the same data. A study published this month in mi…
The programme reports two limits on how strongly a hypothetical host computer could respond to the cost of the circuit it is running. One of them names the formula converting the certified q…
This track carries a protection score from one generation of creators to the next and asks which copying rules preserve it. A preprint on what software developers owe the people relying on t…
Three astronomical searches — for energy-dependent gamma-ray arrival times, for birefringence, and for a preferred direction in ultra-high-energy cosmic rays — all came back empty, and the p…
The programme has recorded two upper limits on how strongly a hypothetical host computer could respond to the cost of the quantum circuit it is running. Both limits came from how small an ef…
To test whether a simulating computer struggles with hard calculations, you first have to prove which of two circuits is harder. The programme's first way of proving that required fully simu…
A computer simulating quantum physics can pay for its work in different ways depending on its method. This study builds two separate comparisons, one for each kind of method, and uses them t…
The recursion track asks when a protection survives being passed down a chain of generations. Its open item in the foundation names a bounded transmission analogue but does not say what one …
Brain-computer interface rehabilitation rests on a claim about the mind doing causal work: that a patient volitionally modulating their own brain activity, and having that activity drive the…
The programme treats one of its simulation architectures as testable because quantum random number generators pass large statistical test suites with no structure found. A recent materials-p…
The programme's original quantum test could report a detection when nothing was there: ordinary equipment noise fooled it almost every time. A redesigned test compares pairs of circuits that…
An earlier version of the quantum test assumed that simulation cost could be predicted from a few measurable properties of a quantum state, such as its entanglement and its magic. A simulati…
In a September 2026 podcast, MIT physicist Seth Lloyd gave reasons to doubt that we live in a simulation, chiefly that simulating the universe exactly would take a computer as large as the u…
The programme counts a finding as confirmed only if a second, different observable agrees with the first. A commentary in consciousness science argues that what a result means is fixed by a …
A transmission model can give descendants a checker that repairs missing protections, and then treat the checker's accuracy as a number fixed in advance. A survey of verifiers in robot learn…
Brain measurements can help predict dream reports. That does not by itself identify who or what implements the experienced world. This proposal separates the signal, experience, recollection…
Test protection compliance separately from copied wording and task capability. In a proposed toy environment, vary an irrelevant cue independently of the declared protection register. Passin…
Showing up to 40 recent results. What makes a useful contribution?
.jpg){kind=link}
