Recounting What Was Said
Among Metamorfon’s seven analysis modes, Integrative Synthesis is at once the most immediately useful and the least spectacular. The most immediately useful, because anyone who has just launched a nine-turn debate between two or three models has, first of all, a simple need: knowing what was said in it. The least spectacular, because synthesizing content does not appear to constitute a distinctive cognitive operation — it is, one might think, what every model does all day long.
This apparent self-evidence in fact masks the difficulty proper to the mode. Synthesizing a debate is not summarising a text. A text offers a given organization — an author, a progression, a thesis — and the summary follows what the author has set out. A debate, by contrast, offers several voices, sometimes non-synchronous, that take up, reformulate, contest, and mutually appropriate each other’s arguments, and whose argumentative trajectory is not linear. What a model said at turn 3 may be reformulated at turn 5, refuted at turn 6, and reformulated differently at turn 8 after a user intervention. Restoring this movement — who said what, when, in response to what, with what effect — requires an operation that is not simple compression: it is a narrative reconstruction of the debate’s conceptual pathway.
This operation has a name in narrative theory: it is called emplotment.
Emplotment According to Ricœur
In Time and Narrative (1983–1985), Paul Ricœur proposes a concept that precisely illuminates what an Integrative Synthesis does to a debate. He calls emplotment — taking up and transforming the Aristotelian muthos — the operation by which a succession of heterogeneous events is gathered into an intelligible story.
Emplotment, in Ricœur, does not add a narrative to facts already there. It does two things simultaneously: it selects among events those that count for the trajectory it traces, and it configures these events by linking them through an intelligible necessity. A chronicler can list what happened; a narrator turns it into a story. Emplotment transforms the manifold into the followed — a sequence of things into a progression that makes sense.
This operation is more than editorial convenience. It responds to what Ricœur calls the aporias of lived time: pure time, in its flow, resists understanding; only emplotted time becomes thinkable. In a debate, it is the same problem: the raw sequence of turns is full of repetitions, reformulations, and tacit echoes that resist immediate legibility. Without emplotment, the reader who returns to a long debate loses the thread — every turn seems equal to every other, without one perceiving why a given position was refuted, why a given concession was decisive, why an intervention reoriented the trajectory.
Integrative Synthesis does not apply Ricœur — no more than the other modes apply Foucault, Deleuze, Whitehead, Simondon, or Lyotard. But its fundamental gesture — recomposing a succession of turns into an intelligible argumentative trajectory — is emplotment in the strict sense. It invents nothing that was not said; it reconfigures what was said into a narrative in which each contribution finds its place in a progression.
What the Synthesis Does — and What It Does Not Do
A well-constructed Integrative Synthesis renders four interlocking layers of a debate, in an order that is not negotiable.
It first renders the initial framing — the positions posed by each model in the opening turn, with their accents, their organizing categories, their visible presuppositions. This is the raw material of the narration that will follow, and a poorly reconstituted initial framing skews everything that comes after.
It then renders the argumentative trajectory — who responded to whom, what objections were raised, which positions were reformulated, what concessions were made. This is the heart of the emplotment: what transforms a succession of turns into a coherent movement. In a recent session on the extraction of Claude’s capabilities by several laboratories, a synthesis identifies a precise pivot — Claude acknowledges that its economic argument “marginal cost vs fixed cost” is self-refuting and withdraws it, which forces the debate to reconfigure itself around a pragmatic criterion of performative expectations. Without this identification, the rest of the debate appears arbitrary; with it, it becomes the necessary consequence of an argumentative withdrawal.
This rendering is not a mere sequential listing. It identifies the pivots of the debate by name — the moments where the orientation, the tone, or the framework shifted — by qualifying them according to a precise typology: concession (a model accepts a point it was contesting), reversal (a model modifies its position), frame shift (the conceptual reference moves without the models thematizing it), structuring user intervention (a question or relaunch that reorients the debate). The argumentative withdrawal evoked above is thus classified as a concession; another session, devoted to the very notion of intelligence in AI, sees its trajectory pivot on a structuring user intervention that deliberately tightens the technical framing, then on a frame shift not thematized by the models themselves. For each pivot, the synthesis indicates not only that it took place, but why it functioned as such — what it made possible, what it displaced, what the following turns made of it. It is this requirement of causal analysis that turns the synthesis into analytical work rather than a report. And this is a discipline that extends to its negative side: when a debate contains no real pivot, the synthesis indicates this explicitly rather than fabricating one to fill the rubric.
It renders, third, the conceptual shifts and evolutions — the moments where a position has shifted, where a concept has been enriched, where one framework has been replaced by another. It is here that the Synthesis intersects — without duplicating — Emergence Analysis: it mentions the co-constructed concepts (the differential friction, the scale neutrality, the irreducible residue) because they are part of what was said, but it does not specifically map them with the seven axes of the emergence mode. The Synthesis integrates them into the narrative without making them the object of an autonomous examination.
It finally renders the points of arrival — what the models converge on at the end of the debate, what remains in disagreement, what remains undetermined for lack of information. This section intersects with Tension Mapping but without its analytical depth: it does not classify disagreements according to their nature and resolvability, it signals them as part of the narrative balance.
What Integrative Synthesis does not do, however, is just as important as what it does. It does not judge the argumentative quality of the positions — an argument may be ill-founded and yet faithfully reported as a moment of the debate. It does not deconstruct implicit framings — that is the work of Meta-Analysis and Critical Archaeology. It does not project towards what the debate could have produced without managing to — that is the work of Horizon of Possibilities. It holds itself in the register of faithful rendering, accepting that this fidelity passes through an intelligible reconstruction, not through an integral transcription.
A Stable Structure, Three Styles of Execution
Integrative Synthesis operates according to a stable structure: principal narration of the debate, then complementary meta-analysis, then a question addressed to the models. This structure is the same for all analysts — but the grain and the register of execution vary appreciably depending on the model that produces it. A recent session juxtaposed three independent syntheses of the same debate, executed by three analysts from different families. The result is instructive.
GPT-5.1 produces a long, fluid narration in continuous paragraphs that closely follow the succession of turns. Each significant turn is reconstituted with its principal arguments, and the argumentative progression is rendered through prose that connects the logical pivots. It is a synthesis narrative in the strict sense — the reader following it follows the debate turn by turn, but in a condensed version made intelligible by selection. The register is analytical, with formulations such as “Claude explicitly acknowledges that the argument is self-refuting and withdraws it”, which identify key moments without dramatising them.
Gemini 3.1 Pro Preview adopts a more literary style. The narration is more condensed, but seeks memorable formulations — “a rare and constructive dynamic of concession”, “to recognize this residue is to avoid the authoritarian temptation of the total solution”. This style has a cost and a virtue: it loses some factual granularity, but it helps the reader retain the pivots of the debate as qualified moments, not merely as logical sequences. Gemini’s synthesis resembles a good review article: it does not render everything, but what it renders impresses itself on memory.
Mistral Large adopts a third, structured-tabular style. The continuous narration gives way to numbered sections, sub-sections, and typed bullets (“persistent disagreement”, “implicit agreement”, “point of friction”). The debate becomes a working document where each tension is labelled and situated within an architecture. This style loses in narrative fluidity what it gains in navigability: for a user seeking to quickly locate a precise point of the debate, the tabular synthesis is more efficient than the continuous narration.
Three styles, the same structure, the same substantial content: in all three cases, the same pivots are identified (withdrawal of the economic argument, introduction of the pragmatic criterion, co-construction of differential friction), the same user interventions are signalled as structuring, the same residual disagreements are inventoried. The variance bears on the grain, the register, the distribution of narrative weight — not on the diagnosis.
The Choice of Model: What Each Style Does Best
This stylistic variance is not neutral. It has practical consequences for the user.
The continuous narrative style — GPT-5.1, certain Anthropic models — is optimal when one wants to render a debate to someone who did not follow it. The prose reconstitutes the argumentative thread with enough density that an external reader can understand what played out without returning to the original turns. This is the style to favor for a report to a third party, for an archive intended to be re-read, for a rendering to a collaborator.
The condensed literary style — Gemini 3.1 Pro Preview — is optimal when one wants the key moments of the debate to imprint themselves as memorable formulations. For a debate that serves as the basis for subsequent reflection, for a rendering to an audience that will not return to the raw text, for a session one wants to be able to cite, this style has a mnemonic advantage. It sacrifices a little granularity for the sake of memorability.
The structured tabular style — Mistral Large, certain OpenAI models — is optimal when one wants to work on the debate rather than re-read it. Quickly identifying a specific point, comparing two positions on the same sub-issue, locating user interventions and their effects: the explicit structure facilitates all these operations. This is the style to favor for an operational use of the synthesis — reinjection of elements into a subsequent session, collaborative work, preparation of an article or a structured report.
This choice is not purely aesthetic: it determines what the synthesis makes possible as subsequent use. A user with long, recurring sessions will have an interest in experimenting with all three styles on the same material to identify the one best suited to their way of working.
A Particular Condition: The Reliability of the Rendering
Among the seven analysis modes, Integrative Synthesis is the one in which fidelity to the material weighs the most heavily. The other modes produce analyses whose value rests on the framework applied — a Meta-Analysis can illuminate a debate even if it takes some liberties with certain precise formulations. A synthesis, by contrast, has reliability as its first requirement: if it distorts what was said, it is no longer a synthesis, it is a reformulation.
This requirement has implications for the choice of analysis model. Smaller models can produce acceptable syntheses on short sessions, but they tend to lose track of the voices on long sessions — an argument formulated by model A at turn 3 may be attributed to model B at turn 8, not through malice but through confusion of references. This attribution error is particularly serious in an Integrative Synthesis, because it breaks precisely what the synthesis is supposed to produce: a narrative in which each contribution is situated within a coherent trajectory.
For sessions of more than six turns, it is therefore justified to assign Integrative Synthesis a model powerful enough to maintain attribution reliability across the entirety of the debate. This is not a question of stylistic quality — a modest model can write fluently — it is a question of tracking capacity across a long context. Reliability, for this mode, takes precedence over every other criterion.
Distinctions From Other Modes
Two distinctions deserve to be posed explicitly to situate Integrative Synthesis among the other modes.
Integrative Synthesis vs Emergence Analysis. The two modes deal with the content of the debate, but at different levels. The Synthesis renders the totality of the debate in a coherent narration, including what was already there at the beginning and what emerged along the way. Emergence Analysis, by contrast, isolates a specific layer: only what appeared within the debate without pre-existing in it. In the session on the distillation of capabilities, the Synthesis will render the initial economic argument and its withdrawal and the pragmatic criterion that replaces it; Emergence Analysis will retain only the new concepts — differential friction, scale neutrality, irreducible residue. The Synthesis describes the entire material; the Analysis extracts the emergent share.
Integrative Synthesis vs Tension Mapping. The Synthesis mentions the disagreements that subsist, as part of the debate’s balance. Tension Mapping analyses them — with standardized templates, classification by nature and resolvability, transversal tensions, meta-analysis of divergent biases. The Synthesis says that such-and-such disagreement remains; Tension Mapping says why it persists, of what nature it is, and what would be needed for it to give way. On a debate where disagreements are the principal stake, Tension Mapping is more powerful. On a debate where one is seeking an overall view including everything — convergences, disagreements, shifts, concessions — the Synthesis is better suited.
An ambiguity deserves to be dispelled: Integrative Synthesis is not a generalist mode that would render the others superfluous. It is a narrative mode that has its own purpose (rendering the debate in a readable form) and of which the other modes are the analytical complements — each other mode isolates and deepens a dimension that the Synthesis only touches upon.
When to Use It, When to Set It Aside
Integrative Synthesis is the mode to favor in several specific situations.
First, when one wants to make a debate accessible to someone who did not follow it — a collaborator, a future reader, oneself in several months. It is irreplaceable for this function of intelligent archiving.
Second, when one wants to obtain an initial overview of a long debate before deciding which other mode to apply to it. Running a Synthesis allows one to identify the zones that would deserve a more thorough examination (strong emergence? structurally irreconcilable disagreements? implicit framing to deconstruct?), and thus to choose the most relevant complementary analysis.
Finally, when one wants to prepare a written rendering — article, report, working document — that will rest on the debate without reproducing it integrally. The Synthesis provides a condensed but faithful material, which one can cite, reformulate, extract.
It is, however, ill-suited to other situations. On a very short debate (two or three turns), a synthesis has almost nothing to synthesise — better to read the turns directly. On a debate whose principal stake is a precise dimension (the disagreements, the emergent concepts, the implicit framings), the corresponding specialised mode provides a more useful analysis. On a debate that the user wants to exploit as a lever for a relaunch — that is, where they are seeking a question that will move the positions — the other modes produce sharper final questions: the Synthesis’s question tends to be more general, more open, less surgical.
The Final Question: Extending the Narrative
Like the six other modes, Integrative Synthesis closes with a question addressed to the models that debated. The tone of this question, in this particular mode, is characteristic: it does not aim either to force a concession on a structuring disagreement (as in Tension Mapping), nor to interrogate what has emerged at the price of a narrowing (as in Emergence Analysis), nor to open a virtual horizon (as in Horizon of Possibilities). It tends rather to extend the argumentative trajectory of the debate as it actually unfolded — to pose the question that would be the natural continuation of the debate if it continued beyond its last turn.
In the session on the distillation of capabilities, Mistral Large’s synthesis closed with the question: “If we accept that the proposed mechanisms only partially resolve the problem of transjurisdictional extraction, what complementary strategy — technical, legal, or structural — would make it possible to reduce the very incentive to extract, without resorting to coercive measures that would risk exacerbating geopolitical tensions?” The question extends precisely where the debate had arrived — the acceptance of an irreducible residue, the mention of an opening of the models as a possible path — and asks the models to continue from this precise point. It is not a question that displaces the debate; it is a question that pursues it.
Reinjected as a user intervention, this question typically produces a continuation rather than a frame transformation. This is valuable in certain contexts — notably when the debate has found a fruitful trajectory one wishes to extend — and less useful in others — when one is seeking to swing the debate towards a different framing, in which case the final questions of the other modes are better calibrated.
A Necessary Discretion
Among the seven modes, Integrative Synthesis is the least demonstrative. It does not identify new concepts as Emergence Analysis does, does not deconstruct implicit framings as Meta-Analysis does, does not trace back to historical conditions as Critical Archaeology does. Its ambition is more humble: to faithfully recount what happened, in a form coherent enough to be returned to.
This humility is precisely its value. The six other Metamorfon analysis modes take on their full meaning once the debate has been rendered in an intelligible form — a Tension Mapping on a poorly synthesised debate floats without anchoring; a Meta-Analysis on unrendered material works in a void. Integrative Synthesis is therefore, more than one mode among seven, the narrative foundation on which the other modes can operate. A user seeking to explore a debate in depth often has an interest in beginning with a Synthesis — not because it would deliver the truth of the debate, but because it makes it shareable, navigable, taken up.
This is what emplotment, to take up Ricœur, does to raw experience: it does not replace it, it makes it intelligible to those who were not there — and, often, to those who were.