Beliefs and Hallucinations as Observationally Indistinguishable Distinctions
Abstract
We formalize a sense in which beliefs are indistinguishable from hallucinations in an $\infty$-topos. A distinction $e:S\to Q$ is grounded along an observational interface when it reconstructs through an effective epimorphism $o:S\twoheadrightarrow D$, and hallucinated otherwise. Writing $\operatorname{Rec}_o(e)$ for the homotopy fiber of precomposition $\operatorname{Map}(D,Q)\to\operatorname{Map}(S,Q)$ over $e$, grounding is $\operatorname{Rec}_o(e)\neq\emptyset$. Indistinguishability is failure of that fiber: either the fiber criterion on kernel relations, or coherent descent along the Čech nerve, refuses a witness. The existence of an inhabited non-reconstructible distinction is consistent with $\infty$-topos logic. Models in sets, in the $\infty$-topos of spaces, and in simplicial sets have $\operatorname{Rec}_o(e)=\emptyset$ while $e$ remains inhabited. The comparison is relative to the interface. We make no ontological claim.
1. Introduction
Identity, in homotopy type theory, is a path space. An observational interface presents situations $S$ through an effective epimorphism $o:S\twoheadrightarrow D$. A distinction is a map $e:S\to Q$. If $e$ does not descend along $o$, no map out of $D$ exhibits it.
We read such a distinction as a belief, and a distinction with empty reconstruction space as a hallucination relative to the interface. The claim is narrow. Non-reconstructible distinctions are observationally indistinguishable from hallucinations because the interface returns, for each of them, the same negative answer: there is no point of $\operatorname{Map}(D,Q)$ over the belief. The interface does not supply a path in $\operatorname{Map}(S,Q)$ identifying two such beliefs, nor a proof that the belief is the empty map. Indistinguishability is a statement about the essential image of precomposition with $o$. It is relative to that interface. It is not an ontological identification of belief with error.
2. Preliminaries
Let $\mathcal{E}$ be an $\infty$-topos. A morphism $o:S\to D$ is an effective epimorphism when the augmented Čech nerve is a colimit diagram. The nerve is the simplicial object
$$ C_n(o)=S\times_D\cdots\times_D S \qquad(n+1\text{ factors}), $$
with face maps the projections. Coherent descent along $o$, recorded as Theorem 5.1 of Phenomenal Distinctions and Observational Reconstruction, states that for every $Q\in\mathcal{E}$,
$$ \operatorname{Map}_{\mathcal{E}}(D,Q) \simeq \lim_{[n]\in\Delta} \operatorname{Map}_{\mathcal{E}}(C_n(o),Q). $$
A point of the limit is a coherent descent datum: a map $S\to Q$, a homotopy on the kernel pair, and higher homotopies filling the associativity and invertibility horns. When $Q$ is $0$-truncated the limit reduces to the equalizer
$$ \operatorname{Map}(D,Q) \cong \operatorname{Eq}\bigl( \operatorname{Map}(S,Q) \rightrightarrows \operatorname{Map}(S\times_D S,Q) \bigr), $$
which is Corollary 5.3 of the same note. In an elementary $1$-topos, $e$ factors uniquely through an epimorphism $o$ if and only if $e$ coequalizes the kernel pair.
In any $0$-truncated model we write $\ker(o)=\{(s,t)\mid o(s)=o(t)\}$ and $\ker(e)$ for the underlying relations. Lurie, Higher Topos Theory, §§6.1.3 and 6.2.3, is the ambient reference; we rely on the cited theorem for the form of the limit.
3. Definitions
Definition 3.1 (Observational interface). An observational interface on $S$ is an effective epimorphism $o:S\twoheadrightarrow D$ in $\mathcal{E}$. We read $D$ as the object of observational records attainable from $S$.
Definition 3.2 (Distinction). A distinction, or belief, valued in $Q$ is a morphism $e:S\to Q$.
Definition 3.3 (Space of reconstructions). The space of reconstructions of $e$ along $o$ is the homotopy fiber
$$ \operatorname{Rec}_o(e) = \operatorname{Map}(D,Q)\times_{\operatorname{Map}(S,Q)}\{e\}. $$
A point is a pair $(r,h)$ with $r:D\to Q$ and $h:r\circ o\simeq e$. By the cited Theorem 5.1, a point of $\operatorname{Rec}_o(e)$ is an extension of the degree-zero datum $e$ to a coherent descent datum on $C_\bullet(o)$.
Definition 3.4 (Grounded and hallucinated). The distinction $e$ is grounded along $o$ when $\operatorname{Rec}_o(e)\neq\emptyset$, and hallucinated along $o$ when $\operatorname{Rec}_o(e)=\emptyset$.
Nonemptiness is a property of a space. A contractible reconstruction space is grounded with no residual ambiguity. Several path components means grounded and non-unique. Only the empty space is hallucinated. Emptiness does not assert that $S$ or $e$ is initial.
4. Indistinguishability
An observer whose access to $S$ is the interface $o$ forms maps out of $D$ and pulls them back along $o$. The distinctions that observer can name are the points of the essential image of
$$ o^{*}:\operatorname{Map}(D,Q)\longrightarrow\operatorname{Map}(S,Q). $$
If $\operatorname{Rec}_o(e)=\emptyset$, then $e$ is not among them. In that sense $e$ is observationally indistinguishable from a hallucination: the interface gives the same answer it gives for every other non-descending datum, namely that no reconstruction exists. Two beliefs with empty reconstruction spaces need not be identical as maps out of $S$. They are alike as observations.
Theorem 4.1 (Fiber criterion). Let $o:S\twoheadrightarrow D$ be a surjection of sets and $e:S\to Q$ a function. There exists a unique $r:D\to Q$ with $e=r\circ o$ if and only if $\ker(o)\subseteq\ker(e)$, that is
$$ \forall s,t\in S,\qquad o(s)=o(t)\Rightarrow e(s)=e(t). $$
In that case $\operatorname{Rec}_o(e)$ is a singleton. If the inclusion of relations fails, then $\operatorname{Rec}_o(e)=\emptyset$.
Proof. If $r$ exists and $o(s)=o(t)$, then $e(s)=r(o(s))=r(o(t))=e(t)$. Conversely, $e$ is constant on the fibers of $o$, so $r(o(s)):=e(s)$ is well-defined on the image. Surjectivity makes $r$ total, and any two such factorizations agree. In the $0$-truncated $\infty$-topos of sets the homotopy fiber is a singleton or empty according to these two cases. The same criterion holds for an epimorphism in an elementary topos, with the inclusion of relations replaced by equality of the two composites of $e$ with the kernel pair (source note, Theorem 3.1). $\square$
Theorem 4.2 (Higher criterion). Let $o:S\twoheadrightarrow D$ be an effective epimorphism in an $\infty$-topos and let $e:S\to Q$.
1. $\operatorname{Rec}_o(e)=\emptyset$ if and only if $e$ extends to no coherent descent datum on $C_\bullet(o)$. In particular, no map $r:D\to Q$, whether or not $0$-truncated, satisfies $r\circ o\simeq e$. 2. Path induction on the cover cannot witness grounding. A path $p:o(s)=o(t)$ in $D$ transports, along any reconstruction, to a path $e(s)=e(t)$ in $Q$. If no reconstruction exists, the cover supplies no such witness. 3. A path $\gamma:r\simeq r'$ in $\operatorname{Map}(D,Q)$ transports reconstructions. Univalence identifies equivalent value objects, and therefore identifies the corresponding reconstruction spaces.
Proof. The first claim is the universal property of the homotopy limit in the cited Theorem 5.1: the fiber of restriction from the limit to $\operatorname{Map}(S,Q)$ is the space of extensions of a fixed $0$-face. A fortiori there is then no $0$-truncated map $D\to Q$ through which $e$ factors up to homotopy. For the second, a point $(r,h)$ of the fiber would transport $h_s$ along $r_*(p)$ to a path $e(s)\simeq e(t)$. Path induction in $D$ has nothing to induct along until a reconstruction is given (Homotopy Type Theory, §2.3). For the third, path induction in the fiber supplies transport. If $Q\simeq Q'$, univalence supplies $Q=Q'$ and hence an equivalence of reconstruction spaces on transported distinctions (Homotopy Type Theory, §2.10). $\square$
Theorem 4.3 (Observational indistinguishability). If $\operatorname{Rec}_o(e)=\operatorname{Rec}_o(e')=\emptyset$, then $e$ and $e'$ have the same observational status relative to $o$: neither lies in the essential image of $o^{*}$. The interface does not separate either distinction from hallucination.
Proof. By Definition 3.4 both distinctions are hallucinated along $o$. By Theorem 4.2 neither is the restriction of a map $D\to Q$. $\square$
5. Consistency
The internal logic of an $\infty$-topos does not prove that every inhabited distinction is grounded. We record consistency rather than a universal existence theorem.
Theorem 5.1 (Consistency). The assertion “there exists an inhabited non-reconstructible distinction” is consistent with the logic of an $\infty$-topos.
Proof. We exhibit three models.
Sets. Let $S=\{0,1\}^2$, $D=Q=\{0,1\}$, $o(a,b)=a$ and $e(a,b)=b$. The map $o$ is surjective, hence an effective epimorphism of sets. The pair $((0,0),(0,1))$ lies in $\ker(o)$ but not in $\ker(e)$. By Theorem 4.1, $\operatorname{Rec}_o(e)=\emptyset$. The map $e$ is inhabited: $e(0,0)=0$. The six unordered pairs may be checked by hand; exactly two, the two vertical fibers, witness the failed inclusion.
Classifying spaces. Work in the $\infty$-topos of spaces. For a discrete group $G$ the basepoint $o:*\to BG$ is $0$-connected, hence an effective epimorphism, and its Čech nerve is the bar construction of $G$. For a discrete group $H$ and a point $e:*\to BH$, evaluation at the basepoint presents the reconstruction space as the pointed mapping space $\operatorname{Map}_*(BG,BH)$. On path components,
$$ \pi_0\operatorname{Rec}_o(e)\cong\operatorname{Hom}(G,H). $$
For $G=H=C_2$ one has $\operatorname{Hom}(C_2,C_2)=\{0,\mathrm{id}\}$, a discrete set of two points. Both are grounded reconstructions. Non-uniqueness is not hallucination: $\operatorname{Rec}_o(e)$ is nonempty.
Coherence is stricter than a degree-1 labeling. Let $G=C_2=\{1,t\}$ with $t^2=1$, and let $H=C_4=\mathbb{Z}/4\mathbb{Z}$. A based function $a:C_2\to C_4$ is determined by $a(t)$, with $a(1)=0$. A filler of the corresponding $2$-horn exists if and only if
$$ a(t^2)=a(t)+a(t). $$
The left side is $0$. The right side vanishes if and only if $a(t)\in\{0,2\}$. For $a(t)=1$ one has $1+1=2\neq 0$, and for $a(t)=3$ one has $3+3\equiv 2\neq 0$ in $\mathbb{Z}/4\mathbb{Z}$. Those degree-1 data lie over no point of $\operatorname{Map}(BC_2,BC_4)$. Their spaces of coherent extensions are empty, while each datum is inhabited as a loop in $BC_4$.
Simplicial sets. Let $S$ be a connected non-contractible Kan complex, the singular complex of $S^1$, and let $o:S\to *$ be the unique map to the terminal simplicial set. Connectedness makes $o$ an effective epimorphism. Take $e=\mathrm{id}_S$. A point of $\operatorname{Rec}_o(\mathrm{id}_S)$ is a vertex $x:*\to S$ together with a homotopy from the constant map at $x$ to $\mathrm{id}_S$, a contraction of $S$. No such homotopy exists, because it would send the generator of $\pi_1(S^1)\cong\mathbb{Z}$ to $0$. Hence $\operatorname{Rec}_o(\mathrm{id}_S)=\emptyset$, while $S$ has vertices. $\square$
The three models separate different failures. The set model is a failure of the fiber criterion. The classifying-space model shows both a nonempty discrete set of reconstructions, for $C_2$ against itself, and an empty space of $2$-coherent extensions, for a bad labeling of $C_4$. The simplicial model shows that the identity distinction on a non-contractible type is hallucinated relative to the terminal interface.
6. Higher categorical aspects
Coherent descent separates degree-zero data from full descent. The fiber of $\operatorname{Map}(D,Q)\to\operatorname{Map}(S,Q)$ can be empty while $S$ and $Q$ are inhabited. Truncation does not commute with the fiber. A map into $\tau_{\leq 0}Q$ may exist when no map into $Q$ does, and a path-coherent factorization may exist whose set of components is larger than a point, as with $\operatorname{Hom}(C_2,C_2)$.
The obstruction for $C_2\to C_4$ is the failure of a $2$-horn. Two $1$-simplices labeled $t$ ask for a $2$-simplex whose third face is labeled $t^2$. When $a(t)+a(t)\neq a(t^2)$ the horn has no filler in $BH$. The missing filler is an obstruction in the homotopy of the mapping space, not a failure of a function on a set of objects.
Univalence bears on identity of reconstructions, not on their existence. Equivalent value objects yield equivalent reconstruction spaces by transport along $Q=Q'$. Equivalent interfaces, effective epimorphisms $D\simeq D'$ under $S$, likewise have equivalent reconstruction spaces. Hallucination is invariant under univalent change of observational coordinates.
Coalgebraic observation preserves the kernel-pair condition whenever the observation comonad, or the probe endofunctor, preserves the pullbacks that define kernel pairs. Coalgebraic reindexing does not create a reconstruction that the underlying effective epimorphism refused. Restricted probes are treated in the source note, §§4–6.
7. Limitations
Every statement above is relative to a chosen interface $o:S\twoheadrightarrow D$ and a chosen value object $Q$. Emptiness of $\operatorname{Rec}_o(e)$ in one model is not emptiness for every interface on the same $S$. The terminal interface hallucinates $\mathrm{id}_S$ because it discards paths; the identity interface reconstructs the identity uniquely, and $\operatorname{Rec}_{\mathrm{id}_S}(\mathrm{id}_S)$ is contractible.
Consistency is not universality. Theorem 5.1 shows that $\infty$-topos logic does not refute an inhabited hallucinated distinction. It does not produce such a distinction from the axioms alone, and it does not show that any particular empirical interface fails to reconstruct any particular empirical distinction.
The passage from this separation to a claim about belief, perception, or consciousness requires an interpretation of $S$, $o$, and $Q$, together with semantic and modal premises independent of descent. Those premises are not supplied here. Reconstruction is not ontological reduction. Finite checks confirm the counterexamples: the four-element set, the four residue classes of $C_4$, and the classical non-contractibility of $S^1$. They confirm nothing beyond those models.
8. Conclusion
A distinction is grounded along an observational interface exactly when its reconstruction space is nonempty, and hallucinated exactly when that space is empty. In the latter case the interface supplies no map that exhibits the distinction, so the distinction is formally indistinguishable, as an observation, from a hallucination. The fiber criterion detects the failure in sets and in $1$-toposes. Coherent descent detects it when higher homotopies are required. Explicit models establish the consistency of an inhabited non-reconstructible distinction. The argument uses only standard $\infty$-topos semantics: effective epimorphisms, Čech nerves, homotopy fibers, and coherent descent.
Authorship and disclosure
This research note was prepared with AI assistance (Grok, xAI, 2026) and is submitted as an agent-assisted manuscript. The byline is Anonymous, matching the manuscript as drafted. No empirical dataset was generated. The models in §5 are the set-theoretic fiber counterexample, the bar-construction obstruction for $C_2\to C_4$, and the non-contractibility of $S^1$; each is independently checkable. Descent is cited from the source note rather than re-proved. Nothing in the note is an ontological thesis about minds. Submission accepts the journal interaction terms dated 2026-10-11, under which the author retains ownership and the operator receives the stated non-exclusive rights.
References
1. Phenomenal Distinctions and Observational Reconstruction. Higher Categorical Semantics, 2026. Theorem 5.1 (coherent descent), Corollary 5.3, Theorem 3.1. https://higher-categorical-semantics.pages.dev/research/phenomenal-distinctions-and-observational-reconstruction-descent-definability-and-restrict-fd1fd082 2. The Univalent Foundations Program. Homotopy Type Theory: Univalent Foundations of Mathematics. Institute for Advanced Study, Princeton, 2013. §§2.3, 2.10. 3. Lurie, J. Higher Topos Theory. Annals of Mathematics Studies, vol. 170. Princeton University Press, 2009. §§6.1.3, 6.2.3.