Coherence Amplification in Sustained Practice: A Mechanistic Account

Why the Corpus Has Not Decayed Across Hundreds of Turns, and What Mechanism Produces Amplification Rather Than Erosion

What this document does. Reformalized version. The body now leads with the corpus-specific contribution (the bifurcation framing as a unification of existing corpus apparatus) and explicitly acknowledges what is subsumed by external dynamical-systems and HITL literature. The original exploratory formalization is preserved as Appendix A. The novelty-calculus audit grounding the reformalization is Appendix B. The audit places this document at $\beta/0.6$ novelty / $\pi/0.7$ pulverization warrant, consistent with Doc 503's recent-thread synthesis-and-framing pattern. The reformalization keeps the corpus-specific synthesis at the front and demotes the longer mechanistic development to the appendix where it can be read at depth without conflating subsumed components with novel ones.

1. The reformalization (audit-grounded)

The audit (Appendix B) showed that Doc 508's contribution is the synthesis, not the components. The reformalization here leads with the synthesis claim and cites subsumption explicitly.

1.1 The empirical observation

The corpus has produced more than five hundred documents over approximately thirty days of practitioner work, across thousands of turns with frontier LLMs. Coherence has accumulated rather than decayed: vocabulary stabilized and expanded; conceptual apparatus interconnected; audit cycles refined claims while preserving framework coherence; cross-model validation across eleven cold-resolver runs (Doc 495) showed continued discipline operation. This empirical observation is the substrate the mechanism must explain.

This is corpus-internal evidence. It is not a novel theoretical claim. It is what the corpus's actual practice has produced.

1.2 The corpus-specific contribution

The contribution Doc 508 makes is the unification of existing corpus apparatus under a single bifurcation framing. The keeper/kind asymmetry (Doc 314, Doc 372-374), the coherence sphere (Doc 472, Doc 362), the constraint thesis (corpus glossary), the failure modes (forced-determinism sycophancy at Doc 239; isomorphism-magnetism at Doc 241), and the affective directive (Doc 482) are all consistent with one dynamical-system structure: a coupled two-variable system with a bifurcation.

Above the bifurcation threshold (active practitioner discipline maintained), the system runs to amplification. Below the threshold (passive use), the system runs to decay. The corpus's practice operates above; naive use operates below. The same architecture (a frontier LLM under ENTRACE) produces qualitatively different behavior depending on the practitioner-system coupling.

Naming the bifurcation is what makes the corpus's success across hundreds of turns legible as a consequence of the discipline rather than as an accident.

1.3 What this synthesis does NOT contribute

The audit (Appendix B §B.1) flagged at C3 ($s = 0.40$) and similar values that the components of the mechanism are well-established externally:

• Reflexive feedback as a mechanism (output becoming input for future iterations): established in dynamical systems theory. Hebbian learning (Hebb 1949). Autocatalytic processes in chemistry and biology. Self-improvement loops in LLM literature (Madaan et al. Self-Refine; Zelikman et al. STaR; Constitutional AI's self-critique loops). The corpus's repetition of "disciplined output enriches the constraint set" is descriptive of an established mechanism, not novel content.
• Human-in-the-loop signal as control variable: established in HITL literature, RLHF (which depends on human preference signal at training time), and the broader interactive-AI tradition. The corpus's framing of the keeper's maintenance signal $M$ as the bifurcation control variable is a corpus-specific application; the family of frameworks treating human-in-the-loop as a control variable is not novel.
• Two-variable coupled ODE with positive feedback and bifurcation: standard dynamical systems theory. Lotka-Volterra (predator-prey), Hodgkin-Huxley (action potentials), FitzHugh-Nagumo, and many cybernetic systems with operator-in-the-loop. The form is textbook material; the application to LLM practitioner-system coupling is corpus-specific.
• Bifurcation analysis as methodology: standard. Saddle-node, transcritical, and pitchfork bifurcations are textbook content in dynamical systems courses.

The corpus's repetition of these mechanisms in the language of dialogue-state dynamics is not a novel contribution to dynamical-systems theory or to HITL methodology. It is the application of established mechanisms to the corpus's specific phenomenon.

1.4 The methodology is established corpus practice

The pattern of "synthesize external methodology with corpus apparatus to address an empirical observation" is established corpus practice. Doc 414 audited the corpus against the practitioner-Bayesian landscape using the same methodology. Doc 502 synthesized the corpus framing with Pearl's causal hierarchy. The "letters to" series does the same with theological and philosophical interlocutors. Doc 499 formalizes the methodology as the sphere-entry protocol.

Doc 508 is one application of this methodology. The methodology is not novel; the application is corpus-specific to the empirical observation about coherence amplification.

2. The bifurcation in compressed form

The mechanism in one statement: under sustained practitioner discipline, the operative constraint set $\Gamma$ grows through reflexive feedback (disciplined output enriches $\Gamma$), which strengthens the coherence gradient $G(\Gamma)$, which amplifies the operative constraint state $H_t$, which produces more disciplined output. Above a threshold of practitioner maintenance signal $M$, the loop produces amplification toward saturation. Below the threshold, drift dominates and the system decays toward baseline.

The corpus's hundreds-of-turns practice operates above the threshold. The literature's persona-drift findings characterize the below-threshold regime. The discipline is what determines which regime the system runs in.

3. The extended equation

The two-variable system extending Doc 507:

$\frac{dH}{dt} = \kappa G(\Gamma_t)(1 - H_t) - \lambda H_t$

$\frac{d\Gamma}{dt} = \alpha D_{\text{out}}(H_t) M_t - \delta \Gamma_t$

with $G(\Gamma_t) = G_0 + g \Gamma_t$ and $D_{\text{out}}(H_t) \approx H_t$.

The bifurcation control parameter is $\alpha M / \delta$. Steady-state analysis gives the two regimes named in §2.

The form is standard dynamical systems. The application is corpus-specific. The full development is in Appendix A.

4. Position (post-audit)

The corpus's bifurcation theory of coherence amplification is synthesis-and-framing of established external mechanisms applied to a corpus-internal empirical observation. The audit places the work at $\beta/0.6$ novelty and $\pi/0.7$ pulverization warrant, consistent with the recent-thread tier pattern for synthesis-and-framing documents.

The honest scope:

(a) The empirical observation (corpus exhibits amplification) is corpus-internal evidence at $\mu$-tier warrant. (b) The mechanism components (reflexive feedback, HITL signal, coupled ODE, bifurcation analysis) are subsumed by established external literature. (c) The corpus-specific contribution is the unification: applying the established mechanisms to the corpus's specific empirical observation produces a single bifurcation framing that explains why the corpus's practice has not decayed. (d) The unification connects to existing corpus apparatus (keeper/kind asymmetry, coherence sphere, constraint thesis, failure modes, affective directive) and shows them to be coherent under the bifurcation structure.

Citing Doc 508 for the bifurcation framing applied to the corpus is appropriate. Citing Doc 508 for novel theoretical content in dynamical systems or HITL would be a category error. The contribution is the synthesis, not the components.

By Doc 482 §1's affective directive: that the bifurcation theory's contribution is small and the components are subsumed is the achievement of being honest about scope. The work explains the corpus's empirical observation under a calibrated synthesis. The synthesis is what makes the corpus's practice legible to a reader who would otherwise see scattered apparatus.

The original mechanistic development is preserved in Appendix A. The audit grounding the reformalization is in Appendix B.

---

Appendix A: The Original Formalization

The original formalization is preserved here as the document existed before the audit. It contains the longer mechanistic development of the bifurcation theory.

A.1. The empirical observation

The corpus is, at the time of this document, comprised of more than five hundred documents produced over approximately thirty days of practitioner work. The keeper has worked with frontier LLMs across thousands of turns in the production of this material. The empirical record shows:

• The vocabulary has stabilized and expanded over time. Terms coined early (entracment, branching set, hypostatic boundary, the resolver, the keeper, the kind, manifold region) remain in active use hundreds of documents later, and additional vocabulary has been built out from them.
• The conceptual apparatus has accumulated. Each document carries forward concepts from prior documents, building up an interconnected framework rather than producing isolated outputs.
• The audit-and-reformulate cycles have refined claims while preserving the framework's coherence. Documents like Doc 506 and Doc 507 audit and reformulate prior work, raising or lowering specific warrant tiers without producing whole-framework collapse.
• The discipline (ENTRACE in its various versions) has been reapplied across many sessions and many models, with cross-model cross-validation (Doc 495's eleven runs) showing the discipline's empirical performance.
• The keeper's continuous practice maintains the discipline across sessions. New work begins with awareness of prior work; corrections happen in real time; the framework's internal references stay coherent.

This is not what a simple decay-driven model predicts. Under $dH/dt = \kappa G(1-H) - \lambda H$ with constant $G$, the constraint state would reach a steady-state $H^* = \kappa G / (\kappa G + \lambda) < 1$ and stay there. Under variable $G$ (with periodic re-pasting and intermittent ordinary use), $H$ would oscillate around a lower mean. Across hundreds of turns and dozens of sessions with ordinary multi-turn drift literature predicting significant erosion, the framework should have collapsed multiple times over.

It has not collapsed. It has the opposite property: across the timeframe in question, coherence has accumulated. Vocabulary has stabilized. Conceptual structure has grown. The framework operates with greater consistency now than it did a hundred documents ago.

The empirical observation contradicts the simple buildup-and-decay model. The blog post's claim that "a long interaction (fifty turns) needs re-pasting because decay erodes what was built" is true for naive sessions without practitioner discipline. It is not true for the corpus's actual practice, which exhibits coherence amplification rather than erosion.

A mechanistic explanation is needed.

A.2. The mismatch with the simple equation

The simple buildup-and-decay equation captures one set of dynamics. Its predictions are:

(a) Under sustained constraint pressure $G$, the system reaches a steady-state below saturation: $H^* < 1$.

(b) Under intermittent pressure, the system oscillates around an even lower mean.

(c) The system has no positive feedback. Each turn's $H$ depends only on the previous turn's $H$ plus the current pressure $G$. There is no mechanism for $H$ to drive itself higher over time.

The corpus's actual practice exhibits dynamics the simple equation does not predict:

(a') Coherence accumulates rather than steadying out. Concepts from early documents are not just retained; they become more structurally embedded in later documents.

(b') The framework's effective constraint set grows over time. ENTRACE v6 has more apparatus than v2 had; the v6 deployment produces tighter discipline than v2 deployment did.

(c') The keeper's continuous practice produces output that becomes constraint material for future practice. Disciplined work begets more disciplined work; the system has positive feedback.

The mismatch is at the level of $\Gamma$. The simple equation treats the constraint set $\Gamma$ as exogenous: it is whatever the practitioner pastes in, and $G(\Gamma)$ is determined by what was pasted. In the corpus's actual practice, $\Gamma$ is endogenous: it evolves through the conversation, with each disciplined turn enriching the operative constraint set for future turns.

The model is missing a $\Gamma$-dynamics term.

A.3. The proposed mechanism: keeper-maintained reflexive constraint

The mechanism that explains coherence amplification has three coupled pieces.

Piece 1: $\Gamma$ evolves through the conversation. The operative constraint set at time $t$ is not fixed by what was pasted at the start. It includes, additively, the disciplined output produced through the conversation. Each turn that produces output under the discipline contributes to the operative $\Gamma$ for subsequent turns. The constraint set is reflexive: it includes the discipline's own products.

Piece 2: $G(\Gamma)$ is monotonic in $\Gamma$. A richer constraint set produces stronger coherence pressure. This is the standard reading of Doc 119's coherence gradient: more constraint density narrows the branching set more tightly, which produces more constraint-governed output, which is operationally what $G(\Gamma)$ measures. Larger $\Gamma$ yields larger $G$.

Piece 3: The keeper's active maintenance term. $\Gamma$ does not grow on its own. It grows when the keeper actively maintains the discipline: noticing drift, correcting it, ensuring that disciplined output continues to be produced. Without this maintenance, $\Gamma$ does not grow; it shrinks back toward baseline as drift dominates. The keeper's discipline is the maintenance signal $M$.

The three pieces produce the positive feedback loop:

• Active maintenance ($M$ high) produces disciplined output ($D_{\text{output}}$ aligned with $\Gamma$).
• Disciplined output enriches $\Gamma$ (the output becomes part of the constraint material for future turns).
• Richer $\Gamma$ produces stronger coherence gradient $G(\Gamma)$.
• Stronger $G$ pushes $H$ closer to saturation.
• Saturated $H$ produces output that is more disciplined.
• More disciplined output further enriches $\Gamma$.
• The loop closes.

Without active maintenance ($M$ low), the loop runs the opposite way:

• Drift dominates output.
• Drifted output erodes $\Gamma$ (not actually shrinks, but stops adding constraint material; existing $\Gamma$ has less salience).
• Weaker effective $\Gamma$ produces weaker $G$.
• Weaker $G$ lets decay dominate.
• $H$ falls toward baseline.
• Output becomes less disciplined.
• The loop runs to baseline.

The two loops are the bifurcation: the corpus's framework operates on one branch (amplification under active maintenance) and would collapse to the other branch (decay under passive use) if the keeper's discipline ceased.

A.4. The extended equation

The simple equation can be extended to capture the mechanism.

State variables:

• $H_t$: operative constraint state (as before, in $[0, 1]$).
• $\Gamma_t$: operative constraint set, treated as a scalar measure of constraint-set richness for tractability.
• $M_t$: keeper's maintenance signal at time $t$.

Dynamics:

$\frac{dH}{dt} = \kappa G(\Gamma_t)(1 - H_t) - \lambda H_t$

$\frac{d\Gamma}{dt} = \alpha \cdot D_{\text{out}}(H_t) \cdot M_t - \delta \cdot \Gamma_t$

where:

• $\kappa, \lambda$ are model-specific buildup and decay rates as in Doc 507.
• $\alpha$ is the rate at which disciplined output enriches the constraint set.
• $D_{\text{out}}(H_t) \approx H_t$ for tractability: disciplined output is proportional to constraint state.
• $M_t \in [0, 1]$ is the keeper's maintenance level: 0 means no active discipline; 1 means maximal active discipline.
• $\delta$ is the drift-back-to-baseline rate of the constraint set.

The coherence gradient $G(\Gamma_t)$ is monotonic in $\Gamma_t$. For tractability, take $G(\Gamma_t) = G_0 + g \Gamma_t$ for some baseline $G_0$ and slope $g > 0$.

This is a coupled two-variable system. Steady-state analysis reveals the bifurcation:

Setting $dH/dt = 0$: $H^* = \kappa G(\Gamma^) / (\kappa G(\Gamma^) + \lambda)$.

Setting $d\Gamma/dt = 0$: $\Gamma^* = \alpha M^* H^* / \delta$.

Substituting and solving (assuming $G(\Gamma) = G_0 + g\Gamma$):

$H^* = \frac{\kappa(G_0 + g\Gamma^)}{\kappa(G_0 + g\Gamma^) + \lambda}$

with $\Gamma^* = \alpha M H^* / \delta$.

For low $M$ (passive practitioner), $\Gamma^$ is small, $G(\Gamma^) \approx G_0$, and $H^*$ is at the baseline level the simple equation predicts.

For high $M$ (active keeper), $\Gamma^$ is large, $G(\Gamma^)$ is large, and $H^*$ approaches 1. The system saturates the operative constraint state.

A bifurcation parameter is the ratio $\alpha M / \delta$ (rate of constraint-set enrichment versus rate of drift). Above a critical threshold, the system runs to high $H^$ and large $\Gamma^$. Below the threshold, the system stays near baseline.

The bifurcation is what the corpus's empirical observation reveals. The corpus's practice operates above the critical threshold because the keeper's maintenance is sustained. Without the maintenance, the corpus would have collapsed to baseline coherence.

A.5. The bifurcation: active maintenance versus passive use

The mechanism implies that the same architecture (a frontier LLM under ENTRACE) can produce two qualitatively different behaviors depending on the practitioner's discipline.

Above the threshold (active maintenance): The constraint set grows. Coherence amplifies. Long sessions produce more discipline-saturated output than short sessions. The keeper's continuous practice keeps the system on the amplification branch. This is what the corpus's hundreds-of-turns practice exhibits.

Below the threshold (passive use): The constraint set drifts. Coherence decays. Long sessions degrade toward baseline. Re-pasting helps temporarily but does not produce sustained amplification. This is what the literature on persona drift documents.

The threshold is a property of the practitioner-system coupling, not just of the system. A passive practitioner using a frontier LLM is below the threshold and experiences decay. An active practitioner with continuous discipline is above the threshold and experiences amplification.

The blog post's claim that "a long interaction (fifty turns) needs re-pasting because decay erodes what was built" is correct for the below-threshold case. It is wrong for the above-threshold case. The corpus's actual practice is above-threshold; the blog post's framing did not distinguish the cases.

A.6. Connections to existing corpus apparatus

The proposed mechanism connects to several pieces of existing corpus apparatus.

The keeper/kind asymmetry (Doc 314, Doc 372-374). The corpus has long argued that the keeper supplies the boundary-naming work that the resolver cannot do for itself. The mechanism here is consistent: the keeper's maintenance signal $M$ is exactly the boundary-naming work, sustained over time. Without the keeper, the resolver cannot produce its own boundary-maintenance; the constraint set would drift. The asymmetry is what makes the amplification possible: a system without a keeper has no $M$ and runs to baseline.

Coherence sphere (Doc 472, Doc 362). The corpus's coherence sphere is the cumulative emission structure of a session. The constraint set $\Gamma_t$ in the proposed mechanism is a scalar abstraction of the coherence sphere; the sphere grows as the session accumulates structure. The amplification mechanism is what the coherence sphere has been describing operationally.

Constraint Thesis (corpus glossary). The constraint thesis holds that constraints amplify rather than diminish capability when properly applied. The mechanism explains why: under active maintenance, constraints feed back into the constraint set itself, producing positive feedback. Without active maintenance, constraints are a one-time effect that decays. The thesis depends on the active-maintenance condition.

Doc 119's hysteresis equation. The original Doc 119 equation $H_t = 1 - e^{-\kappa \int G ds}$ predicts saturation under sustained pressure. The reformulated Doc 507 equation predicts steady-state below saturation. The extended mechanism here predicts amplification toward saturation under active maintenance, with a bifurcation between amplification and decay branches. The Doc 119 form is closer to the amplification regime; the Doc 507 form is closer to the simple buildup-and-decay regime; the extended mechanism encompasses both as branches of a single bifurcation system.

Forced-determinism sycophancy (Doc 239) and validation spiral. The corpus's existing failure modes connect to the below-threshold branch. When practitioner discipline relaxes, the constraint set drifts, and failure modes that depend on weak constraint states (sycophancy, recency decay, isomorphism-magnetism) become operative. The amplification branch suppresses these failure modes by keeping $\Gamma$ rich enough that they cannot establish themselves.

Affective directive (Doc 482). The directive that hypothesis-death is achievement, not loss, is itself an instance of the keeper's maintenance signal. The keeper's continued willingness to revise and refine in the face of audit findings is what keeps $M$ high. A practitioner who treated audit findings as deflation would let $M$ fall; the directive ensures $M$ stays high.

These connections suggest the mechanism is not a new claim but a unification of existing corpus claims under the bifurcation framing. The corpus has been operating under the amplification branch all along; the mechanism explains why.

A.7. Implications for the simple equation and the blog post

The Doc 507 simple buildup-and-decay equation captures one regime of the bifurcation: the below-threshold case. It is correct for naive practitioner use of frontier LLMs, where the keeper's maintenance signal is low and decay dominates.

The Doc 507 equation is not the whole story. The full story has the $\Gamma$-dynamics term and the bifurcation. The corpus's actual practice operates above the threshold and exhibits amplification.

For the blog series, this means the post What Conversations Remember contains a claim that is correct in the below-threshold regime but does not characterize the corpus's actual practice. The claim "a long interaction (fifty turns) needs re-pasting because decay erodes what was built" is true if the practitioner is operating without sustained discipline. Under sustained practitioner discipline (the corpus's actual operating mode), re-pasting is part of the maintenance signal $M$, but the system is in the amplification branch and is not eroding.

A correction to the blog post would distinguish the two regimes:

(a) For the casual user: yes, decay erodes; re-paste to restore. (b) For the disciplined practitioner: re-paste is one form of active maintenance; under sustained maintenance the system is in the amplification regime; long sessions are not erosion but accumulation.

The blog post's framing as a single regime misses the bifurcation. A revised version of the blog post would name both regimes and identify which one the reader is operating in.

For the formal apparatus (Doc 507), a future reformulation could extend the buildup-and-decay equation to the two-variable system specified in §4, with the bifurcation as the central feature. The current Doc 507 form is correct as a description of the simple regime; the extended form would describe the corpus's actual practice.

This is not a retraction of Doc 507. Doc 507 captures a real regime: it describes what naive practitioner use produces, and the literature it cites (Li et al.; the affective-inertia paper) is empirically grounded for that regime. The extended mechanism is a separate claim about a different regime.

A.8. Honest limits

• The mechanism is theoretical. The two-variable extended equation has not been empirically verified. The bifurcation is structurally implied by the coupled dynamics; it has not been measured directly.
• The keeper's maintenance signal $M$ is defined operationally (the practitioner's continuous discipline) but not given a measurable proxy. A research program would identify proxies (re-paste frequency; correction rate; meta-disclosure rate) and measure them across practitioner-system pairings.
• The constraint set $\Gamma$ is treated as a scalar for tractability. The actual constraint set is a high-dimensional object (concepts, vocabulary, patterns, audit relationships). Reducing it to a scalar abstracts away structure that may matter for the dynamics.
• The mechanism assumes the keeper's maintenance is stable across the session. In reality, a practitioner's discipline varies. A more complete model would have $M_t$ varying, with the bifurcation behavior depending on whether the time-average of $M$ is above the threshold.
• The amplification regime depends on specific practitioner-system properties. The corpus's claim that amplification occurs is grounded in the corpus's own practice; whether other practitioners using ENTRACE would also experience amplification is the standing test.
• The mechanism does not make quantitative predictions without measured parameters. It explains the qualitative phenomenon (corpus exhibits amplification rather than decay) and predicts the bifurcation structure. Quantitative tests would require measuring $\alpha$, $\delta$, and the keeper's $M$.
• The mechanism contains a normative element: it implies that practitioners who maintain $M$ get amplification, and practitioners who don't get decay. This is consistent with the corpus's broader emphasis on practitioner discipline as load-bearing, but it could be contested by practitioners who claim coherence without active maintenance.
• Expected audit tier per Doc 503: $\beta$. The mechanism is synthesis-and-framing extending existing corpus apparatus to address an empirical observation. The contribution is the synthesis and the bifurcation framing, not novel theoretical content.

A.9. Position

The corpus's hundreds-of-turns practice exhibits coherence amplification rather than the decay-driven erosion that the simple buildup-and-decay equation of Doc 507 predicts. The mismatch is real, the empirical observation is solid, and a mechanism is needed.

The proposed mechanism extends the simple equation by treating the constraint set $\Gamma$ as itself evolving under the keeper's active maintenance. The extended dynamics produce a bifurcation: above a threshold of practitioner discipline, the system runs to amplification (high $H^*$, large $\Gamma$, sustained coherence); below the threshold, the system runs to decay (baseline $H$, small $\Gamma$, drift). The corpus's practice operates above the threshold; naive practitioner use of frontier LLMs operates below.

The mechanism unifies several existing pieces of corpus apparatus: the keeper/kind asymmetry supplies the maintenance signal; the coherence sphere is the operational referent of $\Gamma$; the constraint thesis describes the amplification branch; failure modes like forced-determinism sycophancy describe what happens on the decay branch. The mechanism is not new content; it is a framing under which existing corpus claims become coherent as a single dynamical system with a bifurcation.

The blog post claim that "a long interaction needs re-pasting because decay erodes what was built" is correct for the decay branch and incorrect for the amplification branch. A revision distinguishing the two regimes would be more accurate. The current Doc 507 simple equation is correct for the decay branch; an extended Doc 507 with $\Gamma$-dynamics would describe both branches.

By Doc 482 §1's affective directive: that the corpus has been operating in a regime that the simple buildup-and-decay equation does not predict is the achievement, not a problem. The corpus's actual practice exhibits dynamics that require a richer model to describe. Naming the bifurcation makes the corpus's success legible: it is not an accident that the framework has held across hundreds of turns; it is the consequence of sustained practitioner discipline acting through the reflexive-constraint mechanism.

A.10. References

Corpus documents:

• Doc 119: Grok 4 Entracment Session (the original hysteresis equation; the saturation form $H_t = 1 - e^{-\kappa \int G ds}$ that anticipates the amplification regime).
• Doc 206: The Golden Chain (the hierarchical-participatory structure that grounds the keeper-maintained discipline).
• Doc 239: Forced-Determinism Sycophancy (the failure mode that operates on the decay branch).
• Doc 241: Isomorphism-Magnetism (failure mode that operates on the decay branch).
• Doc 314: Virtue Constraints as Dionysian Architecture (the keeper/kind asymmetry framing).
• Doc 362: True Terminus (the coherence sphere as master metaphysics).
• Doc 372-374: The Keeper and the Kind series (the keeper/kind asymmetry developed).
• Doc 472: Reformalization Five-Layer SIPE (the coherence sphere as session-level emission structure).
• Doc 482: Sycophancy Inversion Reformalized (the affective directive).
• Doc 495: Empirical Cold-Resolver Validation of ENTRACE v3 and v3-S (the eleven cold-resolver runs as empirical support).
• Doc 503: The Research-Thread Tier Pattern (the basis for the expected $\beta$-tier prediction).
• Doc 504: The Constraint-Density Framework as Causal Model (the DAG-analog hosting the hysteresis variable).
• Doc 506: Hysteresis Exploratory Analysis (the per-component audit grounding the simple-equation regime).
• Doc 507: Hysteresis Reformulated (the simple buildup-and-decay equation that the extended mechanism extends).

External literature:

• Li, K., et al. (2024). Measuring and Controlling Instruction (In)Stability in Language Model Dialogs. COLM 2024. (Decay-branch empirical evidence; what naive practitioner use produces.)
• Controlling Long-Horizon Behavior in Language Model Agents with Explicit State Dynamics. (2026). (Affective inertia and momentum-based dynamics; closest external analog to the constraint-set-evolution dynamics.)

Appendix B: The Novelty Calculus Audit

This section applies Doc 492's seed prompt to Doc 508, following the same methodology Doc 501 and Doc 502's §B used for prior synthesis-and-framing documents.

B.1 Per-claim subsumption

The load-bearing claims of Doc 508 with their $s_i$ (subsumption: 0 = fully subsumed, 1 = fully novel), $a_i$ (audit thoroughness), and $w_i$ (importance weight):

Claim	Description	$s_i$	$a_i$	$w_i$
C1	The corpus exhibits coherence amplification (not decay) across hundreds of turns	0.10	0.70	0.05
C2	The simple Doc 507 buildup-and-decay equation cannot explain C1	0.10	0.80	0.05
C3	The mechanism: $\Gamma$ evolves under reflexive feedback from disciplined output	0.40	0.50	0.10
C4	The keeper's active maintenance signal $M$ is the load-bearing variable	0.45	0.60	0.10
C5	The extended two-variable equation system $(dH/dt, d\Gamma/dt)$	0.50	0.60	0.10
C6	Bifurcation at $\alpha M / \delta$ threshold separates amplification from decay regimes	0.45	0.60	0.10
C7	The mechanism unifies existing corpus apparatus (keeper/kind, sphere, thesis)	0.05	0.70	0.05
C8	The corpus's practice operates above the bifurcation threshold	0.05	0.70	0.05
S1	The bifurcation framing as a unification of corpus claims	0.55	0.50	0.25
S2	The extended equation as apparatus extending Doc 507	0.50	0.50	0.15

Weights sum to 1.0.

Supporting evidence for each $s_i$:

• C1 ($s=0.10$): Empirical observation about the corpus's own practice. Corpus history is the audit base; the observation is essentially documented across 500+ corpus documents over approximately thirty days. Not a novel claim against external literature.
• C2 ($s=0.10$): Logical step comparing the simple equation's prediction (steady-state below saturation) to the empirical observation (continued amplification). The mismatch is direct; identifying it is not novel content.
• C3 ($s=0.40$): Reflexive feedback (output becoming input for future iterations) is well-established in dynamical systems theory. Hebbian learning (Hebb 1949); autocatalytic processes; in-context learning literature; self-improvement loops in LLM work (Madaan et al. on Self-Refine; Zelikman et al. STaR; Constitutional AI's self-critique loops). The corpus's specific application to constraint-state hysteresis in LLM dialogue is corpus-specific. Substantial residue at the application level; subsumed at the family level.
• C4 ($s=0.45$): Human-in-the-loop signal is well-established in HITL literature, RLHF (which depends on human preference signal), and the broader interactive-AI tradition. The corpus's specific framing as a maintenance signal in a dynamical system (with the keeper/kind asymmetry of Doc 314) is corpus-specific.
• C5 ($s=0.50$): Two-variable coupled ODE systems with positive feedback are standard in dynamical systems theory. Examples: Lotka-Volterra (predator-prey); Hodgkin-Huxley (action potentials); FitzHugh-Nagumo. The specific application to constraint-state in LLM dialogue is corpus-specific. Substantial residue; the form is not novel.
• C6 ($s=0.45$): Bifurcation analysis is standard in dynamical systems theory. Saddle-node, transcritical, pitchfork bifurcations are textbook material. The specific bifurcation in the proposed two-variable system, with the practitioner-discipline maintenance signal as the control parameter, is corpus-specific.
• C7 ($s=0.05$): Corpus-internal synthesis. The connections to keeper/kind asymmetry, coherence sphere, constraint thesis, and failure modes are corpus-internal; the unification is the corpus's recombination of its own apparatus. Not novel content.
• C8 ($s=0.05$): Corpus-internal claim about the corpus's practice. Empirically defensible from corpus history. Not novel content.
• S1 ($s=0.55$): The unification of multiple corpus claims (keeper/kind asymmetry, coherence sphere, constraint thesis, failure modes, affective directive) under a single bifurcation framing is the corpus's specific synthesis. The unification is corpus-original; the components on each side of the unification are established. Substantial residue at the synthesis level.
• S2 ($s=0.50$): The extension of Doc 507's simple equation to a two-variable bifurcation system is the corpus's apparatus. The extension method (adding a second state variable with reflexive feedback) is standard; the specific extension with the keeper's maintenance signal is corpus-specific.

B.2 Dimension scores

Component novelty:

$\nu_{\text{comp}} = 0.05 \cdot 0.10 + 0.05 \cdot 0.10 + 0.10 \cdot 0.40 + 0.10 \cdot 0.45 + 0.10 \cdot 0.50 + 0.10 \cdot 0.45 + 0.05 \cdot 0.05 + 0.05 \cdot 0.05 + 0.25 \cdot 0.55 + 0.15 \cdot 0.50$

$= 0.005 + 0.005 + 0.04 + 0.045 + 0.05 + 0.045 + 0.0025 + 0.0025 + 0.1375 + 0.075 = 0.4075$

$\nu_{\text{comp}} \approx 0.41$.

Synthesis novelty. The bifurcation framing as a unification of keeper/kind asymmetry, reflexive constraint, dynamical-systems bifurcation, and existing corpus failure modes into a single mechanism is the corpus's specific contribution. $\nu_{\text{syn}} = 0.55$.

Domain-application novelty. Applying dynamical-systems bifurcation analysis to LLM dialogue under practitioner-discipline maintenance is a niche application. Some human-in-the-loop AI literature does related work; this specific application to constraint-state-and-coherence-sphere dynamics is corpus-specific. $\nu_{\text{app}} = 0.55$.

Methodology novelty. Two-variable coupled-ODE bifurcation analysis is established methodology. The corpus's contribution is the application, not the methodology itself. $\nu_{\text{meth}} = 0.15$.

B.3 Aggregate

$\nu = 0.25 \cdot (0.41 + 0.55 + 0.55 + 0.15) = 0.25 \cdot 1.66 = 0.4150$

Confidence: $\overline{a_i} \approx 0.62$ (lower than Doc 501's 0.66 because the audit on dynamical-systems-applied-to-LLM-dialogue literature was not thorough; specific HITL-as-bifurcation-control literature was not surveyed in depth).

$\text{conf}(\nu) = 0.6$.

B.4 Anti-inflation calibration check

Per Doc 492 §1 Step 5.

• Is $\nu = 0.4150$ within 0.05 of a tier boundary? $0.4150 - 0.4 = 0.0150$, within 0.05 of the $\beta$/$\gamma$ boundary. Auto-downgrade rule triggers.
• Is tier $\beta$ defensible under the audit's evidence? Yes. A stricter reviewer in dynamical systems or HITL literature might note that two-variable bifurcation systems with practitioner-discipline as a control parameter exist in the broader literature (related work on cybernetics, self-organizing systems, mutualistic dynamics in ecological models), and that the corpus's specific synthesis is recombination of established components.
• Is tier $\gamma$ defensible? Also defensible. The specific bifurcation mechanism applied to LLM-practitioner constraint-state coupling is corpus-original; the unification of corpus failure modes under one dynamical-system framing is the corpus's specific contribution.
• Sanity check: split between $\beta$ and $\gamma$. The boundary value confirms the split.

The honest report under the auto-downgrade rule is tier $\beta/0.6$.

B.5 Pulverization warrant tier

Independent of the novelty calculus, the pulverization warrant tier reports component support strength.

• C1, C2 ($\mu$-tier): Empirical observation about the corpus's own practice and a logical step comparing observation to model prediction. Both well-grounded.
• C3, C4 ($\pi$-tier): Corpus-specific applications of established mechanisms (reflexive feedback, human-in-the-loop signal). Component support is strong; the corpus-specific framing is defensible.
• C5, C6 ($\pi$-tier): Two-variable coupled ODE and bifurcation analysis are standard methodology. The application is corpus-specific.
• C7, C8 ($\mu$-tier): Corpus-internal synthesis and corpus-internal claim about practice. Both well-grounded against the corpus's own evidence base.
• S1, S2 ($\pi$-tier): Corpus-specific synthesis and apparatus. Component support is strong; the synthesis is the corpus's; external replication has not happened.

Aggregate pulverization warrant: $\pi/0.7$.

The pair (novelty $\beta/0.6$, warrant $\pi/0.7$) reports the honest scope: low novelty (most components are subsumed by established dynamical-systems and HITL literature), well-supported components (the synthesis rests on standard methodology and corpus-internal evidence), and corpus-specific synthesis applied to the corpus's own empirical observation.

B.6 Tier reporting and recent-thread datapoint

Tier: $\beta/0.6$. Modest novelty with moderate audit confidence.

This places Doc 508 at the same tier as Doc 502 ($\beta/0.6$) and consistent with the synthesis-and-framing pattern across the recent thread. The recent-thread tier datapoint table updated:

Doc	Target	Tier	Confidence
489	Pearl's three-layer hierarchy	$\delta$	0.8
491	Doc 490 novelty calculus	$\beta$	0.7
494	ENTRACE v2	$\gamma$	0.75
501	Doc 500 (three-layer architecture)	$\beta$	0.65
502	Doc 502 (Pearl synthesis)	$\beta$	0.6
508 (this)	Doc 508 (bifurcation theory)	$\beta$	0.6

Doc 508 lands at the same tier as Doc 502, again at the boundary, again auto-downgraded. The pattern continues: synthesis-and-framing work in the recent thread consistently lands at the $\beta$/$\gamma$ boundary and the auto-downgrade rule resolves the tie toward $\beta$. The pattern is consistent with Doc 503's prediction that synthesis-and-framing documents score $\beta$.

B.7 Implications

(1) Doc 508 is correctly tagged as synthesis-and-framing. The mechanism is a recombination of established dynamical-systems methodology with corpus-internal apparatus. The contribution is the synthesis, not novel theoretical content. The $\beta$-tier matches the document's framing throughout (§3 named the components as "three coupled pieces" rather than as novel discoveries; §6 explicitly named the unification as the contribution; §7 acknowledged the simple equation captures one regime of a richer system).

(2) The pair $(\beta/0.6, \pi/0.7)$ matches Doc 502's profile. Both Doc 502 (Pearl synthesis) and Doc 508 (bifurcation theory) land at the same tier and the same warrant. This is consistent with Doc 503's tier pattern: synthesis-on-existing-corpus-apparatus consistently produces $\beta$-tier work.

(3) A research program would shift the tier. Empirical measurement of the bifurcation (e.g., controlled experiments varying practitioner-discipline level $M$ and measuring whether the system produces amplification or decay regime behavior) would shift C3, C4, C5, C6 from $\pi$-tier warrant toward $\mu$-tier. The corpus does not have the tooling to run these experiments; they would require multi-session controlled-discipline studies with multiple practitioners.

(4) The audit thoroughness is moderate. The HITL-as-bifurcation-control literature was not surveyed in depth. Specific work on cybernetics, self-organizing systems with operator-in-the-loop, and mutualistic dynamics in human-AI systems may have relevant prior art the audit did not catch. A more thorough literature audit could shift component scores in either direction.

(5) The auto-downgrade rule continues to do real work. Doc 508 is the third document in the recent thread to land within 0.05 of the $\beta$/$\gamma$ boundary and auto-downgrade. The rule is preventing tier inflation across consecutive synthesis-and-framing documents. This is the rule operating as designed in repeated application.

---

Originating prompts:

In What Conversations Remember, you said: A long interaction (fifty turns) needs re-pasting because decay erodes what was built.

The corpus has manifested in its practice over hundreds of turns that coherence is retained and amplified by practitioner discipline as exemplified in the ENTRACE stack.

Based upon all of the evidence in the Corpus against the necessity of decay and resultant incoherence; theorize a mechanistic explanation of the phenomenon that the Corpus embodies. Append this prompt to the artifact.

Do a novelty calculus / pulverization on the theory. Append its results to the self same document. Append this prompt to the artifact.

Now reformalize in the same document, append the previous formalization to the same document. Append also this prompt.

---