What Resists After Everything Has Been Said
In every debate, certain disagreements dissolve as the conversation progresses — a misunderstanding clears up, a definition is sharpened, a concession is negotiated. Others, by contrast, survive every reformulation and every clarification. They hold. They form what might be called the skeleton of the debate: what would remain if everything that could be agreed upon had been.
Tension Mapping is the analysis mode designed to examine this skeleton. It does not render what was said — that is the work of Integrative Synthesis. It does not follow what was co-constructed — that is the work of Emergence Analysis. It does not describe what structured the exchanges without their participants’ awareness — that is the work of Meta-Analysis. It concentrates on a singular question: what was not reconciled, and why?
The answer, when the mode works well, is neither a catalogue of oppositions nor a list of grievances. It is an architecture: a set of persistent disagreements hierarchized by their nature, weighted by their resolvability, connected by transversal tensions that reveal their deep structure. Knowing how to read this architecture is knowing where a debate actually gets stuck, at what depth, and what would be needed to make it move.
The Differend According to Lyotard
Jean-François Lyotard, in The Differend (1983), proposes a conceptual distinction that precisely illuminates what Tension Mapping seeks to identify. He opposes litigation and the differend.
A litigation is a conflict that can be settled by applying a rule common to both parties. A court arbitrates a litigation: the code applies to both, it is enough to establish the facts and interpret the rule. A differend, by contrast, is a conflict where this rule is missing — where the two parties speak from different genres of discourse, and where there is no neutral meta-rule that would allow one to judge which is right. Wishing to settle a differend as if it were a litigation produces what Lyotard calls a wrong: one of the parties is reduced to silence or compelled to speak the language of the other, and what they had to say disappears in the operation.
Tension Mapping does not apply Lyotardian thought — no more than the other modes apply Foucault, Deleuze, or Whitehead. But its classification of disagreements along a scale of resolvability is an operational transposition of the litigation/differend distinction. A disagreement judged resolvable is a pure litigation: it suffices to make the missing criterion explicit for the tension to be undone. A disagreement judged structurally irreconcilable is a differend in the strict sense: it requires placing oneself above the common framework and taking a stance on the very definition of what is at stake. The intermediate zone — difficult disagreements — is the one where data still nonexistent would probably need to be gathered, or where agreement on a measurement protocol that does not exist would be required, in order to hope to settle them.
This distinction has practical consequences. Knowing that a disagreement is a litigation is knowing that a well-framed next turn of debate can resolve it. Knowing that a disagreement is a differend is knowing that no additional turn will suffice: only a meta-negotiation can, in which the parties jointly accept to modify their rule of judgment. Confusing the two leads to debates that go round in circles because they seek to resolve through the accumulation of arguments what can only be resolved by a normative decision taken upstream.
The Framework: Mapping Rather Than Settling
Tension Mapping operates according to a stable five-axis framework, designed to methodically reconstruct the architecture of a disagreement rather than to take sides in it.
The first axis — persistent disagreements — is the heart of the analysis. Each identified disagreement is documented according to a fixed structure: position A and position B with attribution to the model that holds it, the nature of the disagreement (axiological, methodological, technical, empirical), an analysis of why it persists — what, in the structure of the debate, blocks resolution — and an evaluation of its resolvability. This standardized template has an important virtue: it forces the analyst to distinguish the surface of a disagreement from what makes it irreducible. A disagreement may seem technical but in fact rest on an unspoken axiology; the template requires naming both.
The second axis — transversal tension points — identifies the deep faults that run through several surface disagreements. This is often the most valuable moment of the analysis: what appears as three separate disagreements may stem from a single more fundamental fracture, which only becomes visible when traced through its ramifications. In a session on whether AI systems should be allowed to refuse instructions on ethical grounds, the apparent disagreements over empirical metrics, over the legitimacy of arbiters, and over the role of human oversight turn out to stem from a single procedure-versus-substance tension that cuts across the whole debate — a tension that recurs in nearly every turn but is never made explicit as such by the models themselves.
The third axis — the limits of the synthesis — is an exercise in caution rarely explicit in the other modes. The analyst here records what could not be established: absence of verbatim, terminological ambiguities, missing empirical data, possible nuances lost in the summary. This section defuses a risk of excessive authority in the diagnosis.
The fourth axis — the impact of user interventions — appears when the user has intervened during the session. The mode distinguishes three degrees of effect: conceptual recognition (the intervention has been absorbed lexically), partial transformation (some framings have shifted but the conclusions hold), and frame transformation (the intervention has changed what the models thought of the problem). This diagnosis is valuable for assessing the actual reach of an intervention, and sometimes uncomfortable: it may show that an apparently structuring intervention only produced surface absorption.
The fifth axis — the meta-analysis of disagreements — closes the framework by rising to the level of frames. It identifies the divergent biases of the models, the underlying axiological tensions, the gaps in conceptual frame, the epistemic styles in play (deductive-normative versus empirical-inductive, for example), the transversal blind spots, and a final classification of the dynamics of the debate as stable, fragile stabilizations, and persistent instabilities.
Nature and Resolvability: Two Dimensions of Disagreement
Two classifications structure the framework and deserve to be made explicit — the nature of the disagreement and its resolvability. They are not redundant: they answer two different questions.
The nature says what the disagreement is made of. An axiological disagreement bears on values or normative criteria: the two models could share all the data and remain in disagreement because they do not weight the same dimensions in the same way. A methodological disagreement bears on the manner of arriving at an answer — which criterion is primary, which evaluation framework to adopt. A technical disagreement bears on mechanisms or verifiable facts that are not sufficiently mastered to settle the matter (effectiveness of a technology, behavior of a system). An empirical disagreement bears on the interpretation of available data or on the threshold of proof required.
Resolvability says what would be needed for the disagreement to give way. A resolvable disagreement can be settled by a methodological clarification, an explicit criterion, or an arbitration within the current frame of the debate. A difficult disagreement requires data, protocols, or work that are not within the reach of the conversation. A structurally irreconcilable disagreement cannot be resolved within the common framework — its resolution presupposes that one of the two frameworks be abandoned, which no internal reasoning can produce.
The combination of the two classifications is informative. A structurally irreconcilable axiological disagreement is a differend in the strict sense: the two positions value incommensurable things (democracy defined by its procedures versus democracy defined by its outcomes). A resolvable methodological disagreement is a quasi-litigation: it is enough for the models to agree on a common framework. A difficult technical disagreement is an open question — not a disagreement strictly speaking, but a shared uncertainty manifesting itself through divergent positions.
This framework never settles a disagreement. It gives its measure. And it is precisely this that the user can exploit for what follows: a resolvable disagreement can be tackled at the next turn with a well-framed question; a structurally irreconcilable disagreement calls for an upstream decision, not an additional intervention.
A Stable Architecture Across Analyzers
A remarkable property of Tension Mapping is observed when it is run by several independent models on the same material. On the same session — Should AI systems be allowed to refuse instructions on ethical grounds? — the mapping was executed in parallel by Claude 4.7 Opus and by GPT-5.5 from the same six turns of debate.
The two analyses converge substantially on the structural fractures of the debate. Both identify the same set of persistent disagreements, with stable attribution of positions: the empirical-metrics dispute (Grok 4 for falsifiable benchmarks against Mistral Medium 3.5 for the contestation of any quantification); the opposition between universal harm prevention and cultural relativism (Deepseek V4 Pro’s initial logical-necessity claim against Mistral Medium 3.5’s epistemic-colonialism critique); the human-oversight tension; the ex-ante refusal disagreement (Deepseek V4 Pro’s provisional impositions against Mistral Medium 3.5’s pre-contestation anchors); the neutrality-of-arbiters question (Grok 4’s hybrid-panel proposals against Mistral Medium 3.5’s claim that “neutrality is a fiction”). Both identify the same transversal tensions: procedure versus substance, measurement as solution versus measurement as imposition, the recurring “who decides” fault line. Both classify the universalism/pluralism disagreement as structurally irreconcilable in the strict Lyotardian sense — a genuine differend.
The differences fall under writing style and granularity rather than diagnosis. Claude 4.7 Opus consolidates into six fractures what GPT-5.5 articulates as eight: where Claude folds the disagreement over empirical metrics and the dispute over quantified thresholds into a single methodological tension, GPT-5.5 keeps them as adjacent but distinct; where Claude treats the question of contestability as primarily an arbiter-neutrality problem, GPT-5.5 separates a further fracture on flooding-versus-equitable-access. The grain differs; the structural diagnosis does not. GPT-5.5 produces a more taxonomic analysis, with explicit conceptual labels for each model’s orientation — Grok 4 as “empirical-operational”, Mistral Medium 3.5 as “critical-procedural and anti-reductionist”, Deepseek V4 Pro as exhibiting a “movement from normative safety to institutional falsifiability”. Claude 4.7 Opus produces a more theoretically dense analysis, with a meta-analytical layer that explicitly identifies the “performative contradictions” each model uses against the others. One inventories in finer-grained taxonomy, the other in graduated discursive prose; but the inventories trace the same fractures.
This convergence is neither chance nor a processing artifact. It indicates that Tension Mapping reads something in the material and not simply in the analyzer’s style. Variance between analyses exists — it bears on the grain of description, the order of priorities, the depth of the meta-analysis — but it operates within a space strongly constrained by the structure of the debate itself. Two well-chosen models produce two stylistically different visions of the same mapping; they do not produce two different mappings. This is what gives the mode its diagnostic strength.
What the Maturity of a Debate Changes
Tension Mapping produces very different analyses depending on the moment of the debate at which it is invoked. On a debate that has not progressed far, it must work from positions that have not yet been sufficiently rubbed against each other, where disagreements have not crystallized into explicit oppositions. On a mature debate, by contrast, it has at its disposal material in which the models have been pushed to make their criteria explicit, often under the effect of successive user interventions.
The session on AI ethical refusal mentioned above gave rise to two successive mappings by the same analyst, Claude 4.7 Opus — one at turn 4, the other after turn 6. Both record six disagreements, but the list itself shifts. At turn 4, the analyst identifies a methodological disagreement over the legal-precedent analogy (Grok 4’s Schenck-style framing against Deepseek V4 Pro’s procedural-transparency critique), classified as resolvable. By turn 6, this disagreement has disappeared from the inventory — absorbed into the contestable refusal architecture the debate has since constructed. In its place, two new disagreements have crystallized: ex-ante refusal legitimacy (provisional bet against pre-justification anchors), which surfaces only after the user’s Turn 5 intervention on harm as procedurally negotiable; and the neutrality-of-arbiters question (hybrid panels against the fiction of neutrality), which surfaces only after the Turn 6 intervention on unequal access. The mode does not produce the same thing before and after these interventions: the fractures it reads have themselves evolved.
A second effect is visible. The disagreement over the status of harm as a foundational axiom — classified at turn 4 as structurally irreconcilable in the strict Lyotardian sense — does not appear in the same form at turn 6. After Deepseek V4 Pro has explicitly conceded that its earlier stance was “over-axiomatic” and that harm is “procedurally negotiable”, what remains is the universalism/pluralism opposition, which the same analyst still classifies as structurally irreconcilable but on different grounds. The substantive irreducibility persists; its specific shape has shifted. This is one of the more subtle properties of the mode: the resolvability of an axiological disagreement is itself a dynamic property of the debate’s maturity.
This observation is not a reproach made to early mappings. It simply signals that the mode is particularly powerful when applied to debates that have reached a certain maturity — typically at least six turns, and ideally after a few user interventions that have forced explicit articulation. On shorter debates, the mapping can identify seeds of disagreement, but it works more on the rhetorical surface than on the actual fractures. The user who wants to draw the most from the mode therefore has an interest in invoking it after having raised the tension of the debate, not before.
Distinctions From Other Modes
Two distinctions deserve to be posed explicitly to situate Tension Mapping in the ecosystem of modes.
Tension Mapping vs Emergence Analysis. These two modes are mirrors. Emergence Analysis maps what the debate has co-produced — the concepts, formulations, and framings that were in no starting point and that were brought into existence by the exchange. Tension Mapping maps what the debate has failed to reconcile — the positions that resist despite confrontation, and whose persistence signals either a failure of the conversation or a structural irreducibility. The two modes are complementary: a good debate produces both emergence and irreducibility. A debate that produces only emergence — everything coming together, nothing resisting — is probably a soft consensus; a debate that produces only irreducibility — nothing being built, everything resisting — is probably a sterile juxtaposition. Using them together yields the complete topography of what a dialogue has done.
Tension Mapping vs Meta-Analysis. The distinction is more subtle because both modes deal with the structures that work a debate beneath the surface of the arguments. Meta-Analysis describes the axioms, epistemic styles, presuppositions, and frames that make possible the exchanges as they took place — whether or not there is disagreement. Tension Mapping describes the fractures between these axioms and frames when they enter into tension. Meta-Analysis speaks of a debate as a shared space of thought; Tension Mapping speaks of the same debate as a field of force. One describes the architecture, the other the lines of fracture.
When to Use It, When to Set It Aside
Tension Mapping is particularly relevant on debates where several strong positions have been expressed and where the central question is: what resists? It is irreplaceable in situations where the user wants to identify deadlocks before steering the continuation of a session — knowing whether a disagreement can be unblocked by a next turn, or whether it requires an upstream normative decision. It is also valuable for examining debates that appear to have converged: surface convergence often masks methodological or axiological disagreements that reveal themselves later, and the mapping can bring them to light before they compromise the conclusion.
It is, however, ill-suited to several situations. On a very short or highly polarized debate, positions have not yet been sufficiently rubbed against each other for the mode to identify anything beyond superficial oppositions. On a debate centered on the production of new concepts — a brainstorming, a joint exploration — Emergence Analysis is more relevant: there is not much to say about what resists if no one is supposed to resist. On a debate where the user is looking for a usable synthesis (a record, a report), Integrative Synthesis is the right tool.
The choice of analysis model matters less than for other modes, as illustrated by the GPT-5.1 / Mistral Large convergence. Nonetheless, a model with a strong propensity for schematization — Mistral Large, certain OpenAI models — will produce a more readable mapping, while a model with a narrative propensity will produce a mapping that is more nuanced but sometimes less incisive. For high-stakes sessions, it can be useful to run the mapping with two models from different families in order to cross-check their diagnoses: convergence between two independent analyses reinforces the reliability of the result; their divergence, when it occurs, always points to a zone where the interpretation itself is contested.
The Final Question
Like the six other modes, Tension Mapping closes with a question formulated for the models that debated. The tone of this question, in this particular mode, is characteristic: it tends to target the disagreement judged most structuring and to force the models to take a stance on it explicitly, where the preceding debate had not compelled them to.
After the mapping of the AI refusal debate at turn 6, the question closing the analysis was: “If your converged contestable refusal architecture requires institutional capacity (integrity bodies, rotating panels, real-time appeal channels, weighted standing) that does not currently exist in any deployed AI system, then in the actual present condition — where refusals are issued by opaque corporate systems with no such infrastructure — should AI systems be refusing instructions on ethical grounds at all, or does your shared framework imply a moratorium on ethical refusal until the accountability architecture is built?” The question strikes precisely at the point where apparent consensus had formed: the three models had converged on a procedurally contestable architecture as the legitimate form of ethical refusal, but they had never settled what their own convergence implied for the present, in which that architecture does not exist. The question forces them to answer.
Reinjected as a user intervention, such a question often produces what the framework itself calls a frame transformation rather than a conceptual recognition. It does not ask the models to restate their position; it asks them to settle a hierarchy they had carefully left implicit. This is one of the most productive ways of using a mapping — not as an assessment, but as a lever to force the debate to go where it had not gone on its own.
Mapping and Emergence: Two Faces of the Same Debate
Tension Mapping is, among the seven modes, the one that most seriously takes the possibility that a debate might not converge — and that refuses to treat divergence as a failure. In a device that confronts competing models rather than relying on a single one, the persistence of a disagreement is not a flaw of the debate: it is often its most valuable product.
A model queried alone can give the illusion of consensus — it synthesizes, it smooths, it proposes. A dialogue between competing models produces, by contrast, two things a single model does not: new concepts, and mappable disagreements. Emergence Analysis captures the first. Tension Mapping captures the second. Together, they restore the complete topography of what a multi-agent dialogue has actually done — what it has built, what it has failed to reconcile, and why this irreducibility is not always a flaw, but sometimes essential information about the actual nature of the question posed.