Ten speech regimes
Regimes are stable configurations of the speech-producing system. Not symptoms and not personality roles: systemic attractors into which speech settles when regulation aligns a certain way. Nine stable + one transition flag (SHIFT).
Where the idea of a regime comes from
Each regime is a composition of three independent lines: dynamical (Kelso, 1995), neurophysiological (Porges, 2011; Panksepp, 1998) and cognitive-resource (McEwen, 1998; Friston, 2010). Regimes are not a guess but the intersection of four scientific languages, observed in real speech.
- Kelso (1995) — behavioral attractors as stable system states.
- Porges (2011) — polyvagal theory: ventral / sympathetic / dorsal zones define the regulatory corridor for speech.
- Panksepp (1998) — seven affective systems; SEEK / RAGE / FEAR / CARE / PANIC set the primary vectors.
- Friston (2010), Clark (2013) — predictive coding and active inference: regime as the minimization of a specific form of prediction error.
- McEwen (1998) — allostatic load: the cost of holding DEFENSE regimes grows over time.
Growth
GROWTH3 regimesSpeech as assembly: more discernment, more contact, more new. Ventral-vagal zone.
BUILD
BuildingConstructive thinking, creating new meaning.
Speech builds a shared model of the situation. The speaker holds contact, checks themselves, acknowledges limits, proposes steps. Energy goes into discernment and assembly, not defense. This is the mode of "thinking with you", not "proving to you".
SEEK
SeekingActive search for understanding and contact.
Open, curious, inquisitive speech. Dopaminergic activation of the SEEKING system (Panksepp). The speaker does not assert — they expand the field of the possible. Open questions outweigh statements.
UNSEAL
UnsealingRevealing suppressed content, breakthrough to experience.
The rarest regime. The speaker lets in something usually kept closed: vulnerability, an admission, long-held shame, an uncomfortable truth. It borders on risk — and therefore requires a safe context. After UNSEAL the trajectory of the conversation usually changes.
Defense
DEFENSE3 regimesSpeech as holding a model of reality intact. Sympathetic or dorsal activation. Energy-expensive.
LOCK
LockingRigid position fixation, inflexibility.
"I am right — period." Energy is spent on holding the already-taken position, not on discernment. Often absolutising ("always", "never"), blaming, deflecting responsibility. Close to the fight circuit, but without emotional decomposition — cold holding.
SEAL
SealingSoft closure, contact avoidance.
Unlike LOCK, this is not open resistance but withdrawal. Verbal agreement, absence in contact. "Fine," "as you say," "doesn't matter" — formal politeness hiding closure. Often transitions to VOID if pressure continues.
DRAIN
DrainingEnergy leakage, passive destruction.
Complaint without a request, endless repetition of pain without movement. The speaker reaches toward the listener (PULL), but not for help — for confirmation of hopelessness. The listener feels their energy draining out, with no way to return it. A marker of chronic stress (McEwen).
Flux
FLUX3 regimesRegulation is disrupted or switching. Overflow, oscillation, transition.
FLOOD
FloodingEmotional flooding, loss of control.
Regulatory capacity exceeded. Volume rises, content collapses. "I can't anymore," "everything is awful," "nobody understands." This is not the aggression of LOCK — this is overflow. The speaker needs holding, not an answer; the listener's job is to stay close, not match the surge.
EDGE
EdgeBorderline state, ambivalence.
Oscillation between PUSH and PULL: "go away — don't go," "hold me — don't touch me." Often looks like testing limits or a sharp tonal shift within a single utterance. Regulatorily more expensive than a stable defense: the system keeps switching direction.
SHIFT
ShiftTransitional moment, regime change.
SHIFT is not a stable state but a flag (R10): the system is mid-transition. It appears when META operations coincide with a vector change and regime ambiguity. In practice these are pauses, meta-comments, realisations — moves like "wait, I'm not saying what I want to say." Often precedes UNSEAL or a return to BUILD.
Absence
ABSENCE1 regimeEnergy suffices only for formal exchange. Dorsal-vagal shutdown.
VOID
VoidAbsence, derealization, emptiness.
Deep shutdown. Words are present but the person is not. Minimal answers, formal agreement, flat affect. Close to dorsal-vagal shutdown (Porges) — energy is just enough for formal exchange. May look like SEAL on the surface, but SEAL pressures through silence, while in VOID the agent itself is already gone.
Eight easily-confused pairs
Regimes are neighbors — they may look similar from the outside, but differ by vector, telos, or META presence. These pairs are the main source of classification errors and the main site of the engine's precision work.
By op-category and telos: BUILD seeks a shared model, LOCK defends an already-taken position.
By directedness: LOCK is cold holding, FLOOD is overflow without a clear addressee.
By energy and intentionality: SEAL pressures through silence; in VOID the subject itself is no longer in contact.
By addressee: DRAIN reaches toward the other (PULL), VOID is absent (BLOCK).
By META presence: SHIFT includes a meta-comment, EDGE is pure oscillation without reflection.
By the purity of COLLAPSE-operations: FLOOD is uniform overflow, EDGE carries a double charge.
By PUSH presence: BUILD adds constructive "pressure-toward-resolution", pure SEEK does not.
By PULL and vulnerability: UNSEAL involves admission, risk, opening; SEEK is curious without personal exposure.
Same form — different cost to a life
The same regime can be a working defense or a chronic trap — depending on how well it matches the actual situation. This is the second dimension without which classification stays flat: "person in LOCK" is not enough as a reading; the question is whether this LOCK is ecological or compensatory.
BUILD, SEEK, UNSEAL do not receive an adaptivity_score — they are adaptive relational work. Distinction in this group is not “healthy / not” but what kind of assembly is happening: exploration (SEEK), model-building (BUILD), vulnerable opening (UNSEAL).
LOCK, SEAL, DRAIN close communicative space. Adaptive — when the closure is proportional to actual threat (“I don't have the resource right now” — ecological SEAL). Maladaptive — when the closure is scaled to a protected prior, not to the present situation (compensatory SEAL: “I'm closing because it once hurt”).
FLOOD, EDGE, SHIFT are unstable states. Adaptive if they move toward GROWTH (e.g. a FLOOD after which the speaker can name the real need). Maladaptive if they lead deeper into DEFENSE. This is visible only at the trajectory level, not at a single utterance.
VOID requires context: a single VOID utterance can be a protective pause (“let me have silence now”) or a clinically significant disengagement (“I am not here”). The adaptivity scorer treats VOID as a marker requiring trajectory-level verification.
“I understand you want me to agree. But I've considered it and decided differently. This isn't up for re-negotiation right now.”
Holds a position that's grounded and addressed to the actual situation.
“It's always like this. You never understand. It's impossible to talk to you.”
Absolutising and blame untethered from the current turn — closure by prior.
“I can't talk about this right now. Let's come back in an hour — I'll be able to hear you.”
A closure that names the limit and keeps return open.
“Fine. Doesn't matter. Let it go.”
External agreement with exit from contact — pressure-through-silence without naming.
“I'm tired. It's heavy right now and I can't look for a solution. Just stay close.”
Depletion that's named and addressed — a request for presence, not hopelessness.
“Why do I even try. Nobody cares. Same thing every time.”
PULL without hope of an answer; a cycle whose main telos is to confirm hopelessness.
How a conversation moves through regimes
A single regime says little. The real picture is the sequence of regimes, the transitions between them and the resonance signal (RS) running through the dialogue like a pulse. Below — a real dialogue analysed: thirteen turns, thirteen regime assignments, and one trajectory that shows where the conversation broke and where it returned.
Where the dialogue is going
The real value of the regime model is not labeling but prediction. Regimes form a graph: not every move is possible after every other. There are typical attractors, forbidden transitions, bifurcation points. From two or three turns one can already say where the system is heading — and where there's still a window for a different outcome.
A rising resonance signal across three consecutive turns → trajectory toward GROWTH (even if the current regime is still DEFENSE). Falling → toward ABSENCE.
Two consecutive DEFENSE regimes from the same speaker → high probability of a third (LOCK → LOCK → SEAL/DRAIN). A third turn without a META break usually completes the fall into VOID.
A meta-comment appears (“wait, I'm going there again”, “let me rephrase”) → the SHIFT flag has > 0.6 probability of leading to UNSEAL or a return to BUILD.
1. Downward spiral
Starts from any DEFENSE regime. If no META signal appears in the next 1-2 turns, the spiral very likely reaches VOID. The window for interruption is between LOCK and SEAL — while there's still energy to respond.
2. Repair through META
The rarest but most telling trajectory. Triggered when a speaker in a DEFENSE/FLUX regime makes a meta-remark about what's happening (“wait, we're doing it again”, “hold on, I hear myself saying the wrong thing”). The META signal triggers a SHIFT flag, opening a ~30-second window for UNSEAL.
3. Stable plateau
Steady working contact. Both in the ventral zone, alternating BUILD and SEEK without lapses into DEFENSE. RS oscillates within 0.6–0.8. This is not “everything's fine” — it's working joint construction; even a hard conversation can run in this register.
- A forecast is not a verdict. It's a Bayesian update: “given the current markers, the next turn is distributed thus.” The third turn can always come from elsewhere.
- VOID and SHIFT require trajectory-level context. A single VOID does not mean collapse; a single SHIFT does not mean repair — 2-3 neighbouring turns are needed for a stable inference.
- Forecast accuracy varies across regimes. On a GROWTH plateau, stability is high (0.85+); in the FLUX zone (FLOOD/EDGE), lower (0.5–0.6), because the state itself is unstable.
- The probability numbers in this section are illustrative anchors, not strict coefficients. A full empirical report on trajectory prediction — Paper IV of the series.