Extended Wigner’s Friend Paradoxes Overview

• Extended Wigner’s friend paradoxes are multi-agent quantum scenarios where internal observers record outcomes while external agents perform interference measurements.
• They reveal conflicts in cross-agent reasoning, challenging standard collapse rules, state assignments, and the absoluteness of observed events in quantum mechanics.
• Circuit-level and multipartite implementations provide practical insights into coherence, contextuality, and error management in scalable quantum experiments.

Searching arXiv for recent and foundational papers on extended Wigner’s friend paradoxes. Searching for “extended Wigner's friend paradox”. Extended Wigner’s friend paradoxes are multi-agent extensions of Wigner’s original thought experiment in which one or more “friends” perform measurements inside sealed laboratories while external “Wigner-type” observers describe those laboratories as coherent quantum systems. In this setting, the central issue is not merely the coexistence of collapse and unitary evolution, but the status of cross-agent reasoning, the conditions under which quantum states may be assigned, and whether “facts” or “observed events” are absolute, relative, or context-dependent. The literature does not treat the paradoxes as a single theorem with a single lesson: some analyses argue that standard quantum mechanics remains internally consistent once one conditions correctly on surviving records and on Heisenberg-cut choices, whereas others recast the paradoxes as no-go results against Absoluteness of Observed Events, Local Agency, setting-independence, or related assumptions, and still others identify the underlying resource as logical contextuality rather than specifically Bell nonlocality (Sokolovski et al., 2021, Vilasini et al., 2022, Nurgalieva et al., 6 Feb 2025).

1. Canonical extended setup and Frauchiger–Renner-type protocol

A standard extended Wigner’s-friend arrangement uses two perfectly isolated laboratories, $\bar L$ and $L$. Inside $\bar L$, Friend $\bar F$ measures a two-level “coin” system $\mathrm C$; inside $L$, Friend $F$ measures a two-level “spin” system $\mathrm S$. Outside $\bar L$ sits observer $\bar W$, and outside $L$0 sits observer $L$1. In the presentation analyzed by Sokolovski and Matzkin, the experiment unfolds at times $L$2: the coin is prepared in $L$3, $L$4 measures it, a conditioned coin–spin coupling sends $L$5 to itself and $L$6 to $L$7, $L$8 measures the spin, and finally $L$9 and $\bar L$0 each perform an interference-type measurement on the entire laboratory in a $\bar L$1 basis (Sokolovski et al., 2021).

Agent	Location/system	Role
$\bar L$2	Inside $\bar L$3	Measures the coin
$\bar L$4	Inside $\bar L$5	Measures the spin
$\bar L$6	Outside $\bar L$7	Measures the whole lab $\bar L$8
$\bar L$9	Outside $\bar F$0	Measures the whole lab $\bar F$1

The Frauchiger–Renner protocol, as summarized in later critical discussions, has the same basic architecture: a quantum coin is measured by $\bar F$2, the outcome determines whether an electron is prepared in $\bar F$3 or $\bar F$4, $\bar F$5 measures the electron in the $\bar F$6 basis, and the external observers measure the sealed laboratories in rotated $\bar F$7-type bases. Under the usual quantum-mechanical rules plus the meta-assumptions (C) Consistency, (S) single outcomes, (Q) universal validity of the Born rule, and an additional assumption (H) about collapse for insiders but unitary evolution for outsiders, the protocol yields a contradiction between a nonzero probability for $\bar F$8 and a chain of certainty-claims implying that $\bar F$9 cannot occur (Muciño et al., 2020).

2. How the paradoxical inference chains are generated

The extended paradoxes do not arise from a single probability calculation; they arise from chaining together inferences that are individually valid only under specific conditions. In the Feynman-rules analysis, four distinct experimental situations are separated by which records survive until the end. If both laboratories are measured in the $\mathrm C$0 basis, the friends’ records are erased and the relevant alternatives interfere. In that case one obtains

$\mathrm C$1

If instead only one external observer performs an interference measurement while the other leaves the friend’s record intact, one obtains context-specific implications such as

$\mathrm C$2

The apparent contradiction comes from chaining these implications across different final measurement arrangements, even though “none of these four situations can be mixed: each corresponds to a different experimental question (different choice of final measurement basis) and hence to a different statistical ensemble” (Sokolovski et al., 2021).

A second line of diagnosis isolates the assumption responsible for the certainty-chain itself. Muciño and Okon argue that the contradiction depends crucially on the hidden premise (H): when an observer inside a sealed lab performs a measurement, she treats it as causing wave-function collapse, while an outside observer treats the entire lab as evolving unitarily. On this analysis, the certainty-statements

$\mathrm C$3

$\mathrm C$4

and the subsequent transitive reasoning are not consequences of quantum mechanics alone, but of an agent-relative collapse/no-collapse split. Without (H), “either all agents use collapse (no paradox) or all use unitary evolution (no definite single outcomes)” (Muciño et al., 2020).

A recurring misconception is therefore that extended Wigner’s-friend paradoxes, by themselves, show that standard quantum mechanics is formally inconsistent. The contrary diagnosis, strongly represented in the literature, is that the contradiction appears only when statements derived under mutually incompatible ensembles, or under unacknowledged extra assumptions, are combined.

3. Conditioning, Heisenberg cuts, records, and state assignment

Several contemporary frameworks resolve the paradox by making the conditioning structure explicit. Vilasini and Woods formalize an Extended Wigner’s Friend Scenario as an augmented quantum circuit in which each measurement channel has a binary setting $\mathrm C$5 encoding a Heisenberg cut: $\mathrm C$6 means a purely unitary channel, while $\mathrm C$7 means a decoherent measurement producing a classical record. A setting-conditioned prediction is then computed by fixing the $\mathrm C$8’s and applying the Born rule within one circuit. Their main result is a completeness–consistency–causality theorem: conventional predictions are special cases of setting-conditioned predictions, no two statements assign different probabilities to the same event under the same conditions, and cut-choices outside the causal past of the conditioned outcomes do not matter. In this framework, Frauchiger–Renner-type contradictions disappear because the linked implications actually use different hidden conditions $\mathrm C$9, and “classical logic cannot chain statements when their hidden conditions $L$0 differ” (Vilasini et al., 2022).

A related reconstruction is the “bubble” picture. Del Santo and collaborators distinguish a friend’s bubble, in which the friend has a definite outcome $L$1 and applies the usual state-update rule, from Wigner’s bubble, in which Wigner assigns an entangled state to the friend-plus-system composite. The two descriptions coexist because each observer has different information “in principle available” to them. Their central criterion states that an observer in bubble $L$2 may adopt and verify the state assignment of bubble $L$3 for measurements in $L$4 if and only if she conditions on the full classical information in her bubble together with all quantum-mechanically reversible records that can be coherently transferred from $L$5 to $L$6 without disturbing the relevant degrees of freedom. In their explicit example, after an additional unitary interface $L$7, Wigner predicts $L$8, while the friend predicts $L$9; but if the friend coherently receives Wigner’s description via a swap, she can then predict $F$0 exactly as Wigner does (Santo et al., 2024).

Other approaches formulate admissibility conditions on cross-agent knowledge. Baltag and Smets define “records,” “persistent records,” and “admissible observers” relative to a protocol. A subsystem may function as an observer for another only if its information can persist in the relevant epistemic sense; if a later global measurement is destructive and no external persistent record exists, higher-order knowledge attributions to that subsystem are illegitimate. On this view, in the standard Frauchiger–Renner protocol neither $F$1 nor $F$2 is an admissible observer for the external Wigners, so nested knowledge transfer involving them is blocked (Baltag et al., 2023).

The same basic moral appears in analyses of information leakage and in the timeless Page–Wootters formulation. Baumann shows that a tunable quasi-classical communication channel between friend and Wigner interpolates smoothly between a fully unitary regime and an effectively collapsed regime; full which-outcome information destroys interference, while unbiased messaging preserves it. Losada, Laura, and Lombardi argue that generalized two-time probability rules are required in a timeless description, and that a genuine joint two-time probability for Wigner’s and the friend’s outcomes is well-defined only when Wigner’s later measurement does not disturb the friend’s memory; the same condition matches the consistency condition of the consistent-histories framework (Baumann, 2023, Baumann et al., 2019).

4. No-go theorems, local friendliness, and contextuality

A major branch of the literature treats extended Wigner’s-friend paradoxes as theory-independent or near-theory-independent no-go results. In the Local Friendliness program, the central assumptions are Absoluteness of Observed Events (AOE) and Local Agency (LA). Bong-style scenarios then derive inequalities analogous to Bell inequalities, but with observer records inside sealed laboratories. A possibilistic strengthening replaces probabilistic Local Agency by Possibilistic Local Agency (PLA), requiring only that interventions cannot influence possibilities outside their future light cone. Under AOE and PLA, a Hardy-type pattern of possible/impossible events leads to contradiction; standard quantum predictions satisfy the required pattern and therefore violate possibilistic Local Friendliness (Haddara et al., 2022).

Subsequent work shows that nonlocal correlations are not the unique resource behind such paradoxes. Brown, Cavalcanti, and Friend construct a no-go theorem on a single qutrit using the 5-cycle contextuality scenario. Their setup employs five binary observables $F$3, unitary implementation of each measurement, successive undoing operations by Wigner, and a metaphysical assumption called Commutation Irrelevance: if a unitary $F$4 commutes with a later measurement $F$5, then the joint outcome statistics of $F$6 and $F$7 are independent of whether $F$8 is performed before or after $F$9. The contradiction is then derived from contextual, but local, correlations. The explicit conclusion is that “Extended Wigner’s friend paradoxes do not require nonlocal correlations” (Walleghem et al., 2023).

Nurgalieva and Vilasini generalize this point further. They define a multi-agent setup in a theory $\mathrm S$0, together with four assumptions A1–A4: common knowledge of theory and setup, reasoning only about compatible agents, setting-independence with respect to Heisenberg cuts, and non-contradictory outcomes. Their main theorem states: if a theory admits a Wigner’s-Friend-type multi-agent paradox under A1–A4, then there exists a measurement scenario in that theory whose empirical model is logically contextual. In particular, the Frauchiger–Renner paradox becomes a proof of logical contextuality. They also show that theories admitting extremal vertices of $\mathrm S$1-cycle contextuality scenarios admit Wigner’s-Friend-type paradoxes without post-selection, whereas any quantum Wigner’s-Friend paradox based on the $\mathrm S$2-cycle scenario must involve post-selection (Nurgalieva et al., 6 Feb 2025).

The scope of “absoluteness” has also been widened from outcomes to choices. An extended Wigner’s-friend argument based on the Pusey–Barrett–Rudolph theorem internalizes the friends’ allegedly free choices as recorded quantum events. Under Absoluteness of Observed Events and Local Agency, together with the PBR orthogonality structure, the resulting argument yields the conclusion that “free choices cannot be both absolute and local” (Walleghem, 14 May 2026).

5. Multipartite extensions and circuit-level implementations

Extended Wigner’s-friend paradoxes are not confined to the minimal two-laboratory case. A scalable tripartite version introduces three laboratories, each containing a friend and an outside observer, with the outside observers choosing between measuring the friend’s recorded fact in the computational basis or the joint signal-plus-record system in a Bell basis. The signal state is

$\mathrm S$3

augmented by ancillary Bell pairs inside each laboratory. Local “fusion” transformations are implemented by a CNOT followed by post-selection, and the resulting correlators $\mathrm S$4 violate a generalized tripartite inequality with quantum maximum $\mathrm S$5 at $\mathrm S$6 (Ding et al., 2021).

This construction was studied both in Q# simulation and on IBM quantum hardware. The Q# implementation used eight operations, one per measurement setting, each run for $\mathrm S$7 shots and repeated $\mathrm S$8, yielding

$\mathrm S$9

in excellent agreement with the ideal $\bar L$0. The IBM implementation used the 5-qubit devices ibmq_lima and ibmq_belem; at the time of experiment the average single-qubit error was $\bar L$1, the average CNOT error was $\bar L$2, and typical $\bar L$3. W-state tomography agreed with theory to within $\bar L$4, and post-selected six-qubit correlations showed deviations up to $\bar L$5, attributed mainly to two-qubit gate errors and readout infidelity (Ding et al., 2021).

This line of work suggests that the unitary-versus-collapse tension can be embedded in scalable circuit architectures and studied at the level of correlation polynomials, post-selection overhead, and hardware error budgets, rather than only as an abstract philosophical puzzle.

6. Broader significance, analogies, and alternative resolutions

Some authors argue that the significance of extended Wigner’s-friend paradoxes extends beyond quantum foundations. Jones and Müller formulate “Restriction A”: a theory may fail to provide a single joint probability distribution over all agents’ observations, even when it does provide a probabilistic description of external physical variables. They argue that versions of the Wigner’s-friend structure can be reproduced by classical thought experiments involving duplication or merging of agents, and that the same structural difficulty appears in the Boltzmann-brain problem. On this view, the essential issue is not uniquely quantum, but concerns reliable prediction in situations where outcomes can be privately but not intersubjectively verified (Jones et al., 2024).

A second strand connects extended Wigner’s-friend paradoxes with black-hole paradoxes. Hausmann and Renner argue that the firewall paradox depends on a combination rule or “consistency of knowledge” rule analogous to the rule used in extended Wigner’s-friend reasoning, and that the usual conclusion of the firewall paradox is therefore challenged by the fact that such a rule already conflicts with quantum theory without gravity. Later work constructs paradoxes that merge nonlocal Wigner’s-friend protocols with black-hole evaporation and interprets complementarity as a possible response; still later analysis argues that taking the analogy seriously appears to favor intrinsic relationality and some kind of retrocausality (Hausmann et al., 4 Apr 2025, Walleghem, 7 Jul 2025, Adlam, 20 Apr 2026).

The interpretational range is correspondingly wide. Some papers maintain that standard quantum mechanics is internally consistent once amplitudes and probabilities are combined correctly and hidden conditioning choices are made explicit. Others maintain that the paradoxes reveal the failure of Absoluteness of Observed Events, of Local Agency, or of setting-independence. Still others propose alternative ontologies or dynamics. One recent example is a framework of quantum mechanics on the hybrid space $\bar L$6, where collapse is not an independent postulate but a dynamical consequence of a non-local Schrödinger equation: wavefunctions localize onto compact supports during measurement interactions, both Friend and Wigner are modeled as classical apparatuses, and the observer-dependent assumptions driving extended Wigner’s-friend no-go theorems are thereby vacated (Zúñiga-Galindo, 30 Jun 2026).

Taken together, these developments show that “extended Wigner’s friend paradoxes” name a research program rather than a single paradox. The program encompasses formal inconsistency claims, ensemble-sensitive resolutions, circuit frameworks with explicit Heisenberg cuts, contextuality theorems, multipartite simulations, and analogies to problems in cosmology and black-hole physics. What unifies these disparate lines is the attempt to specify, with maximal precision, which outcomes are recorded, which records persist, which descriptions are conditioned on which information, and under what circumstances one observer may inherit or combine another observer’s quantum-mechanical inferences.