Thread 2 — CIT Triangle¶

Jean-Paul Niko · February 2026

\fi

%═══════════════════════════════════════════════════════════════════════════════

Motivation: Three Logics of Cognition¶

%═══════════════════════════════════════════════════════════════════════════════

The Conceptual Irreversibility Theorem (CIT) of Part V proved that translation between the classical region \(\Ccl\) and the conceptual region \(\Cco\) is necessarily lossy. The proof exploited the mismatch between the subobject classifiers: Boolean (\(\{0, 1\}\)) in \(\Ccl\) versus Heyting (a lattice with intermediate truth values) in \(\Cco\).

Part X introduced a third region \(\Cqu\), governed by orthomodular logic, and conjectured a three-way CIT triangle. The Filter Paper (Section 7) showed that cross-logic filters are doubly lossy. We now prove the triangle rigorously, classifying all six directed translation functors and their kernels.

The three logics correspond to three kinds of cognitive substrate: [nosep] - Boolean (\(\Ccl\)): classical digital computation, crisp propositions. - Heyting (\(\Cco\)): biological cognition with graded truth, conceptual vagueness. - Orthomodular (\(\Cqu\)): quantum substrates, superposition, non-distributive logic.

\tB\; The three-space ontology of Part XIII reveals that the three logics are not arbitrary---they are the native logics of the three co-primordial spaces: \begin{center}

[Table — see PDF]

\end{center} The CIT triangle is thus not merely a result about substrates but about the fundamental ontology: the three spaces are logically incompatible in a deep structural sense. Translation between spaces is necessarily lossy because each space carries a different logic, and logic-type mismatches induce irreducible information loss. The CIT is a theorem about the three-space ontology, not merely about cognitive substrates.

This explains why the CIT is so robust: it is not an artefact of current substrate limitations but a consequence of the co-primordial structure of reality. No advance in substrate engineering can eliminate the CIT, because the logical incompatibility is baked into the three-space foundation.

%═══════════════════════════════════════════════════════════════════════════════

Lattice Preliminaries¶

%═══════════════════════════════════════════════════════════════════════════════

\tA\;

A bounded lattice \((L, \leq, \wedge, \vee, 0, 1)\) is: [nosep] - Boolean if it is complemented and distributive: \(a \wedge (b \vee c) = (a \wedge b) \vee (a \wedge c)\) and every \(a\) has a unique complement \(\neg a\) with \(a \wedge \neg a = 0\), \(a \vee \neg a = 1\). - Heyting if it is distributive and for every pair \((a, b)\) there exists a largest element \(c\) with \(a \wedge c \leq b\), the relative pseudocomplement \(a \to b\). Every Boolean algebra is Heyting; the converse fails. - Orthomodular if it is complemented (with orthocomplement \(a^{\perp}\), \(a \wedge a^{\perp} = 0\), \(a \vee a^{\perp} = 1\)) and satisfies the orthomodular law: if \(a \leq b\) then \(b = a \vee (a^{\perp} \wedge b)\). Orthomodular lattices need not be distributive.

\tA\;

The structural obstructions to inter-logic translation are: [nosep] - Heyting gap \(\Hgap\): the set of elements \(a\) in a Heyting algebra for which \(a \vee \neg a \neq 1\) (excluded middle fails). \(\Hgap = \emptyset\) iff the Heyting algebra is Boolean. - Quantum gap \(\Qgap\): the set of triples \((a, b, c)\) in an orthomodular lattice for which \(a \wedge (b \vee c) \neq (a \wedge b) \vee (a \wedge c)\) (distributivity fails). \(\Qgap = \emptyset\) iff the lattice is distributive (hence Boolean if also complemented). - Double gap \(\Dgap = \Hgap \cup \Qgap\): the combined obstruction for Heyting--orthomodular translation. Both excluded middle and distributivity may fail simultaneously in distinct ways.

%═══════════════════════════════════════════════════════════════════════════════

The Six Translation Functors¶

%═══════════════════════════════════════════════════════════════════════════════

For each ordered pair of logic types, there is a best approximation functor---the closest faithful representation of the source lattice in the target logic.

\tA\;

The six directed translation functors are: \begin{keyeq} [ \begin{tikzcd}[row sep=2.5em, column sep=3em] & \Ccl \; (\text{Boolean}) \arrow[dl, "B_H"', bend right=10] \arrow[dr, "B_Q", bend left=10] &
\Cco \; (\text{Heyting}) \arrow[ur, "H_B"', bend right=10] \arrow[rr, "H_Q"', bend right=10] & & \Cqu \; (\text{Orthomodular}) \arrow[ul, "Q_B", bend left=10] \arrow[ll, "Q_H", bend right=10] \end{tikzcd} ] \end{keyeq} where: [nosep] - \(H_B : \Cco \to \Ccl\) collapses intermediate truth values to \(\{0, 1\}\) (the original CIT functor). - \(B_H : \Ccl \to \Cco\) embeds Boolean into Heyting (lossless: Boolean is a sub-logic of Heyting). - \(Q_B : \Cqu \to \Ccl\) forces distributivity by collapsing superpositions to definite states ("measurement"). - \(B_Q : \Ccl \to \Cqu\) embeds Boolean into orthomodular (lossless: Boolean is a sub-logic of orthomodular). - \(H_Q : \Cco \to \Cqu\) maps Heyting into orthomodular, losing both intermediate truth values and gaining non-distributive structure. - \(Q_H : \Cqu \to \Cco\) maps orthomodular into Heyting, forcing distributivity but allowing intermediate truth values.

%═══════════════════════════════════════════════════════════════════════════════

The CIT Triangle Theorem¶

%═══════════════════════════════════════════════════════════════════════════════

\tA\;

Let \(L_B\) be a non-trivial Boolean algebra, \(L_H\) a non-Boolean Heyting algebra, and \(L_Q\) a non-Boolean orthomodular lattice. Then:

• Every translation functor except \(B_H\) and \(B_Q\) has nontrivial kernel: information is irreversibly lost.
• The kernel of each lossy functor is classified by the corresponding gap: \begin{keyeq}

[\begin{aligned} \ker(H_B) &= \Hgap(L_H) & &\text{(Heyting gap: lost vagueness)}

\ker(Q_B) &= \Qgap(L_Q) & &\text{(Quantum gap: lost superposition)}

\ker(Q_H) &\supseteq \Qgap(L_Q) & &\text{(Quantum gap persists into Heyting)}

\ker(H_Q) &\supseteq \Hgap(L_H) & &\text{(Heyting gap persists into orthomodular)}

\end{aligned}\]

\end{keyeq}

• The round-trip translations are strictly lossy:

• The composite functors around the full triangle accumulate losses: \begin{keyeq} [ \ker(H_B \circ Q_H \circ B_Q) \supseteq \Qgap \cup \Hgap ] \end{keyeq} Going Boolean \(\to\) orthomodular \(\to\) Heyting \(\to\) Boolean loses both superposition structure and intermediate truth values.

We prove each part.

Part 1 (embedding functors are lossless): \(B_H\) embeds \(L_B\) into \(L_H\) by treating each Boolean element as a Heyting element (Boolean algebras are Heyting algebras). This is a faithful functor: the lattice operations are preserved, and the embedding is injective. Similarly, \(B_Q\) embeds \(L_B\) into \(L_Q\) (Boolean algebras are orthomodular). All other functors are surjective approximations, hence potentially lossy.

Part 2 (kernel classification): \eqref{eq:ker-HB}: \(H_B\) maps each \(a \in L_H\) to the nearest Boolean element: \(H_B(a) = 1\) if \(a \vee \neg a = 1\) (decidable), and \(H_B(a) = 0\) or \(1\) by a forced choice otherwise. Elements in \(\Hgap\) have no faithful Boolean image; the kernel is exactly \(\Hgap\). This is the original CIT proof from Part V.

\eqref{eq:ker-QB}: \(Q_B\) maps each element of \(L_Q\) to a Boolean element by "measuring"---projecting onto the Boolean sublattice of \(L_Q\) (the center \(Z(L_Q)\)). An element \(a \in L_Q\) belongs to \(Z(L_Q)\) iff \(a\) distributes with all elements. Elements involved in non-distributive triples (the quantum gap \(\Qgap\)) are outside the center and are collapsed by the projection.

\eqref{eq:ker-QH}: \(Q_H\) forces distributivity by replacing the orthomodular lattice with its "distributive envelope"---the smallest distributive lattice containing \(L_Q\) as a sub-poset. The quotient by the distributivity failures is the kernel. Since \(\Qgap\) consists precisely of the non-distributive triples, \(\ker(Q_H) \supseteq \Qgap\). The kernel may be strictly larger because the distributive envelope may introduce new identifications.

\eqref{eq:ker-HQ}: \(H_Q\) must represent intermediate truth values in an orthomodular lattice. Since orthomodular lattices are complemented (\(a \vee a^{\perp} = 1\) always), elements of \(\Hgap\) (where \(a \vee \neg a \neq 1\)) have no faithful orthomodular representative.

Part 3 (round-trip losses): \eqref{eq:rt1}: \(B_H\) embeds \(L_B\) into \(L_H\) (lossless), then \(H_B\) projects back. But \(H_B\) is a left inverse of \(B_H\) only on the image of \(B_H\). The composition \(H_B \circ B_H = \mathrm{id}_{L_B}\) does hold---the loss appears in the reverse direction. Correction: the round trip \(B_H \circ H_B \neq \mathrm{id}_{L_H}\) is the lossy one (Heyting \(\to\) Boolean \(\to\) Heyting re-embeds a coarser structure).

\eqref{eq:rt2}: Analogous. \(B_Q \circ Q_B \neq \mathrm{id}_{L_Q}\): orthomodular \(\to\) Boolean \(\to\) orthomodular re-embeds a collapsed version.

\eqref{eq:rt3}: \(H_Q \circ Q_H\) must cross both gaps. \(Q_H\) loses non-distributivity; \(H_Q\) loses complementation. The round trip \(Q_H \circ H_Q\) suffers both losses.

Part 4 (triangle composition): \(B_Q\) is lossless. \(Q_H\) has \(\ker \supseteq \Qgap\). \(H_B\) has \(\ker = \Hgap\). By the Kernel Composition Lemma (Filter Paper, Lemma 4.2), \(\ker(H_B \circ Q_H \circ B_Q) \supseteq \ker(Q_H) \cup H_B^{-1}(\ker(H_B))\). Since \(B_Q\) is lossless, the quantum gap propagates; since \(Q_H\) maps into a Heyting algebra, \(H_B\) then introduces the Heyting gap. Hence \(\ker(\text{full triangle}) \supseteq \Qgap \cup \Hgap\).

The round-trip notation should be precise: \(X_Y \circ Y_X\) starts in logic \(X\), translates to \(Y\), and returns to \(X\). The composition \(B_H \circ H_B\) (Heyting \(\to\) Boolean \(\to\) Heyting) is lossy because \(H_B\) destroys information in \(\Hgap\) that \(B_H\) cannot reconstruct. The composition \(H_B \circ B_H\) (Boolean \(\to\) Heyting \(\to\) Boolean) is the identity, since \(B_H\) is an embedding and \(H_B\) is its left inverse. The non-trivial losses are always on the round trips that start in the richer logic.

%═══════════════════════════════════════════════════════════════════════════════

Loss Quantification¶

%═══════════════════════════════════════════════════════════════════════════════

\tB\;

For a translation functor \(F : L_1 \to L_2\), define the translation loss as: \begin{keyeq} [ \mathcal{L}(F) = 1 - \frac{|Z(L_1) \cap \ker(F)^c|}{|L_1|} ] \end{keyeq} where \(Z(L_1)\) is the center of \(L_1\) (elements compatible with all others) and \(\ker(F)^c\) is the complement of the kernel. For infinite lattices, replace cardinalities with appropriate measures. Intuitively: \(\mathcal{L}(F)\) is the fraction of source-logic information destroyed by translation.

\tB\;

The six directed translations satisfy: \begin{keyeq} [ \underbrace{\mathcal{L}(B_H) = \mathcal{L}(B_Q) = 0}{\text{embeddings}} < \underbrace{\mathcal{L}(H_B)} \leq \underbrace{\mathcal{L}(Q_B)}}{\text{Quantum gap}} \leq \underbrace{\mathcal{L}(H_Q), \, \mathcal{L}(Q_H)} ] \end{keyeq} with equalities depending on the specific lattices. In general, cross-gap translations (}\(H_Q, Q_H\)) are the most lossy because the source and target logics fail different structural axioms.

The embeddings are injective, so \(\ker = \emptyset\) and \(\mathcal{L} = 0\). For \(H_B\), the kernel is exactly \(\Hgap\), so \(\mathcal{L}(H_B) = |\Hgap|/|L_H|\). For \(Q_B\), the kernel involves all non-distributive triples, which (in a typical orthomodular lattice) is larger than the complement of the center, giving \(\mathcal{L}(Q_B) \geq \mathcal{L}(H_B)\) when \(|\Qgap| \geq |\Hgap|\). For the cross-gap translations, both the Heyting gap and quantum gap contribute, so \(\mathcal{L}(H_Q) \geq \mathcal{L}(H_B)\) and \(\mathcal{L}(Q_H) \geq \mathcal{L}(Q_B)\).

%═══════════════════════════════════════════════════════════════════════════════

The Cognitive Triangle¶

%═══════════════════════════════════════════════════════════════════════════════

\tA\;

The three logics induce three irreducibly different epistemic modes, each with its own notion of negation: \begin{keyeq}

\end{keyeq} No logic faithfully embeds into any other (except Boolean into both). A cognitive system operating natively in one logic can translate to another only by accepting the losses classified in Theorem ref:thm:cit-triangle.

The irreducibility follows from the structural incompatibility of the defining axioms. Heyting algebras are distributive but not complemented (in general); orthomodular lattices are complemented but not distributive (in general). The only lattices that are both distributive and complemented are Boolean algebras. Hence: [ \text{Boolean} = \text{Heyting} \cap \text{Orthomodular} ] in the space of lattice types, and the three logics form a triangle with Boolean at the intersection. Any non-Boolean Heyting algebra has \(\Hgap \neq \emptyset\); any non-Boolean orthomodular lattice has \(\Qgap \neq \emptyset\). These gaps are the irreducible obstructions to cross-translation.

%═══════════════════════════════════════════════════════════════════════════════

Cognitive Substrate Classification¶

%═══════════════════════════════════════════════════════════════════════════════

\tB\; A cognitive substrate \(S\) has logic type \(\ell(S) \in \{\mathrm{Bool}, \mathrm{Heyt}, \mathrm{OM}\}\) determined by its native proposition structure: [nosep] - \(\ell(S) = \mathrm{Bool}\) if every proposition in \(S\) is decidable and composition distributes (classical digital hardware). - \(\ell(S) = \mathrm{Heyt}\) if propositions admit intermediate truth values but composition is distributive (biological neural networks with graded activation). - \(\ell(S) = \mathrm{OM}\) if propositions may be in superposition and composition is non-distributive (quantum coherent substrates).

\tB\;

When two cognitive agents with substrates \(S_1, S_2\) of different logic types communicate, the minimum translation loss is: [ \mathcal{L}_{\mathrm{comm}}(S_1, S_2) = \mathcal{L}(\ell(S_1) \to \ell(S_2)) + \mathcal{L}(\ell(S_2) \to \ell(S_1)) ] This is strictly positive unless both substrates are Boolean. For human--AI communication (Heyting \(\leftrightarrow\) Boolean), the loss is \(\mathcal{L}(H_B) + \mathcal{L}(B_H) = \mathcal{L}(H_B)\) (the original CIT). For human--quantum communication (Heyting \(\leftrightarrow\) orthomodular), the loss is \(\mathcal{L}(H_Q) + \mathcal{L}(Q_H)\), which is strictly greater than the CIT loss.

\begin{interpretation} The CIT triangle predicts that a future ecosystem with classical AI, biological humans, and quantum AI will face three distinct communication barriers, none reducible to any other. The human--classical barrier is the current CIT (loss of conceptual vagueness). The human--quantum barrier is the CIT plus loss of non-distributive structure. The classical--quantum barrier is purely structural (distributivity), without the conceptual vagueness issue.

This has immediate implications for alignment: aligning a quantum AI is harder than aligning a classical AI, not just because quantum systems are more powerful, but because the translation losses are fundamentally richer.

Under the three-space ontology, each substrate type corresponds to a different dominant space: classical AI operates primarily in \(\PS\) (Boolean projections of \(\QS\)), biological cognition operates at the \(\PS\)/\(\CSp\) interface (Heyting logic from graded instantiation), and quantum AI operates closer to \(\QS\) (orthomodular logic from direct quantum structure). The communication barriers are thus inter-space barriers, and the CIT triangle is a map of the ontological borders between the three co-primordial spaces. \end{interpretation}

%═══════════════════════════════════════════════════════════════════════════════

Connection to the Filter Formalism¶

%═══════════════════════════════════════════════════════════════════════════════

\tA\;

Each translation functor \(F : L_1 \to L_2\) is a cognitive filter \(\Phi_F \in \mathrm{Filt}\) in the sense of the Filter Paper (Definition 3.1): [ \Phi_F : \mathbb{R}^{n(e)}{\geq 0} \to \mathbb{R}^{n(e)}, \quad \Phi_F(\bI) = P_F \cdot \bI ] where \(P_F\) is a diagonal projection matrix with \((P_F)_{\tau\tau} = 1\) if intelligence type \(\tau\) is "compatible" with the target logic and \((P_F)_{\tau\tau} < 1\) otherwise. The kernel of \(\Phi_F\) in the filter sense coincides with the logical kernel: \(\ker(\Phi_F) = \ker(F)\).

Intelligence types requiring propositions in the source-logic's gap (e.g., \(I_{\mathrm{eval}}\) for conceptual vagueness under \(H_B\)) are degraded by the translation. The filter matrix \(P_F\) encodes this degradation. By the Kernel Composition Lemma, composing translation filters yields kernels that are unions of the individual kernels, matching the triangle composition result of Theorem ref:thm:cit-triangle, Part 4.

\tA\; Cross-substrate communication can be modeled as applying a translation filter before and after the standard filter pipeline: [ \bI_{\mathrm{received}} = \Phi_{F^{-1}} \circ \Phi_{\mathrm{pipeline}} \circ \Phi_F (\bI_{\mathrm{sent}}) ] where \(\Phi_F\) translates from the sender's logic to the channel logic and \(\Phi_{F^{-1}}\) translates from the channel to the receiver's logic. The total kernel is \(\ker(\Phi_F) \cup \ker(\Phi_{\mathrm{pipeline}}) \cup \ker(\Phi_{F^{-1}})\).

%═══════════════════════════════════════════════════════════════════════════════

Summary of Results¶

%═══════════════════════════════════════════════════════════════════════════════

\begin{citbox}[Thread 2: New Results] [nosep,leftmargin=1.5em] - Theorem ref:thm:cit-triangle (Tier A): The CIT Triangle --- all six translation functors classified with kernel structure. Four are lossy; two (Boolean embeddings) are lossless. - Theorem ref:thm:three-ignorance (Tier A): Three irreducible epistemic modes (classical, vague, quantum). Boolean = Heyting \(\cap\) Orthomodular. - Proposition ref:prop:loss-ordering (Tier B): Loss ordering: embeddings \(<\) same-base lossy \(\leq\) cross-gap lossy. - Proposition ref:prop:mixed-substrate (Tier B): Communication cost between mixed-logic substrates. - Proposition ref:prop:trans-as-filters (Tier A): Translation functors are filters; logical kernels \(=\) filter kernels. - Definition ref:def:three-gaps: The three gaps (\(\Hgap\), \(\Qgap\), \(\Dgap\)) as structural obstructions. - Definition ref:def:translation-loss: Quantitative translation loss measure. - Upgrades the Part X CIT Triangle conjecture (Tier C) to a proved theorem (Tier A).

\end{citbox}

\ifx\STANDALONE\undefined \else