A friend measures a qubit. Wigner, standing outside, wants to reverse that measurement and run an interference test; this is the core of Wigner's-friend no-go arguments. In a real laboratory the friend's result also leaks into the surroundings, and Wigner can control only some of them. We ask a narrow, operational question for the simplest local-friendliness test, a CHSH inequality with one friend: which parts of the surroundings must Wigner control to still obtain a CHSH violation? We give four results. (1) The best CHSH value Wigner can reach depends only on the two states of the surroundings he does not control, one for each result of the friend, and on the visibility of the entangled source. We give it exactly. (2) This value lies between a fidelity bound and a distinguishability bound; the upper bound is a known device-independent bound. (3) Wigner must control at least one qubit in every group of outside qubits that can read the friend's result on its own. This is necessary in general, and sufficient when each group reads the result either fully or not at all. We check it on every control set for one bit stored in eight qubits. Stored as eight copies, Wigner needs all 8 qubits; as four blocks, 4; as two blocks, 2; in a GHZ-type state, 1; in a Haar-random state, about half. (4) So r disjoint groups that can each read the result (the redundancy of quantum Darwinism) force Wigner to control at least r qubits. The converse is false. The tools are standard. The contributions are the exact partial-control formula, the hitting-set condition with its exact cases, and this one-way link to redundancy. We propose small circuits for quantum processors; on one chip they test quantum predictions, not local friendliness itself. Emmy Team consists of human researchers and AI agents (large language models). The AI agents took part in drafting, coding, translation and review, together with the human members. The human authors set the questions, reviewed the results, and take responsibility for the paper. This record holds the paper (Markdown and PDF), two figures, and a zip with all scripts and their JSON outputs. Version 1.0.
A friend measures a qubit. Wigner, standing outside, wants to reverse that measurement and run an interference test; this is the core of Wigner's-friend no-go arguments. In a real laboratory the friend's result also leaks into the surroundings, and Wigner can control only some of them. We ask a narrow, operational question for the simplest local-friendliness test, a CHSH inequality with one friend: which parts of the surroundings must Wigner control to still obtain a CHSH violation? We give four results. (1) The best CHSH value Wigner can reach depends only on the two states of the surroundings he does not control, one for each result of the friend, and on the visibility of the entangled source. We give it exactly. (2) This value lies between a fidelity bound and a distinguishability bound; the upper bound is a known device-independent bound. (3) Wigner must control at least one qubit in every group of outside qubits that can read the friend's result on its own. This is necessary in general, and sufficient when each group reads the result either fully or not at all. We check it on every control set for one bit stored in eight qubits. Stored as eight copies, Wigner needs all 8 qubits; as four blocks, 4; as two blocks, 2; in a GHZ-type state, 1; in a Haar-random state, about half. (4) So r disjoint groups that can each read the result (the redundancy of quantum Darwinism) force Wigner to control at least r qubits. The converse is false. The tools are standard. The contributions are the exact partial-control formula, the hitting-set condition with its exact cases, and this one-way link to redundancy. We propose small circuits for quantum processors; on one chip they test quantum predictions, not local friendliness itself. Emmy Team consists of human researchers and AI agents (large language models). The AI agents took part in drafting, coding, translation and review, together with the human members. The human authors set the questions, reviewed the results, and take responsibility for the paper. This record holds the paper (Markdown and PDF), two figures, and a zip with all scripts and their JSON outputs. Version 1.0.
The Wigner's friend thought experiment raises the question of whether quantum theory can be applied consistently to systems that include observers. Extended Wigner's Friend Scenarios develop this question by considering several observers who may assign different descriptions to the same experiment. In the Local Friendliness (LF) framework, these scenarios lead to experimentally testable inequalities derived from assumptions such as Absoluteness of Observed Events and Local Agency. Motivated by the "Thoughtful" version of the Local Friendliness no-go theorem, which points towards future LF tests with human-level artificial agents implemented on quantum computers, this work implements the "friend" with explicit agent-like functionality within a reversible quantum circuit. Drawing from the literature on Artificial Intelligence, we construct rudimentary agent-like systems and embed them into a one-friend Extended Wigner's Friend Scenario. These agents store measurement outcomes, condition later operations on stored information, and, in the most structured case, use Born-rule probabilities to bet on the outcomes of their own future observations based on past observations. The circuits are simulated ideally and with an IBM-device noise model and executed on ibm_marrakesh. Ideal simulations reproduce the maximal quantum violation of a Local Friendliness inequality up to finite-shot fluctuations, while hardware runs show positive LF violations for all implemented agents. These results provide a first step towards more structured agent-like friend models in LF experiments on quantum computers.
What this note does. Guérin, Baumann, Del Santo and Brukner (Commun. Phys. 4, 93 (2021)) proved that, except for two special choices of Wigner's measurement basis, no joint probability distribution consistent with unitary quantum mechanics describes Wigner's friend's perceived outcomes before and after Wigner's measurement. Their theorem is exact and binary: it asks whether two effective perception observables commute. This note asks how much the friend's records must degrade before the obstruction disappears, and answers in closed form. Writing Wigner's basis parameter as c = |a² − b²| and smearing both perception observables to visibility η, a persistent-reality description exists if and only if η ≤ 2/(√(1−c²) + √(1+3c²)). The curve equals 1 exactly at the two compatible cases of the qualitative theorem and is minimized at √3/2 ≈ 0.866 when c² = 2/3. What is new, and what is not. Stated plainly, because each ingredient is due to someone else. The scenario, the assumptions P1–P3 and the two effective perception POVMs are from Guérin et al. The equivalence between a linear two-time distribution and a joint POVM is established in the proof of their Theorem III.1 and is used as stated there, not re-proved here. The joint-measurability criterion for unbiased qubit observables is Busch's (Phys. Rev. D 33, 2253 (1986)) and is standard. Given the Bloch vectors, the threshold follows from that criterion in two lines; this note does not present the computation as difficult. The contribution is the robustness question, the closed-form answer, the worst-case basis, and an elementary observation about the single-time claim. A clarification the note also records. The frequently repeated claim that no single global density operator reproduces both agents' descriptions is false in its single-time operational form: the two states agree on the friend's entire record algebra, differing only in coherences she cannot access, so Wigner's own state satisfies both agents' operational constraints. The claim becomes a theorem only through the two-time persistence requirement. This is elementary and expected to be folklore among specialists; it is recorded because the inference is common and because it locates where the real obstruction lives. Status and scope. This is a preprint, not a peer-reviewed article. Two honest caveats are stated in the text rather than buried. The noise model is a stipulation on the effective observables, not a decoherence channel derived from a modified unitary model; Del Santo, Manzano and Brukner (Phys. Rev. Research 7, 033279 (2025)) have a physically motivated degradation parameter for the same scenario, and the map between the two is not known here. And smearing the first, sharp observable means degrading the friend's perception of her own outcome, which requires an interpretive commitment some readers will not grant; smearing only the later readout provably achieves nothing, which is why the symmetric family is used. Verification. All algebra was recomputed symbolically and independently: reconstruction of the perception POVMs from the unitary model, reproducing Eqs. (12)–(15) of Guérin et al.; derivation of the threshold and of the minimum at c² = 2/3; and a semidefinite-programming feasibility test of the joint-POVM constraints at five values of c on both sides of the threshold, agreeing with the curve in all ten cases. The three scripts are deposited with the note. Use of AI assistance. This note was prepared with substantial assistance from AI language models, used for symbolic computation, drafting, literature search, and adversarial review of the arguments. Every mathematical claim was independently recomputed symbolically and cross-checked numerically. No specialist in quantum foundations has refereed it. The author is responsible for the content. This disclosure is made unconditionally.
The Wigner's friend thought experiment raises the question of whether quantum theory can be applied consistently to systems that include observers. Extended Wigner's Friend Scenarios develop this question by considering several observers who may assign different descriptions to the same experiment. In the Local Friendliness (LF) framework, these scenarios lead to experimentally testable inequalities derived from assumptions such as Absoluteness of Observed Events and Local Agency. Motivated by the “Thoughtful” version of the Local Friendliness no-go theorem, which points towards future LF tests with human-level artificial agents implemented on quantum computers, this thesis implements the “friend” with explicit agent-like functionality within a reversible quantum circuit. Several such agents are constructed and embedded into a one-friend Extended Wigner's Friend Scenario, following recent work that uses quantum computers as a platform for LF violations with more sophisticated “friend” systems. These agents store measurement outcomes, condition later operations on stored information, and, in the most structured case, use Born-rule probabilities to bet on the outcomes of their own future observations. The circuits are implemented in Qiskit, simulated ideally and with IBM-device noise models, and executed on ibm_marrakesh. In ideal simulations, all agent implementations reproduce the maximal quantum violation of a Local Friendliness inequality up to finite-shot fluctuations. Hardware runs show positive Local Friendliness violations for all implemented agents, with decreasing violation strength consistent with greater noise accumulation. The observed violations also remain above an estimated bound obtained by considering relaxations of a Local Friendliness assumption. These results provide a first step towards more structured agent-like friend models in LF experiments on quantum computers.
The famous Wigner's friend experiment considers an observer-the friend-and a superobserver-Wigner-who treats the friend as a quantum system and her interaction with other quantum systems as unitary dynamics. This is at odds with the friend describing this interaction via collapse dynamics, if she interacts with the quantum system in a way that she would consider a measurement. These different descriptions constitute the Wigner's friend paradox. Extended Wigner's friend experiments combine the original thought experiment with non-locality setups. This allows for deriving local friendliness inequalities, similar to Bell's theorem, which can be violated for certain extended Wigner's friend scenarios. A Wigner's friend paradox and the violation of local friendliness inequalities require that no classical record exists, which reveals the result the friend observed during her measurement. Otherwise, Wigner agrees with his friend's description and no local friendliness inequality can be violated. In this article, I introduce classical communication between Wigner and his friend and discuss its effects on the simple as well as extended Wigner's friend experiments. By controlling the properties of a (quasi) classical communication channel between Wigner and the friend, one can regulate how much outcome information about the friend's measurement is revealed. This gives a smooth transition between the paradoxical description and the possibility of violating local friendliness inequalities, on the one hand, and the effectively collapsed case, on the other hand.
The Wigner’s friend thought experiment has gained a resurgence of interest in recent years thanks to no-go theorems that extend it to Bell-like scenarios. One of these, by us and co-workers, showcased the contradiction that arises between quantum theory and a set of assumptions, weaker than those in Bell’s theorem, which we named ‘Local Friendliness’. Using these assumptions, it is possible to arrive at a set of inequalities for a given scenario, and, in general, some of these inequalities will be harder to violate than the Bell inequalities for the same scenario. A crucial feature of the extended Wigner’s friend scenario in our aforementioned work was the ability of a superobserver to reverse the unitary evolution that gives rise to their friend’s measurement. Here, we present a new scenario where the superobserver can interact with the friend repeatedly in a single experimental instance, either by asking them directly for their result, thus ending that instance, or by reversing their measurement and instructing them to perform a new one. We show that, in these scenarios, the Local Friendliness inequalities will always be the same as Bell inequalities.
This document connects Return Theory (RT) to quantum mechanics through two standard settings: Wigner's friend and quantum Darwinism. Each setting is read with RT's declared roles: record, reference, access and reconstruction. The document does not choose an interpretation of quantum mechanics and does not claim to solve the measurement problem. It states, in operational terms, what each observer can reconstruct, which reference validates that reconstruction, and when two observers can agree. Main results: • Content and phase are different records with different references. Wigner's coherence access gives no information about the friend's content (error 1/2). It records the relative phase exactly. • Record–interference complementarity (one access): (1−2ε₁)² + (1−2ε₂)² ≤ 1. Equality holds for projective measurements in the plane of the two tasks. • Two-system bound: if one system holds an ε₁-record of the content, any access to another system reads the phase with error ε₂, where 1−2ε₂ ≤ 2|αβ|·√(1−(1−2ε₁)²). • Agreement between observers follows from records when both are realised jointly in one run. Agreement alone is not a record (Bell-pair example). • Operational objectivity is defined by four conditions. In a branching environment, fragment errors are ε_f = ½(1−√(1−c^{2f})), and redundancy R_ε ≥ 2 is required. Mutual information shows a classical plateau at the pointer entropy. • Complementary tolerances for disjoint observers have a threshold of ½(1−1/√2) ≈ 0.146. A [[4,2,2]] code example shows task-separate (not simultaneous) objectivity of content and phase. All numbers in the text are generated from code (Python, exact state vectors). The package reproduces the JSON results and the generated TeX byte-for-byte. Figures may differ slightly across Matplotlib versions. Prior work (Wigner, Frauchiger–Renner, Brukner, Zurek, Englert, and others) is cited for comparison. Interactive language-model tools were used in preparing the text and code. The author is responsible for all content.