Counterfactual Local Friendliness in Quantum Foundations

• Counterfactual Local Friendliness (CLF) is a quantum-foundational framework that reformulates local friendliness no-go theorems using counterfactual and causal models to pinpoint conflicts in agent-level facts.
• It employs potential-outcome variables and d-separation criteria to derive CHSH-type inequalities that establish stricter bounds than traditional Bell tests.
• The framework extends to interaction-free measurement protocols with ε-bounded disturbance, providing a precise tool to diagnose inconsistencies in single-world interpretations.

Searching arXiv for the quantum Local Friendliness / Counterfactual Local Friendliness literature to ground the article. arXiv search: query = "Local Friendliness Wigner's Friend Bong 2020 Counterfactual Local Friendliness" Counterfactual Local Friendliness (CLF) is a quantum-foundational framework that recasts the Local Friendliness no-go theorem in explicitly counterfactual and causal terms, and in later work extends that framework to an interaction-free, disturbance-bounded Wigner’s-friend paradox. In the counterfactual formulation, CLF expresses the conjunction of Absoluteness of Observed Events (AOE), No Superdeterminism (NS), and counterfactual Locality within a structural-causal or potential-outcome language; in the later interaction-free formulation, it sharpens the same tension by replacing in-lab measurements with $\epsilon$-counterfactual interaction-free modules while preserving a nonzero paradoxical post-selection (Cavalcanti et al., 2021, Liechtenstein, 1 Sep 2025).

1. Conceptual placement and defining assumptions

The Local Friendliness theorem of Bong et al. (2020), as reformulated causally, states that no model satisfying three assumptions can reproduce the quantum statistics of an Extended Wigner’s Friend protocol without contradiction: Absoluteness of Observed Events (AOE), according to which every recorded outcome is a single, absolute event; No Superdeterminism (NS), according to which measurement-setting choices are statistically independent of any prior variables; and Locality (L), according to which the probability of an outcome cannot be influenced by a space-like-separated setting intervention, even conditional on other past events (Cavalcanti et al., 2021).

In the counterfactual presentation, these assumptions are encoded through potential-outcome variables

$A_{x,c}, \qquad B_{y,d},$

meaning the outcome Alice would record if she chose setting $x$ and the friend’s outcome was $c$, and analogously for Bob. The locality conditions are then written as

$$A_{x,c} \perp\!\!\!\perp (Y,D), \qquad B_{y,d} \perp\!\!\!\perp (X,C),$$

together with

$$P(A_{x,c}=a)=P(A=a\mid X=x,C=c), \qquad P(B_{y,d}=b)=P(B=b\mid Y=y,D=d),$$

plus the consistency requirement that the actually observed $A$ and $B$ agree with their appropriate potential outcomes. In this sense, CLF is the conjunction of AOE, NS, and counterfactual Locality cast in the standard potential-outcome language of structural causal models (Cavalcanti et al., 2021).

A later development introduces a different but closely related use of the same term. There, Counterfactual Local Friendliness denotes a Wigner’s-friend-type logical collision built from interaction-free flags whose disturbance on the probed object is bounded by a tunable parameter $\epsilon$. The four assumptions are (Q) universal unitarity for outside observers, (S) single-outcome facts, (C) cross-agent consistency, and (IF-$\epsilon$) $A_{x,c}, \qquad B_{y,d},$0-counterfactuality of the friends’ internal modules (Liechtenstein, 1 Sep 2025). This later formulation preserves the central issue of Local Friendliness—compatibility of agent-level facts—while targeting the possible objection that the paradox is an artifact of invasive measurement.

2. Structural-causal formulation

The causal model underlying CLF uses the variables

$A_{x,c}, \qquad B_{y,d},$1

where $A_{x,c}, \qquad B_{y,d},$2 and $A_{x,c}, \qquad B_{y,d},$3 are Alice’s and Bob’s measurement settings, $A_{x,c}, \qquad B_{y,d},$4 and $A_{x,c}, \qquad B_{y,d},$5 are the “friend” outcomes inside each laboratory, $A_{x,c}, \qquad B_{y,d},$6 and $A_{x,c}, \qquad B_{y,d},$7 are the superobservers’ final recorded outcomes, and $A_{x,c}, \qquad B_{y,d},$8 is a latent common-cause variable residing in the joint past of $A_{x,c}, \qquad B_{y,d},$9 and $x$0 (Cavalcanti et al., 2021).

Space-time separation leads to the directed acyclic graph

$x$1

with no arrows such as $x$2 or $x$3. By the Causal Markov Condition, any joint distribution compatible with that graph must factorize as

$x$4

One may equivalently work with the potential-outcome variables $x$5 and $x$6, together with the consistency constraints

$x$7

The corresponding d-separation statements are

$x$8

and these d-separation relations are exactly the counterfactual Locality assumptions underwriting the theorem (Cavalcanti et al., 2021).

This formulation is significant because it places Extended Wigner’s Friend scenarios within the same formal idiom used in causal inference and structural-causal modeling. A plausible implication is that CLF is best understood not merely as an interpretational slogan but as a precise incompatibility claim about which joint distributions and counterfactual variables may consistently coexist.

3. CLF inequalities and their status relative to Bell-type bounds

From the factorized LF model, one derives, for each value of $x$9, correlators

$c$0

and then the CHSH-type expression

$c$1

which obeys

$c$2

Averaging over $c$3 yields

$c$4

where

$c$5

This is the simplest CLF inequality (Cavalcanti et al., 2021).

Although formally identical to CHSH, the inequality has a different logical standing. It holds under the Local Friendliness assumptions, which the cited analysis describes as strictly weaker in the relevant sense, because $c$6 and $c$7 appear explicitly as local inputs in the model and AOE is needed to justify the well-defined joint distribution $c$8 (Cavalcanti et al., 2021). The same framework also permits further “mixed” inequalities involving correlations of $c$9 with $$A_{x,c} \perp\!\!\!\perp (Y,D), \qquad B_{y,d} \perp\!\!\!\perp (X,C),$$0 and of $$A_{x,c} \perp\!\!\!\perp (Y,D), \qquad B_{y,d} \perp\!\!\!\perp (X,C),$$1 with $$A_{x,c} \perp\!\!\!\perp (Y,D), \qquad B_{y,d} \perp\!\!\!\perp (X,C),$$2; these are facets of the full LF-polytope in the 16-dimensional space of $$A_{x,c} \perp\!\!\!\perp (Y,D), \qquad B_{y,d} \perp\!\!\!\perp (X,C),$$3.

The comparison with Bell’s theorem is therefore central. The causal analysis argues that Local Friendliness puts stronger bounds on quantum reality than Bell’s theorem, because the usual Bell escape route through quantum causal models targets an additional decorrelation or factorization assumption that LF does not require. This suggests that CLF should be read as a strengthening of Bell-type no-go reasoning in scenarios where internal observers’ records are themselves treated as physical events.

4. Quantum realization and experimental status

A quantum implementation of the Extended Wigner’s Friend protocol proceeds by preparing an entangled Bell pair $$A_{x,c} \perp\!\!\!\perp (Y,D), \qquad B_{y,d} \perp\!\!\!\perp (X,C),$$4. Alice’s friend, Charlie, measures qubit $$A_{x,c} \perp\!\!\!\perp (Y,D), \qquad B_{y,d} \perp\!\!\!\perp (X,C),$$5 in the $$A_{x,c} \perp\!\!\!\perp (Y,D), \qquad B_{y,d} \perp\!\!\!\perp (X,C),$$6 basis and records outcome $$A_{x,c} \perp\!\!\!\perp (Y,D), \qquad B_{y,d} \perp\!\!\!\perp (X,C),$$7; Bob’s friend, Debbie, does likewise on qubit $$A_{x,c} \perp\!\!\!\perp (Y,D), \qquad B_{y,d} \perp\!\!\!\perp (X,C),$$8 and records $$A_{x,c} \perp\!\!\!\perp (Y,D), \qquad B_{y,d} \perp\!\!\!\perp (X,C),$$9. Alice and Bob then choose either to read out the friend’s result, corresponding to settings $$P(A_{x,c}=a)=P(A=a\mid X=x,C=c), \qquad P(B_{y,d}=b)=P(B=b\mid Y=y,D=d),$$0 or $$P(A_{x,c}=a)=P(A=a\mid X=x,C=c), \qquad P(B_{y,d}=b)=P(B=b\mid Y=y,D=d),$$1, or to reverse the friend’s unitary coupling and measure the bare qubit in an $$P(A_{x,c}=a)=P(A=a\mid X=x,C=c), \qquad P(B_{y,d}=b)=P(B=b\mid Y=y,D=d),$$2 basis, corresponding to settings $$P(A_{x,c}=a)=P(A=a\mid X=x,C=c), \qquad P(B_{y,d}=b)=P(B=b\mid Y=y,D=d),$$3 or $$P(A_{x,c}=a)=P(A=a\mid X=x,C=c), \qquad P(B_{y,d}=b)=P(B=b\mid Y=y,D=d),$$4 (Cavalcanti et al., 2021).

The resulting joint statistics $$P(A_{x,c}=a)=P(A=a\mid X=x,C=c), \qquad P(B_{y,d}=b)=P(B=b\mid Y=y,D=d),$$5 violate the CLF inequality up to the Tsirelson bound,

$$P(A_{x,c}=a)=P(A=a\mid X=x,C=c), \qquad P(B_{y,d}=b)=P(B=b\mid Y=y,D=d),$$6

provided one can in principle perform the friend-erasing unitaries and $$P(A_{x,c}=a)=P(A=a\mid X=x,C=c), \qquad P(B_{y,d}=b)=P(B=b\mid Y=y,D=d),$$7-measurements. In the photonic proof-of-principle of Proietti et al. (2019), the “friend” was embodied by the photon’s path qubit, and a violation of order

$$P(A_{x,c}=a)=P(A=a\mid X=x,C=c), \qquad P(B_{y,d}=b)=P(B=b\mid Y=y,D=d),$$8

was observed (Cavalcanti et al., 2021).

Two points follow from this implementation. First, the violation is not merely abstract: it arises from a concrete protocol in which internal records and external measurements are jointly modeled. Second, the role of the friend is not eliminable. The variables $$P(A_{x,c}=a)=P(A=a\mid X=x,C=c), \qquad P(B_{y,d}=b)=P(B=b\mid Y=y,D=d),$$9 and $A$0 are not hidden auxiliaries but the internal event-records whose absolute status is under pressure. That feature distinguishes CLF from a standard Bell experiment, even where the inequality takes an algebraically similar form.

5. The interaction-free $A$1-bounded CLF paradox

The 2025 formulation introduces a new paradox called Counterfactual Local Friendliness in which every decisive inference is obtained by interaction-free flags whose disturbance on the probed object is bounded by $A$2 (Liechtenstein, 1 Sep 2025). The setting consists of two spatially separated laboratories, $A$3 and $A$4, each containing a friend who implements an interaction-free measurement oracle on a local two-level system (“bomb”) $A$5, with $A$6. Each oracle is a unitary device acting on a mediator qubit $A$7, the bomb $A$8, and a flag qubit $A$9. When the flag is $B$0 (“Dark”), the bomb is left undisturbed up to trace-distance $B$1 and one can be certain that $B$2; when the flag is $B$3 (“Bright”), one infers $B$4 (Liechtenstein, 1 Sep 2025).

An upstream coin qubit is prepared in

$B$5

and a controlled isometry $B$6 coherently copies it into two lab registers:

$B$7

with

$B$8

The bomb encoding is arranged so that

$B$9

and specifically so that a Dark flag in laboratory $\epsilon$0 implies $\epsilon$1, whereas a Dark flag in laboratory $\epsilon$2 implies $\epsilon$3. A routing qubit sends the mediator $\epsilon$4 in superposition to both oracles, allowing both friends to register Dark simultaneously. One then post-selects on

$\epsilon$5

The assumptions are:

Assumption	Content
(Q)	Universal unitarity for outside observers
(S)	Single-outcome facts
(C)	Cross-agent consistency
(IF-$\epsilon$6)	$\epsilon$7-counterfactuality of the IFM oracles

The formal condition for $\epsilon$8-counterfactuality is that, for every bomb state $\epsilon$9 in the relevant set $\epsilon$0 and every mediator state $\epsilon$1,

$\epsilon$2

Thus, whenever the Dark flag fires, the bomb’s reduced state changes by at most trace-distance $\epsilon$3 (Liechtenstein, 1 Sep 2025).

The paradox follows directly on the post-selected runs. Under (IF-$\epsilon$4), each Dark flag implies that the corresponding bomb was in the live state with certainty; under (C), all observers must agree on that fact. But by construction,

$\epsilon$5

whereas

$\epsilon$6

Hence on the same runs one agent must assert $\epsilon$7 and another must assert $\epsilon$8, contradicting (S). The post-selected event has nonzero probability

$\epsilon$9

and for symmetric routing the amplitude for both flags dark is $A_{x,c}, \qquad B_{y,d},$00, so

$A_{x,c}, \qquad B_{y,d},$01

The contradiction is therefore not confined to a null set (Liechtenstein, 1 Sep 2025).

6. Noncontextuality bound, interpretational consequences, and points of dispute

The same 2025 analysis derives an $A_{x,c}, \qquad B_{y,d},$02-IF three-box noncontextual bound. For the standard three-box pre-/post-selection setup with

$A_{x,c}, \qquad B_{y,d},$03

one uses an $A_{x,c}, \qquad B_{y,d},$04-counterfactual interaction-free measurement to test whether the particle is in box $A_{x,c}, \qquad B_{y,d},$05 or box $A_{x,c}, \qquad B_{y,d},$06. In any single-world, noncontextual model satisfying exclusivity and $A_{x,c}, \qquad B_{y,d},$07-stability, one must have

$A_{x,c}, \qquad B_{y,d},$08

or equivalently, in the notation of the detailed derivation,

$A_{x,c}, \qquad B_{y,d},$09

with $A_{x,c}, \qquad B_{y,d},$10 independent of $A_{x,c}, \qquad B_{y,d},$11 (Liechtenstein, 1 Sep 2025).

Quantum theory violates this bound. By the Aharonov–Bergmann–Lebowitz rule,

$A_{x,c}, \qquad B_{y,d},$12

and replacing the projective probes by $A_{x,c}, \qquad B_{y,d},$13-counterfactual IFMs leaves these probabilities at $A_{x,c}, \qquad B_{y,d},$14. In the ideal limit,

$A_{x,c}, \qquad B_{y,d},$15

for arbitrarily small $A_{x,c}, \qquad B_{y,d},$16 (Liechtenstein, 1 Sep 2025).

Two controversies are addressed directly in the literature. The first concerns whether the paradox can be attributed to hidden disturbance or energy exchange. The interaction-free version is designed to undercut that objection: the decisive inferences are obtained from Dark-port events whose disturbance is explicitly bounded by $A_{x,c}, \qquad B_{y,d},$17, and the argument states that the paradox is therefore not an energy-exchange paradox but a conflict among agent-level facts in a single-world narrative (Liechtenstein, 1 Sep 2025). The second concerns whether quantum causal models provide an escape. The causal analysis argues that proposals such as those of Leifer–Spekkens, Pienaar–Brukner, Costa–Shrapnel, Allen–Barrett–Spekkens, and Oreshkov–Costa–Brukner keep the same relativistic DAG and freedom of settings while dropping only the classical decorrelation step. That move can evade Bell’s 1976 theorem but does not help with Local Friendliness, because the LF inequalities already assume only AOE, Locality, and No Superdeterminism, not the additional decorrelation assumption those models sacrifice (Cavalcanti et al., 2021).

The resulting interpretational pressure is accordingly sharper than in ordinary Bell analysis. If one wishes to retain relativistic causal order together with the universal validity of quantum predictions in Wigner–Friend scenarios, the cited work argues that the remaining assumption to relinquish is the Absoluteness of Observed Events. In that case there is no globally well-defined joint distribution $A_{x,c}, \qquad B_{y,d},$18 on which the CLF inequalities can be imposed; instead, event records become relative to the agents who record them. A plausible implication is that CLF functions as a diagnostic for exactly where single-world, agent-independent narratives fail: not at the level of setting independence or signal locality, but at the level of jointly absolute facts (Cavalcanti et al., 2021).