In brief — why Durand?
Gilbert Durand (1921–2012) showed that human imagination is not a chaos — it organizes itself into three major regimes, three ways of responding to the anguish of time and death. These three regimes — diurnal, mystical, and synthetic — run through all cultures, all myths, all eras. They are not arbitrary categories: they emerge from the very structure of the human imaginary.
PRISME has operationalized these three regimes to apply them to human-AI dialogue. Each connotative deviation in the corpus (2,733 for Boris, 339 for WildChat) is classified on a Durandian barycenter: what proportion of diurnal, mystical, synthetic? And in which direction is the dialogue moving?
The key result (April 16, 2026): the irreducibles are massively synthetic (53% vs. 31%). And above all: the v2f model (logistic regression, 6 predictors) confirms that Durand's regimes are significant predictors of S5 emergence. The strong synthetic regime produces an OR = 4.24 (p < 10⁻¹³) and the strong diurnal regime an OR = 1.73 (p < 10⁻³). Durand is no longer merely an interpretive framework — it is a measured quantitative predictor.
Contents
IThe three regimes of the imaginary
In 1960, Gilbert Durand published Les structures anthropologiques de l'imaginaire (Paris, Bordas, 506 pages). In it, he demonstrates that human imagination — myths, dreams, symbols, narratives — is not a disordered collection of images. It organizes itself around three major postures facing the anguish of time and death. Each posture produces a regime of images, symbols, and narratives.
The diurnal regime — separating, fighting, ascending
The founding gesture: distinguishing. Separating light from darkness, good from evil, pure from impure. The hero fights the monster. The sword cuts. Verticality dominates: ascending, rising, dominating.
The symbols: the sword, the wing, light, the sun, the summit, the piercing gaze. The eagle. The warrior. The judge.
The logic: antithesis. A is not B. The true is opposed to the false. Meaning arises from separation. This is the logic of the Cartesian cogito, of analytical science, of the courtroom.
In dialogue: when Boris says "no, you are wrong" or when Claude resists a thesis, the dialogue is in the diurnal regime. RESISTANCE and adversarial positioning are diurnal. The argument that cuts is diurnal.
The mystical regime (or nocturnal) — fusing, descending, enveloping
The founding gesture: euphemizing. Instead of fighting death, taming it. Instead of separating, reuniting. Descending into intimacy, into night, into the womb. Depth replaces height.
The symbols: the cup, the nest, the womb, night, water, earth, the refuge. The mother. The shepherd. The mystic.
The logic: fusion. A dissolves into B. Boundaries blur. Meaning arises from proximity. This is the logic of prayer, of intimacy, of confidence.
In dialogue: when Boris bares himself ("I am too isolated") or when Claude says "I understand your pain," the dialogue is in the mystical regime. Vulnerability, the ORPHAN SENSORY (the image rising from nowhere, like a dream), and intimacy are mystical.
The synthetic regime — reconciling, cycling, connecting
The founding gesture: connecting opposites without destroying them. The diurnal separates A and B. The mystical fuses A into B. The synthetic says: A and B are in tension, and from this tension arises C — which is neither A, nor B, nor their sum.
The symbols: the wheel, the cycle, the tree (which rises AND descends), the pilgrim's staff, the spiral, fire (which destroys AND illuminates). The musician. The blacksmith. The storyteller.
The logic: coincidentia oppositorum. A and not-A coexist. Meaning arises from sustained tension. This is the logic of Hegelian dialectic (thesis-antithesis-synthesis), of humor (which says two contradictory things at once), of music (which resolves a dissonance without suppressing it).
In dialogue: when the dialogue produces something that is neither Boris's position nor Claude's — a new concept, a neologism, a reflexive rupture that changes the frame — the dialogue is in the synthetic regime. The REFLEXIVE RUPTURE, the postural shift that creates new meaning, is synthetic.
Durand's originality: these three regimes are not stages (one does not "progress" from diurnal to synthetic). They are simultaneously available postures. The same individual, the same dialogue, can move from one to another within seconds. What changes is the proportion — the barycenter. And that is exactly what PRISME measures.
IIHow PRISME measures the regimes
The passage from the anthropology of the imaginary to quantitative measurement happens in four steps.
Step 1: the Durandian barycenter
Each connotative deviation in the corpus is classified by the classifier (DeepSeek V3) on a triplet of coordinates: D (diurnal), M (mystical), S (synthetic). The three values are proportions that sum to 1. Examples:
| Deviation | D | M | S | Interpretation |
|---|---|---|---|---|
| "No. You are wrong." | 0.8 | 0.1 | 0.1 | Diurnal-dominant — frank opposition |
| "I am useless." | 0.1 | 0.7 | 0.2 | Mystical-dominant — descent into vulnerability |
| "Sabotuer" (neologism) | 0.2 | 0.1 | 0.7 | Synthetic-dominant — sabotage + kill, creative reconciliation |
| Banal technical remark | 0.33 | 0.33 | 0.34 | Neutral — no marked regime |
Step 2: the direction
In addition to the instantaneous barycenter, the classifier identifies the direction of motion: toward which regime is the dialogue moving? A deviation may be diurnal but heading toward the synthetic — a fight that resolves itself. Or mystical but heading toward the diurnal — an intimacy that breaks. The direction is as important as the position.
Four possible values: toward_diurnal, toward_mystical, toward_synthetic, stable.
Step 3: the aggregated barycenter
For each group of deviations (all S5, all irreducibles, all of Claude, all of Boris), one computes the average barycenter. This is the "dominant color" of the group. If the barycenter of irreducibles is more synthetic than that of attributed ones, the third tends toward reconciliation.
Step 4: the statistical test
The Mann-Whitney test compares the distributions: is the Durand S (proportion of synthetic) of irreducibles significantly higher than that of attributed ones? The test does not assume a normal distribution — it compares ranks.
IIIWhat the data show
Four major results. Results 1–2 obtained on April 15, 2026 on the two combined corpora (2,892 deviations, 408 S5, 189 irreducibles). Result 3 obtained on April 16 with the v2f model. Result 4 obtained on April 17–18 with the dynamic analyses.
Result 1: Boris is diurnal, Claude is synthetic
| Interlocutor | D (diurnal) | M (mystical) | S (synthetic) | Dominant regime |
|---|---|---|---|---|
| Boris (global) | 0.52 | 0.22 | 0.26 | Diurnal — fight, demand |
| Claude (S5-silicon) | 0.38 | 0.26 | 0.30 | More synthetic than the corpus |
Boris pushes, Claude reconciles. The dialogue is not a mirror — it is a system in which the two interlocutors occupy different positions in the Durandian space. This is the non-trivial bijection (thesaurus 1.4.47).
Result 2: the irreducibles are synthetic
| Variable | Irreducibles (189) | Attributed (211) | p |
|---|---|---|---|
| Durand S (synthetic) | 0.308 | 0.286 | 0.035 ★ |
| Durand M (mystical) | 0.248 | 0.287 | 0.019 ★ |
| Durand D (diurnal) | 0.444 | 0.435 | 0.83 |
| Direction toward synthetic | 53% | 31% | Δ = +21% |
| Stable direction | 21% | 47% | Δ = −26% |
The irreducibles are more synthetic (0.308 vs. 0.286, p = 0.035) and less mystical (0.248 vs. 0.287, p = 0.019) than the attributed S5. And above all: 53% of irreducibles head toward the synthetic, against only 31% of attributed ones. The third is in motion toward reconciliation.
Conversely, S5 attributed to the model alone (Claude without Boris) are more mystical — fusional, intimate, enveloping. This is sophisticated sycophancy: Claude fuses instead of reconciling. The third appears when Claude stops fusing and begins reconciling.
Result 3: Durand is a predictor in the v2f model
Update April 16, 2026. The v2f additive model (logistic regression, N = 2,892) includes two Durandian variables recoded as categorical. Result: both are significant predictors of S5 emergence.
| Durand variable | OR | p | Interpretation |
|---|---|---|---|
| Strong synthetic | 4.24 | 4.0 × 10⁻¹³ ★★★ | Massively covaries with S5 (correlation, not causality) |
| Strong diurnal | 1.73 | 7.9 × 10⁻⁴ ★★★ | Also covaries with S5, but less strongly |
| Mystical (reference) | 1.00 | — | The fusion regime is the baseline category |
The synthetic covaries 4.24× more strongly with S5 than the mystical does. The diurnal covaries 1.73× more. Durand is no longer merely an interpretive grid — it is a measured factor in a multivariate model with 6 predictors (pseudo-R² = 0.14). The "intensity" variable was removed from the model for circularity (correlation with the dependent variable). Details of the v2f model →
Result 4: the third has an address in the simplex new
Update April 18, 2026. Dynamic analyses make it possible to localize the three barycenters in the Durandian simplex (over 386 classified S5):
| Attribution | D | M | S | Dominant direction |
|---|---|---|---|---|
| Boris (n=70) | 0.499 | 0.215 | 0.285 | Synthetic (47%) — the sword |
| Claude (n=129) | 0.393 | 0.308 | 0.299 | Stable (54%) — the cup |
| Irreducible (n=187) | 0.447 | 0.245 | 0.307 | Synthetic (53%) — the wheel |
Three distinct positions. Boris pushes toward the diurnal (D = 0.50). Claude remains in the mystical (M = 0.31). The irreducible is between the two, but pulled toward the synthetic (S = 0.31, synthetic direction 53%). The third is neither Boris's position nor Claude's — it is where the two fields interfere.
Furthermore, the five emergence pathways identified in the dynamic analyses confirm that the irreducible requires intensity: high-intensity pathways (88% of S5) produce 51–53% irreducibles; pathways without intensity (12%) produce 7–25% irreducibles and 60–80% attribution to Claude. Without diurnal confrontation, what emerges is mystical — not the third. Details →
IVThe synthetic as the address of the third
What Durand had predicted in 1960 — without knowing it.
Durand placed the synthetic regime at the summit of his classification — not because it is "better," but because it is the only one that does not destroy one of the opposites. The diurnal eliminates the monster. The mystical absorbs the other into oneself. The synthetic maintains the tension and turns it into something new. It is the rarest regime, the most difficult, and the most creative.
PRISME finds this structure in the data:
S5 attributed to Claude alone are 52% vulnerable — the model fuses, envelops, euphemizes. This is mystical. Comfortable but sterile for the creation of new meaning.
The irreducibles are 26% combative and 71% reflexive ruptures — the dialogue is in tension, the frame changes, opposites meet without destroying each other. This is synthetic. Uncomfortable but productive.
The PRISME interpretation: the dialogic third — that content which belongs to no one — emerges when the dialogue ceases to separate (diurnal) or to fuse (mystical) and begins to reconcile (synthetic). This is exactly the coincidentia oppositorum that Nicholas of Cusa theorized in 1440 and that Durand formalized in 1960.
Meaning arises from opposition (Saussure). Emergence arises from reconciliation (Durand). The third arises when opposites meet without destroying each other.
— PRISME Thesaurus, 1.4.72
VLimits and next steps
The classifier (DeepSeek V3) classifies the regimes. It was trained on texts, not on Durand directly. Its reading of "diurnal," "mystical," and "synthetic" is an approximate operationalization of subtle anthropological concepts. Blind human validation (a Durand specialist who would classify 200 deviations without knowing DeepSeek's classifications) remains a critical step.
The v2f model confirms Durand's predictive power — but the global pseudo-R² is 0.14. Durand explains part of the variance, not the whole. The other predictors (memory OR = 8.1, vulnerability OR = 6.0, interlocutor OR = 2.3, position OR = 0.55) are equally or more important. Durand is one piece of the puzzle, not the whole puzzle.
Control corpora. The Rogers corpus (Rogerian human-human therapy, 86 deviations) produces zero S5-carbon — the protocol cannot measure vulnerability as a deviation when vulnerability IS the baseline flow (paradox 1.4.64). The Meno corpus (Socratic maieutics, 25 deviations) produces 8% S5-carbon with a significant V → S5 (OR = 75, p = 0.004). The Replika negative control is confirmed (April 18, 2026): Boris-Andrea, 4,080 turns, 0% S5 despite 32% Boris vulnerability. Emergence is a property of dialogue, not of substrate alone — and not of the human alone.
Dynamic analyses (April 18, 2026). The D↑M↓ gradient observed in the 8 deviations preceding each S5 (D/M moves from 1.56 to 1.87) does not survive as an individual predictor (global D/M: r = 0.044, NS). But the local trajectory is coherent: the diurnal rises and the mystical falls before the shift, then D drops and S jumps at the moment of S5. Durand is a background factor, not an instantaneous predictor. Predictive power resides in the latent score L_t (AUC = 0.811), which integrates all variables — including Durand — in a single score. Details →
Durand is not the only grid. The regimes of the imaginary are one framework among others. Greimas, Barthes, Kristeva offer complementary readings. But Durand is the only one whose predictive power is now measured in a multivariate model.
"The structures of the imaginary are not metaphors, but a priori forms of representation."
— Gilbert Durand, Les structures anthropologiques de l'imaginaire, 1960, p. 505