---
title: Strict-Tolerant Logic in Non-Classical Reasoning
url: https://www.emergentmind.com/topics/strict-tolerant-logic
type: topic
---

# Strict-Tolerant Logic in Non-Classical Reasoning

Searching arXiv for recent and foundational papers on strict-tolerant logic and related metainferential work.
Strict-Tolerant Logic is a non-classical framework in which validity is defined by a mixed preservation condition: from the truth of the premises must follow the non-falsity of the conclusion. In its standard formulation, this modifies the ordinary notion of logical consequence without abandoning classical validity at the level of extension. The framework has been developed in connection with vagueness, tolerance, and naive truth, and it has also been extended to intuitionistic and minimal settings, where its behavior is markedly sensitive to the underlying semantics. Recent work shows that strict-tolerant inference over intuitionistic semantics collapses to ordinary classical consequence, while the corresponding minimal version becomes inferentially trivial; at the metainferential level, however, classical, intuitionistic, and minimal strict-tolerant systems separate from one another [2509.10322].

## 1. Core notion of strict-tolerant consequence

Let \(L\) be a propositional language, and let interpretations \(\mathcal I\) be equipped with a notion of truth \(\mathcal I \vDash A\) and falsity \(\mathcal I \vDash \neg A\). The strict-tolerant consequence relation is defined to capture the condition that “from the truth of all premises follows at least the non-falsity of the conclusion” [2509.10322]. For a set of premises \(\Gamma \subseteq L\) and a set of conclusions \(\Delta=\{A_1,\dots,A_n\}\), one has
\[
\Gamma \Rightarrow_{ST} \Delta
\]
iff for every interpretation \(\mathcal I\), if \(\mathcal I \vDash B\) for every \(B\in\Gamma\), then \(\mathcal I \not\vDash \neg A_i\) for some \(i\in\{1,\dots,n\}\) [2509.10322]. In the single-succedent case,
\[
\Gamma \vdash_{ST} A
\quad\Longleftrightarrow\quad
\forall \mathcal I \Bigl[
(\forall B\in\Gamma)\,\mathcal I\vDash B
\;\Longrightarrow\;
\mathcal I\not\vDash\neg A
\Bigr].
\]

A closely related three-valued presentation distinguishes strict consequence, tolerant consequence, and their mixed form. In that setting, a valuation \(v\) takes values in \(V=\{0,\tfrac12,1\}\), where \(1\) means “strictly true,” \(\tfrac12\) means “borderline,” and \(0\) means “strictly false” [2207.12786]. The mixed strict-tolerant relation is then
\[
\Gamma\vdash_{ST}\varphi
\quad\text{iff}\quad
\nexists v\;[\,(\forall\gamma\in\Gamma\,v(\gamma)=1)\ \wedge\ (v(\varphi)=0)\,].
\]
This formulation makes explicit the mixed character of the logic: premises are evaluated strictly, conclusions tolerantly [2207.12786].

In first-order multiple-conclusion form, a sequent \(\Gamma\Rightarrow\Delta\) is ST-satisfied by a model just in case either some premise fails to be strictly true or some conclusion is tolerantly true [2602.23568]. The resulting notion preserves all classically valid laws, a fact expressed by the equivalence
\[
\models_{\text{ST}}\Gamma\Rightarrow\Delta
\;\Longleftrightarrow\;
\models_{\text{CL}}\Gamma\Rightarrow\Delta
\]
for multiple-conclusion consequence [2602.23568]. This classical coincidence at the level of extension is one of the central structural features of the framework.

## 2. Semantics and the strict/tolerant distinction

The best-known semantic presentation of Strict-Tolerant Logic uses Strong Kleene truth-functions on the three-valued set \(V=\{0,\tfrac12,1\}\) [2207.12786]. For any formulas \(A,B\),
\[
v(\neg A)=1-v(A),\qquad
v(A\wedge B)=\min\{v(A),v(B)\},\qquad
v(A\vee B)=\max\{v(A),v(B)\},
\]
and
\[
v(A\to B)=\max\{1-v(A),v(B)\}.
\]
Quantifiers are interpreted by infimum and supremum over variants of a valuation [2207.12786]. The same Strong Kleene basis is used in first-order ST-models, where predicates are interpreted as maps into \(\{0,\tfrac12,1\}\) and functions in the ordinary way [2602.23568].

Within this semantics, strict and tolerant statuses are sharply separated. A sentence is strictly true iff it receives value \(1\), tolerantly true iff it receives either \(1\) or \(\tfrac12\), strictly false iff it receives \(0\), and tolerantly false iff it receives \(0\) or \(\tfrac12\) [2602.23568]. That distinction yields three corresponding consequence relations. Strict consequence requires strict truth of premises and strict truth of conclusion; tolerant consequence requires at least borderline truth of premises and conclusion; strict-tolerant consequence combines strict premises with tolerant conclusion [2207.12786].

A notable consequence is that the three relations differ in how they treat borderline cases while coinciding extensionally with ordinary first-order classical entailment [2207.12786]. This does not mean that the framework is merely notationally classical. Rather, what differs is which semantic configurations count as countermodels when premises or conclusions take the borderline value \(\tfrac12\). A plausible implication is that ST is best understood not as a rival set of theorems but as a refined account of consequence under mixed standards of premise acceptance and conclusion toleration.

## 3. Tolerance, vagueness, and object-language principles

Strict-Tolerant Logic has been developed as a logical framework for vagueness, especially for capturing tolerance without giving up classical validities. In the three-valued setting with a vague predicate \(P\) and a similarity relation \(\sim_P\), one may state the tolerance principle directly in the object language as
\[
\forall x\forall y\bigl(P(x)\wedge (x\sim_P y)\to P(y)\bigr)
\]
[2207.12786]. A proof-theoretic counterpart is the metainference rule
\[
\inferrule*[left=Tol]
{\,\vdash_{ST}\;P(t)\quad\quad \vdash_{ST}\;t\sim_{P}u}
{\,\vdash_{ST}\;P(u)}
\]
which is sound in ST-models satisfying the relevant similarity constraint [2207.12786].

The significance of this construction lies in the claim that continuum many degrees are not required to account for tolerance phenomena. “Tolerance and degrees of truth” argues that both Smith’s fuzzy approach and the strict-tolerant approach can be subsumed under a common three-valued framework, and that three values suffice to satisfy Smith’s central desiderata, internalize the closeness principle as an object-language tolerance principle, and reproduce classical entailment in the extended language [2207.12786]. In that account, the unique proper symmetric open parameter with only three values is
\[
(V,T,F)=\bigl\{0,\tfrac12,1\bigr\},\ \{1\},\ \{0\},
\]
and its associated consequence is exactly ST-logic [2207.12786].

This line of work places ST at the intersection of non-classical semantics and conservative inferential behavior. The framework validates a tolerance principle while preserving a fully classical core of inference [2207.12786]. This suggests that its primary role is not to revise theoremhood but to alter the semantic standards under which inferential legitimacy is assessed in the presence of borderline cases.

## 4. Extensions to intuitionistic and minimal logic

A major recent development is the systematic application of the strict-tolerant approach to intuitionistic and minimal logic [2509.10322]. In the Kripke-style setting used there, falsity of \(A\) is defined strongly by
\[
\mathcal I\vDash\neg A
\quad\Longleftrightarrow\quad
A\text{ is false in }\mathcal I,
\]
and in the intuitionistic setting one always has \(\mathcal I\vDash\bot\) never holds [2509.10322].

For intuitionistic semantics, strict-tolerant consequence is defined by
\[
\Gamma\vdash^{i}_{ST}A
\quad\iff\quad
\Gamma\Rightarrow_{ST}A
\text{ holds in every intuitionistic interpretation}
\]
[2509.10322]. The key result is the collapse theorem:
\[
\Gamma\vdash^{i}_{ST}A
\quad\Longleftrightarrow\quad
\Gamma\models_c A
\]
for all \(\Gamma\cup\{A\}\subseteq L\) [2509.10322]. The proof uses a generalized Glivenko theorem: if \(\Gamma\models_c A\), then \(\Gamma\models_i\neg\neg A\), and from \(\neg\neg A\) in any intuitionistic model making all of \(\Gamma\) true, it follows that \(A\) is not false, exactly as required by the ST condition [2509.10322]. The stated corollary is that there is no novel “intuitionistic strict-tolerant” logic at the inferential level: all strict-tolerant intuitionistic inferences are already classical [2509.10322].

For minimal logic, the situation is more extreme. Minimal strict-tolerant consequence is defined by
\[
\Gamma\vdash^{m}_{ST}A
\quad\iff\quad
\Gamma\Rightarrow_{ST}A
\text{ holds in every minimal interpretation}
\]
[2509.10322]. The corresponding theorem states that no inferences are valid:
\[
\Gamma\;\cancel{\vdash^{m}_{ST}\,A
\]
for every \(\Gamma\cup\{A\}\subseteq L\) [2509.10322]. The argument uses a trivial minimal model with a single world \(w\) and \(v_w(B)=1\) for every formula \(B\), including \(\bot\); in that model, every premise is true but every conclusion is also “falsely false,” so no non-vacuous strict-tolerant inference survives [2509.10322].

A further example clarifies why the intuitionistic collapse does not extend to minimal logic. The classical tautology \(\neg a\to(a\to b)\) satisfies \(\models_c\) but fails \(\nvDash_m\neg\neg(\neg a\to(a\to b))\) in a suitable two-world minimal model, exhibiting the non-Glivenko behavior of minimal logic [2509.10322]. The contrast with intuitionistic logic is thus not merely technical; it is tied to the availability of the double-negation bridge used in the intuitionistic collapse proof.

## 5. Metainferential strict-tolerant logics

Strict-tolerant methods can be lifted from ordinary inferences to metainferences, where premises and conclusion are themselves inferences [2509.10322]. If
\[
\Theta = (\Gamma_1\Rightarrow A_1),\dots,(\Gamma_n\Rightarrow A_n)
\quad,\quad
S=(\Gamma_{n+1}\Rightarrow A_{n+1}),
\]
then the strict-tolerant metainference relation \(\Rightarrow^*\) holds iff for every interpretation \(\mathcal I\), at least one of the following occurs: for some premise-inference, \(\mathcal I\vDash\Gamma_k\) and \(\mathcal I\vDash\neg A_k\); or the succedent’s premises fail; or the succedent’s conclusion is not false [2509.10322]. The corresponding semantic notions are
\[
\Theta\vDash^{i}_{STM}S,\qquad
\Theta\vDash^{m}_{STM}S,\qquad
\Theta\vDash^{c}_{STM}S,
\]
according as the condition is evaluated over intuitionistic, minimal, or classical models [2509.10322].

At this metainferential level, the systems do not collapse together. Although \(\vdash^{i}_{ST}=\models_c\) and \(\vdash^{m}_{ST}\) is trivial, the resulting metainferential logics are distinct [2509.10322]. Two separation results are central.

First, the rule of conjunction introduction,
\[
(\;\Rightarrow A,\;\Rightarrow B\;)
\;\Rightarrow^*\;
(\;\Rightarrow A\land B\;)
\]
is valid in \(\vDash^c_{STM}\) but fails in \(\vDash^i_{STM}\), and hence also in \(\vDash^m_{STM}\) [2509.10322]. The countermodel is a Kripke frame with root \(w\) and two immediate successors \(w'\), \(w''\), where \(A\) holds but \(B\) does not at \(w'\), and \(B\) holds but \(A\) does not at \(w''\). Then each of \(\Rightarrow A\) and \(\Rightarrow B\) is strictly-tolerant valid, but \(A\land B\) fails at every world, so the metainference is invalid [2509.10322].

Second, the “explosion” metainference
\[
(\;\Rightarrow A,\;\Rightarrow\neg A\;)\Rightarrow^*(\;\Rightarrow B\;)
\]
is valid in \(\vDash^i_{STM}\) but fails in \(\vDash^m_{STM}\) [2509.10322]. In the intuitionistic setting, if both \(A\) and \(\neg A\) are not-false, falsity of \(\bot\) follows and hence any \(B\) is not-false; in minimal logic, a three-world model can separate branches for \(A\) and \(\neg A\) while preventing \(\bot\) from forcing falsity uniformly, so \(B\) may remain false [2509.10322].

These results show that metainferential strict-tolerant logic preserves distinctions that are obliterated at the inferential level. A plausible implication is that the strict-tolerant transformation interacts more delicately with rule validity than with ordinary consequence, especially in constructive and paraconsistent settings.

## 6. Proof theory and first-order calculi

The proof theory of first-order ST exhibits a non-trivial divergence between ordinary derivability in cut-free classical sequent calculi and local metainferential ST-validity. The classical sequent calculus without Cut is only partially aligned with the latter: the relations coincide only upon the addition of elimination rules and only within the propositional fragment, due to the non-invertibility of the quantifier rules [2602.23568]. In the first-order case, quantifier introduction rules such as
\[
\infer[\forall\text{L}]{\forall x\,\varphi(x),\Gamma\Rightarrow\Delta}{\varphi[x\mapsto t],\,\Gamma\Rightarrow\Delta}
\qquad
\infer[\exists\text{R}]{\Gamma\Rightarrow\Delta,\exists x\,\varphi(x)}{\Gamma\Rightarrow\Delta,\varphi[x\mapsto t]}
\]
are not invertible, so one cannot safely add quantifier eliminations in the same way [2602.23568].

Two calculi have been proposed to recover full correspondence with local metainferential ST-validity in the first-order case [2602.23568].

| Calculus | Main device | Reported properties |
|---|---|---|
| \(\mathcal{ST}^H\) | Henkin expansion with witness rules and invertible quantifier rules | soundness, completeness, Cut admissible |
| \(\mathcal{MQST}\) | discharge of sequent-assumptions | soundness, completeness, normalisable, interpolation |

The first system, \(\mathcal{ST}^H\), expands the language by adding Henkin constants for universal and existential formulas and uses witness-introduction, witness-elimination, and invertible quantifier rules without eigenvariable conditions [2602.23568]. Completeness is proved by a canonical model construction using prime, \(S\)-consistent Henkin theories, and the system is cut-free with admissible Cut [2602.23568].

The second system, \(\mathcal{MQST}\), achieves invertibility by allowing rules that discharge entire sequent assumptions [2602.23568]. It operates with multiset sequents, includes generalized identity and contraction, and provides quantifier introduction, elimination, and discharge rules [2602.23568]. The calculus is normalisable: every proof can be transformed into a normal one, and it also admits interpolation, with finite interpolant sets restricted to the relation-symbols and free variables common to the relevant premises and conclusion [2602.23568].

The broader significance of these results is that every local metainferential ST-valid inference among first-order sequents can be derived without postulating a non-analytic Cut rule [2602.23568]. This places ST within a technically robust Gentzen-style environment while preserving the mixed semantic character that motivates the logic.

## 7. Related uses, scope, and misconceptions

Strict-Tolerant Logic is associated in the literature with naive theories of truth and vagueness, respectively including a fully disquotational truth predicate and an unrestricted tolerance principle, without jettisoning any classically valid laws [2602.23568]. In the first-order setting, one can add a unary truth predicate \(\mathit{T}(·)\) together with biconditional axioms recovering \(T(\ulcorner\varphi\urcorner)\leftrightarrow\varphi\) in ST, and likewise add unrestricted tolerance principles for vague predicates; none of these extensions invalidates any classical sequent, because ST-validity coincides with classical validity [2602.23568].

A common misconception is that ST introduces new valid first-order theorems in virtue of its three-valued semantics. The cited work states instead that strict, tolerant, and strict-tolerant consequence all coincide extensionally with ordinary first-order classical entailment [2207.12786]. What changes is the treatment of borderline semantic values and, correspondingly, the space of acceptable countermodels. Another misconception is that constructive bases automatically yield new strict-tolerant object logics. Recent results show that this is false for intuitionistic logic, where the strict-tolerant relation collapses exactly to classical consequence, and also false in a different way for minimal logic, where no object-level inferences are valid at all [2509.10322].

The phrase “strict-tolerant” also occurs in an unrelated engineering context, namely fault-tolerant coding in 3-bit Hamming space [0903.4046]. There it refers to a coding scheme with poles \(000\) and \(111\), nearest-pole decoding, and tolerant Boolean operators that auto-correct any single-bit fault on the next gate [0903.4046]. This usage is terminologically similar but conceptually distinct from strict-tolerant logical consequence. The logical literature concerns mixed truth-preservation and non-falsity-preservation; the coding literature concerns error correction in Boolean circuits.

Taken together, the cited work presents Strict-Tolerant Logic as a framework with a classical inferential profile, non-classical semantic machinery, and substantial metainferential structure. In vagueness, it provides a formal setting for tolerance without continuum-valued commitments [2207.12786]. In constructive settings, it exhibits collapse and triviality phenomena at the object level but genuine differentiation at the metalevel [2509.10322]. In proof theory, it supports cut-free first-order calculi aligned with local metainferential validity [2602.23568]. These features define the contemporary research landscape of the subject.

Source: https://www.emergentmind.com/topics/strict-tolerant-logic