Kevin L. Michel · A guide to the working paper

What would count as evidence for Many Worlds?

Quantum experiments can strengthen the physics behind Many Worlds without showing that it is the only explanation. The question that makes the difference: compared with what?

Working paper · Version 1.0 · 24 August 2026 · Abstract & citation

Shared successOne result can support several theories.

A real limitIdentical predictions cannot pick a winner.

An open questionWhat would distinguish the worlds themselves?

Start here

What is Many Worlds actually saying?

A quantum system can be in a superposition: a combination of alternatives that can interfere with one another. This is more than a hidden coin that has already landed heads or tails.

When the system interacts with a detector, their states become linked. Information then spreads into the surroundings. This process, called decoherence, makes interference between the alternatives hard to detect locally.

In Everett’s account, the quantum state keeps evolving. No special collapse deletes the other outcomes. Stable patterns in that state form what are often called branches. Many Worlds treats all qualifying branches as real.

That last claim goes beyond saying that the quantum calculations work. This paper asks how evidence reaches each part of the account.

A schematic of the claim

One quantum state, two correlated record patternsA combined A and B quantum state evolves into a superposition of a detector recording A with environmental records A, and a detector recording B with environmental records B. Neither is shown as removed. This illustrates the Everettian account, not experimental proof of worlds. ψOne state Detector records ADetector records B Environment: AEnvironment: B Both terms remain
This is a simplified diagram of the Everettian account, not a picture of observed universes. Branches are approximate patterns, not a known exact count. Decoherence alone does not select a single outcome.

The paper’s central distinction · §2

Five claims that headlines often mix together.

These are related questions, not steps that automatically lead to “Many Worlds proved.” Each asks something different of the evidence.

01

Coherence

Quantum alternatives can interfere

Under the tested conditions, alternatives still act together like waves. This goes beyond ordinary uncertainty about which path a particle took.

Where the evidence stops

A test covers particular masses, distances, times, and conditions. It does not prove that quantum rules work everywhere.

02

Universal unitarity

The quantum law never breaks

The same smooth quantum evolution applies to particles, detectors, people, and the universe. No extra physical collapse interrupts it.

Where the evidence stops

Experiments can extend the range we have tested. A finite set of tests cannot establish a law with no exceptions.

03

Model-specific exclusion

A proposed collapse is absent

Some theories predict that superpositions physically collapse. A test can look for a signal that a named model should produce.

Where the evidence stops

Ruling out one model or range of settings does not rule out every possible collapse theory.

04

Decoherence and records

Stable records can emerge

A system leaves information in its surroundings. Many parts of that environment can carry matching records, helping explain a shared, stable-looking world.

Where the evidence stops

Showing how records form does not, on its own, show that every branch is physically real.

05

Everettian ontology

All qualifying branches are real

Many Worlds treats the different, stable branches of the quantum state as physically real. Ontology means a theory’s account of what exists.

Where the evidence stops

This stronger claim needs a reason to choose it over rival accounts that make the same observable predictions.

Evidence lab

Change the comparison. Change the conclusion.

A result counts in favour of one theory over another when that result was more likely under the first theory. Merely fitting the result is not enough.

The same observation

An interference pattern survives.

Assume the apparatus is understood and ordinary noise has been checked. What does this result support?

It can favour the shared quantum dynamics.

If the collapse model makes this pattern less likely at its specified parameters, observing the pattern counts against that model. Everett’s odds against that collapse model improve, as do the odds of rivals that predict the same quantum pattern.

Conceptual comparisons. The actual strength of evidence requires each model’s quantitative predictions and a treatment of experimental uncertainty.

01 · A family can gain together

More support for Everett. The same odds against Bohm.

Suppose we begin with 40% confidence in a family of quantum theories. Within it, Everett and equilibrium Bohmian mechanics each start at 15% overall. Other members start at 10%.

Now let evidence favour the whole family. Every member grows by the same factor when they make identical predictions for that evidence.

No preference: 1×Strong preference: 100×
EverettEquilibrium BohmOther in familyOutside family
Before the evidence
After the evidence
Illustrative confidence before and after this evidence
Theory groupBeforeAfter
Everett15%32.1%
Equilibrium Bohm15%32.1%
Other family members10%21.4%
Outside the family60%14.3%

40% → 85.7%Confidence in the shared family

1 : 1 → 1 : 1Everett’s odds against equilibrium Bohm

Teaching example, not a measurement. These are the paper’s illustrative starting probabilities. They are not scientists’ estimates of the chance that Many Worlds is true. A Bayes factor multiplies prior odds; it is not a probability of truth.

02 · Shared evidence has a limit

How much does this result single out Everett?

“Not Everett” includes more than collapse theories. It can include rival accounts that use the same quantum dynamics. The more weight those rivals have, the less the result distinguishes Everett from its full set of alternatives.

100×For the shared dynamics

1.98×For Everett against all non-Everett alternatives

With half of the rival probability already on the same dynamics, a 100× result multiplies Everett’s prior odds by about 1.98, not 100.

The exact calculation and assumptions
BFM,¬M = rqr + (1 − q)

Here M means Everett, U means the shared unitary family, r = P(e | U) / P(e | ¬U), and q = P(U | ¬M) before the evidence. The example assumes that this evidence has the same likelihood under all members of U. For q greater than zero, the inherited factor approaches a ceiling of 1/q as r grows. At q = 1, it remains exactly 1.

Illustrative assumptions only. Changing q changes the prior mix of alternatives; it does not change the experimental result. Source: the paper’s Proposition 4, §3.6.

What the experiments reach · §5

Real progress. Specific conclusions.

The paper reviews the following cases. Each moves a particular question forward. None supplies an outcome distribution that favours Everett over every operationally equivalent rival.

Interference2026

Large objects behaving like waves

Pedalino and colleagues reported interference in sodium clusters centered near 172 kilodaltons, containing more than 7,000 atoms. The inferred separation of the alternatives was 133 nanometres.

What it supports

Quantum coherence survives in a demanding tested regime.

What it does not establish

It does not choose Everett over another complete theory with the same interference predictions. Clusters near one megadalton passed through the apparatus, but clear quantum/classical discrimination was below about 200 kilodaltons.

Pedalino et al. · NaturePaper §5.1

Collapse constraints2026

Listening for collapse

The XENONnT study searched for radiation that specified collapse models predict. It found no significant excess in a 1–140 keV electron-recoil search with 1.16 tonne-years of exposure.

What it supports

Tighter limits on the tested collapse models and their parameters.

What it does not establish

A missing predicted signal can count against a model. It does not exclude all collapse theories, and this experiment did not create a giant superposition.

XENON Collaboration · Physical Review LettersPaper §5.2

Quantum Darwinism2025

How the environment keeps a record

Engineered superconducting circuits showed how information about a system can spread into environmental qubits. Separate pieces can carry matching information about a preferred observable.

What it supports

A mechanism for stable, redundant records in controlled systems.

What it does not establish

The qubits are not conscious observers. Redundant records neither pick a unique outcome nor prove that every record-bearing branch is a world.

Zhu et al. · Science AdvancesPaper §5.3

Local Friendliness2020–2025

Can everyone’s observations be absolute?

Local Friendliness results expose a conflict among assumptions about observed events, locality, freedom of experimental settings, and scalable quantum control. Later circuits explored larger versions of these tests.

What it supports

A challenge to keeping all the tested assumptions together.

What it does not establish

Rejecting a combination of assumptions does not tell us which one must go. Laboratory ‘friends’ are physical registers, not people. A circuit’s branch factor is not a count of worlds.

Bong et al. · Nature PhysicsPaper §5.4

Proposal and circuit preprints2026

A message between worlds?

Violaris proposed an ‘interbranch communication’ thought experiment. Altman implemented an inspired five-qubit circuit using 20,000 shots, testing coherent information transfer.

What it supports

A proposal and a hardware demonstration about controlled quantum information.

What it does not establish

The hardware result is inter-register transfer. It does not establish messages between autonomous worlds or discriminate between quantum interpretations. Both works are identified as preprints in the paper.

Altman · arXiv preprintPaper §5.5

The research agenda · §6

So what would count?

The answer depends on how strong a conclusion we want. The paper identifies several useful routes, with different burdens of proof.

A

A prediction that separates complete theories

Specify an Everettian package and a rival package. Give each enough detail to predict outcomes. Find an accessible experiment where those predictions differ, then test it.

The demand: a real likelihood difference, not a new name for the same quantum circuit. This route cannot distinguish theories that remain exactly equivalent in the tested domain.

B

More demanding tests of the shared physics

Extend interference and reversibility tests. Search carefully for the signals that particular collapse theories predict. Control ordinary noise and report which parameter ranges were tested.

The gain: stronger support for a family that includes Everett. The gain need not favour Everett over other members of that family.

C

A stronger account of branches and probability

Show why stable records form, which patterns qualify as branches, and how the theory justifies the probabilities used to connect itself to observations.

The demand: explicit assumptions and comparisons with alternatives. A persuasive explanation can matter, but it is not automatically a new laboratory signature.

For any proposed breakthrough, ask:

“Which rival would have expected a different result?”

If no complete rival predicts differently, the experiment may test shared physics. Its result alone does not choose a unique interpretation.

Keep the complexity where it belongs · §3

The formal argument, with the terms unpacked.

The paper organizes familiar probability tools around a difficult quantum question. It does not claim to have invented Bayes’ theorem. Its contribution is the structure of the comparison and the analysis of how support is inherited.

01 Identical predictions give a Bayes factor of one

A complete theory needs more than a story about reality. It needs dynamics, an account of what exists, state and parameter assumptions, and a probability or typicality rule that produces observable predictions.

BF12(D) = P(D | H1)P(D | H2) = 1

If H₁ and H₂ predict the same conditional probabilities for every allowed step, they assign the same probability to the whole record D. This remains true when later experimental choices depend on earlier results.

Scope: the result applies within the domain of equivalence, for a possible record with well-defined likelihoods. It is not a proof that all future theories and experiments must remain equivalent. See Proposition 1.

02 A family can gain without its members changing rank
P(Hi | e)P(Hi) = P(F | e)P(F)

When the evidence treats all family members alike, each receives the same proportional increase in probability. Their odds against one another stay fixed. To know whether the family gains overall, compare it with its full complement, weighted by the priors on its alternatives. See Proposition 2.

03 Shared support and distinctive support are separate factors
P(M | e)P(M) = P(F | e)P(F) × P(M | F, e)P(M | F)

The first factor measures how much the shared family gains. The second measures whether the evidence favours this particular member within the family. That second factor can remain 1 while the first grows.

The paper applies the same logic to bridges between unitary dynamics, robust records, and Everettian ontology. Evidence does not automatically strengthen every bridge. See Proposition 3.

04 The Born-rule question cannot be skipped

In ordinary quantum calculations, outcome weights follow the Born rule: the weight of an outcome is the squared magnitude of its amplitude.

pi = |αi|2

If every qualifying outcome occurs in a branch, why should an observer use these weights when making predictions? Everettian proposals appeal to ideas such as decision theory, typicality, and uncertainty about one’s own location in the quantum state.

“Every outcome happens somewhere” does not answer that question. A complete package must explain its weighting rule. Counting branches is not a shortcut: branches are approximate and their number is not supplied as a simple fixed count. The paper maps this issue; it does not settle the debate. See §2.3.

Use the distinction

A better way to read a quantum headline.

First, name the observation. Did researchers see interference, rule out a predicted signal, or create redundant records? Then name the actual theory comparison.

Next, ask what was assumed. A rival with no finished prediction does not automatically lose. Its likelihood is undefined, not necessarily zero.

Finally, match the claim to the evidence. A result may be compatible with Everett, favour shared dynamics, support a mechanism for records, or discriminate among theories. Those are different achievements.

Does this paper argue that Many Worlds is false?

No. It allows indirect support for Everett through successful shared physics. It also identifies the limits of that support against equally predictive rivals.

Does it say interpretations are beyond science?

No. It separates experimental discrimination from other scientific arguments. Explanatory reach, coherence, simplicity, and the cost of extra assumptions can matter. They should not be presented as a measured difference in experimental outcomes.

Does decoherence prove there is only one outcome?

No. Decoherence helps explain why interference becomes inaccessible locally and why stable records emerge. On its own, it neither selects one outcome nor proves that all branches are real.

Can an experiment ever change the comparison?

Yes, if the complete theories differ in their predictions in an accessible domain. The equivalence result applies where their predictions are the same. It does not forbid new physics or a new discriminating test.

Is this a peer-reviewed article?

The supplied canonical document is a working paper, Version 1.0. This page explains its argument. The experimental sources include journal articles and clearly identified preprints.

The complete argument

Read the canonical paper.

What Would Count as Evidence for Many Worlds? Operational Equivalence and the Evidential Architecture of Everettian Quantum Mechanics

Kevin L. Michel · 24 August 2026 · Version 1.0 · 38 pages

The download is the author-supplied canonical PDF, unchanged. No account or payment is required. The explainer’s sliders are teaching models, not new experimental findings.

Selected sources and further reading

Experimental sources are linked beside each case above. The full literature review, proofs, qualifications, and reference list are in the paper.