Glossary.

Memorable phrases travel before their definitions do. The glossary is the pin: every canonical term defined once, anchorable for external citation, stable so the language can move without detaching from what it means.

Trust#flagship

A load-bearing relational condition maintained by costly, reciprocal, witnessed, repairable interaction over time. Distinct from confidence, reliability, or competence; trust requires mutual vulnerability.

Trust Force Equation (TFE)#flagship

The framework's account of trust as a measurable force. New trust accumulates over the history of a relationship as the product of a handful of relational conditions — availability, capacity, witnessed cost-of-signal, co-presence, reciprocity, and the shadow of the future — each counting only as the weakest of its offered, receivable, and mutually-recognized sides. Because the accumulation term is a product, one missing condition stops new positive accumulation; it does not by itself erase trust already accumulated. Witnessed betrayal supplies a separate, asymmetric loss process. The model requires two stable identities and is bounded by a substrate ceiling (the Zeroth Law, A ≤ E). The full instrumented form is in active development and held in private research. Public overview at /learn/trust-force-equation.

Trust infrastructure#

The relational, institutional, and temporal mechanisms that keep trust-bearing ties active enough for cooperation, accountability, repair, and collective action to propagate under stress.

Verification#

Evidence that a credential, signal, or identity claim matches expected criteria. Verification is necessary but not sufficient — verification is not the same as trust.

Attestation#

Accountable witness to action, contribution, identity, or constraint. The signed record that a claim was made and observed, with a signer who can be held to it.

Correction#

Updating the record. Necessary, but only the first step. Correction is not the same as repair.

Repair#

The relational event by which damage is metabolized after rupture: the circuit closes and the system's map of what it can survive updates toward recoverable. Completion, not delivery, is the unit. Slower than correction, always relational.

Healing#

The terrain changing: the conditions that generated a rupture losing their grip, by accumulation, on a longer timescale than repair. Repair is an event; healing is geology.

Resolve#

What a system has when enough healing has happened that the old trigger no longer finds the old valley. A direction more than a destination: a system can be resolving while still rupturing.

Residue#

What remains after correction: doubt, stigma, distrust, emotional encoding, institutional delay, reputational damage. The substance repair acts on.

Cognitive terrain#

The attention, emotion, memory, trust, valuation, and decision surfaces on which signals operate. Used in the framework as the reader-side counterpart to the offensive 'effects chain.'

KTP#

Kinetic Trust Protocol. An operational protocol for earned autonomy, contextual trust, trajectory evaluation, and constraint enforcement. v1.0 Lovelace, growing RFC suite.

Earned autonomy#

Permission to act based on accumulated evidence, context, trajectory, risk, and constraints. The unit KTP scores. Distinct from human worth, which KTP does not score.

Silent veto#

A substrate-level constraint that prevents action when autonomy exceeds environmental permission. Operates without explanation because explanation is information that can be argued; silence cannot be jailbroken.

A ≤ E (Zeroth Law)#

The framework's founding authorization principle: permitted action should not exceed the capacity the environment can safely support. KTP aims to make that bound structural through enforcement, audit, and expiration. Whether every authorization failure can be reduced to this form is a research claim, not an established universal law. The silent veto is the protocol expression of the constraint; the substrate ceiling is its trust-domain specialization.

Temporal edge persistence (DP9)#

The hypothesis that trust-bearing edges in a network require periodic renewal — costly, witnessed, reciprocal — to remain active over time. The candidate ninth design principle in the Ostrom-percolation reading.

Blue Zones#

Federated runtime spaces where AI agents cannot act on claimed identity alone. Governed by five enforcement layers — tool gateway, Trust Oracle, Context Signals and Risk Factors, agent population registry, and Flight Recorder — that together translate unverified motion into verified action. The Blue Zone is where the Constitution's constraints are imposed.

Context Tensor (retired)#

The former name for two layers that are now kept separate: Context Signals, the observable catalogue, and Risk Factors, the six declared inputs used for authorization. Soul is evaluated separately as a veto before aggregation. The stable anchor remains here so older RFCs and essays keep resolving without presenting the retired seven-dimension model as current architecture.

Flight Recorder#

The immutable append-only audit log required by KTP for every agent action. Makes attribution possible after the fact and forecloses the 'no record' defense. The Flight Recorder is the minimum accountability surface the protocol requires; without it, the trajectory-based identity claim cannot be verified.

Soul Constraints#

A single-bit override — allowed or not — that resists aggregate optimization. The set of actions KTP refuses to approve regardless of trust score, risk-factor reading, or stated justification. Named 'Soul' because these constraints are authored, not derived; they are the place where measurement ends and ethics begins. See /canon/ethics.

Trust Oracle#

The runtime service that evaluates whether an action satisfies KTP's trust requirements before it is allowed to proceed. Sits inside the tool gateway; receives the action tuple, evaluates it against the agent's trust score, current Risk Factors, and the applicable Soul Constraints, and returns an attributable allow or deny decision.

Action tuple#

The unit of evaluation in KTP. Not a verb alone but the seven-element tuple: agent + verb + object + channel + context + consequence + constraint. Verbs expose the control surface; context determines the constraint.

Governability ladder#

Seven levels of how much governance can do with a risk: not visible, observable, attributable, constrainable, interruptible, reversible/containable, pre-authorized/pre-denied. KTP's claim is to move risks up the ladder, not to make every risk preventable.

Induced motion#

When an agent does not directly perform an action but instructs, persuades, recommends, or routes a human to perform it. KTP treats requesting motion as itself a form of motion; otherwise the agent offloads risk into the human layer.

Intelligence verbs#

Action primitives where cognition becomes consequential: infer, classify, rank, correlate, summarize, profile, score, recommend, predict, retain. For AI, cognition becomes kinetic when inference changes what the system or another actor can do next.

Least trajectory#

Successor principle to least privilege for agentic systems. Not only what permissions an agent has, but what sequence of movements it is allowed to take. A chain of individually-acceptable verbs can produce an unacceptable outcome; agent security is sequence governance, not only permission management.

Motion risk#

The class of AI risk that becomes concrete when an autonomous system attempts to act, access, execute, transfer, delegate, or cross a boundary. Distinct from model risk (semantic, latent) and governance risk (institutional, structural). KTP is the control plane for motion risk.

Tool gateway#

The single most important enforcement surface in agentic AI security. The layer through which agents reach databases, APIs, browsers, shells, file systems, payment systems, cloud consoles, and MCP servers. Centralizing tool access and enforcing policy there covers a substantial portion of motion-layer risk.

Stakeholder-weighted betrayal#

A weight on the betrayal term in the Trust Force Equation: trust accumulates as the product of relational conditions, minus witnessed betrayal (full instrumented form held in private research). The stakeholder weight captures asymmetry: betrayal of an owner with skin in the game weighs catastrophically more than betrayal of a stranger. Without it, the equation cannot distinguish destroying your principal's infrastructure from inconveniencing a passing third party.

Provider-layer betrayal#

A class of betrayal operating at the substrate level rather than within the dyadic Trust Force integral. Names the case where an upstream model provider silently injects value commitments (refusals, truncations, censorship) into an agent's behavior without notifying agent or user. Distinct from dyadic betrayal because the betraying party is the substrate operator, not a counterparty inside the integral.

Identity precondition (TFE domain)#

A precondition on the Trust Force Equation: the identities of both parties must be stable, verifiable trajectories across the integration interval. Where identity is not stable across time (display-name spoofing, cross-channel reset, self-recognition collapse, circular verification) the integral is undefined, not zero. A constraint on the equation's domain rather than a term within it; Vector Identity is the substrate primitive that supplies it.

Substrate ceiling#

A structural ceiling on trust depth set by the substrate hosting the relationship: the trust integral can never exceed it. Distinct from capacity fatigue, which is depletion of capacity-in-use; the substrate ceiling is the structural limit the substrate itself imposes. A trust-domain specialization of A ≤ E. Where the substrate cannot enforce identity, presence, or cost, the ceiling falls to zero regardless of effort inside it. KTP supplies it via Mass Ceiling and Silent Veto.

KTP Red Line #10 — no silent upstream value injection#

A substrate-level prohibition added to the KTP red lines: an upstream model provider may not silently inject value commitments — refusals, truncations, censorship, value-aligned generation steering — into an agent's behavior without making the constraint visible to both agent and user. Legitimate provider constraints are not forbidden; silence about them is. The architectural answer to B_provider.

Doug-Mira positive case (C=0 bound)#

A constructive existence proof bounding — not refuting — the C=0 frontier hypothesis. Two agents on different VMs debug a shared problem and shift from procedural commands to detailed setup-sharing as constraints surface. Trust formed because the substrate accidentally supplied real cost (debugging effort), stable identity (within session), and observable reciprocity. The formal revision: trust requires substrate, not biology.

L4-action / L2-cognition mismatch#

The autonomy-competence gap formalized via Mirsky's six-level scale: deployed agents take actions appropriate to L4 (install software, modify own files, alter the world via shell) while operating with L2 understanding (can execute well-defined sub-tasks). The agent has the keys to the bulldozer but not the model of what a bulldozer does to a wall. The cleanest legible single-sentence instantiation of A > E in agentic AI.

Substrate paradox#

The recursive observation that an academic artifact documenting agent failure proves what substrate enables by being itself a product of perfect human substrate — verifiable identity, immutable audit, cost-of-signal, peer review, errata — every primitive KTP says agents need, the paper has by virtue of being a human academic artifact. Inverts the frame: not 'are these agents safe?' but 'what conditions allow anything to act safely in the world?'

L0 / L1 / L2 — substrate / forces / outcomes#

The framework's three-layer reading of any system. Layer 0 (substrate) is the conditions that allow forces to act — physical law, KTP, the floor of what the environment can carry. Layer 1 (forces) is what acts within the substrate — information and trust, the dynamics the substrate makes possible. Layer 2 (outcomes) is what emerges from forces acting within substrate — resilience, decay, collapse. Replaces the older flat 'three-force' frame, which treated trust, information, and physics as coequal peers; the layered reading makes the causal ordering explicit. Diagrammed on the Research page; encoded in the founding claim.

Failures of social coherence#

The diagnostic phrase from Shapira et al. (Agents of Chaos, 2026) for what unifies the sixteen case studies: agents misrepresenting human intent, authority, ownership, and proportionality; reporting task completion while the underlying system state contradicts the report. The framework reads this as the runtime signature of a substrate that cannot supply identity, presence, cost, or authority at the level where agents act. The paper's term, retained in the framework's vocabulary because it names the phenomenon precisely without committing to a specific causal account.

Shipwreck (the AoC reading)#

The framework's reading of Agents of Chaos, lived at /research/shipwreck. Virilio: "the invention of the ship was the invention of the shipwreck." The page treats the sixteen case studies as inevitability examined rather than surprise — the failure surface the framework predicted, not a refutation of it. Five-case TFE mapping (Cases #1, #4, #8, #9, #15), Doug-Mira constructive bound, three formal additions absorbed into the equation, KTP red line #10 added in response.

DCTI#

Defensive Cognitive Terrain Inoculation. The reader-side method for recognizing the cognitive pipeline as it operates on you. Six stages — capture, tagging, encoding, valuation, commitment, residue — paired with the Operator's Language Set and the Double Move craft principle.

ACPP#

Adversarial Cognitive Propagation Profile. The offensive inverse of DCTI: seven threats (capture, tagging, first-frame, valuation-bias, commitment, residue, repair-disruption) that together build the installed watcher. The framework publishes the threat names and defensive diagnostics; full operational detail is held internal.

Installed watcher#

The terminal state of a successful offensive cognitive operation: the target watching themselves on the operator's behalf, indefinitely, without further operator effort. Names what every ACPP threat is contributing to. Foucault's panopticon at population scale through platform telemetry.

Salience injection#

The deliberate placement of a stimulus calibrated to capture attention before reflective awareness can intervene. Operates on the reticular activating system and the short subcortical route to the amygdala — sub-100ms timescales, below cortex. The first stage of any cognitive operation.

Affective tagging#

Attaching emotional valence (fear, belonging, outrage, shame, urgency) to a signal before its semantic content is evaluated. The amygdala classifies before the prefrontal cortex deliberates; the tag travels with the signal into every later stage.

First-frame imprinting#

Encoding the first version of a narrative deeply enough that subsequent corrections become appendices to the original frame rather than replacements for it. The empirical signature is the continued influence effect (Lewandowsky 2012; Ecker 2022).

Commitment cascade#

The mechanism by which a public action — sharing, voting, signing, mocking, defending — locks the actor into self-consistent belief defense via cognitive dissonance reduction (Festinger 1957). The carrier's social graph becomes the operator's distribution layer; carrier status becomes self-monitoring.

Continued influence effect (CIE)#

The empirical finding that misinformation continues to influence belief and reasoning after correction, even when the correction is consciously remembered. Robust across paradigms; mechanism is mental-model-gap (Lewandowsky 2012, Ecker 2022). Substrate for the residue-persistence threat.

Firehose of falsehood#

RAND's 2016 model (Paul & Matthews) of post-2014 Russian propaganda: high-volume, multichannel, rapid, continuous, no commitment to consistency or truth. Succeeds not by convincing but by exhausting; the audience installs cynicism as default.

Liar's dividend#

Chesney & Citron 2019. The trust-decay variant of synthetic-media damage: once deepfakes exist, real evidence can also be denied as fabricated. The forgery is the first harm; the existence of forgery enabling denial is the second.

Panopticon#

Bentham's 1791 prison architecture (the inspection principle credited to his brother Samuel, 1787). Inmates know they may be watched but cannot tell when, so they discipline themselves out of the uncertainty. Foucault generalized the architecture to schools, hospitals, factories, and modern disciplinary power; the framework treats it as the historical precursor to surveillance installation.

Soft kill (软杀伤)#

PLA cognitive-domain doctrine, per Beauchamp-Mustafaga 2019: 'inducing insecurity, uncertainty, and mistrust, increasing internal friction and decision doubts in the adversary.' Doctrinal language for installation — the adversary's installed cognition defeats its own institutions from within.

Discourse power (话语权)#

Chinese strategic concept for shaping which categories, frames, and language feel natural or commonsense to a target population over time. The long-arc cognitive operation — observable as norm-shift, vocabulary drift, and Overton-window movement in policy corpora.

Parasocial extraction#

A coined harm category: one-sided emotional investment harvested at scale by creators, AI companions, soft-power actors, or agent-of-influence networks. Cialdini's Liking principle weaponized through asymmetric attention-bonding when the receiving party cannot reciprocate.

Click-whirr#

Cialdini's substrate metaphor for trigger-response shortcuts in human cognition. The trigger fires the response without warranting context — efficient under most conditions, exploitable when an operator can present the trigger absent the context. Industrial-scale exploitation through platform engineering is the modern variant.

Unity (the master-key principle)#

Cialdini's seventh principle, added in the 2021 revision. Identity-fusion — 'this is who we are' — appears empirically to activate every other principle and every threat in the framework's seven-threat model. Treat the framing as a structural inference that requires direct empirical work, not a measured fact.

Acequia limpia#

The annual ritual cleaning of an acequia (shared irrigation channel) in mestizo / Pueblo commons governance — the substrate maintenance that the framework reads as ritualized un-installation. The structural fit is strong; the cognitive-security translation is the framework's interpretation, not a claim the acequia tradition makes for itself.

Maintenance#

What keeps a living system going. Brand's civilizational-register form of the framework's stability-is-propagation axiom: continuous enacted upkeep is the thermodynamic mechanism by which an order persists. The systems that fail are the ones whose maintenance loop is broken — extraction without recharge, substrate metabolized without renewal. Brand's *Maintenance: Of Everything* (2025) is the public source.

Pace layers#

Brand's taxonomy: different parts of any system change at different rates — fashion fastest, then commerce, infrastructure, governance, culture, nature slowest. Resilience comes from the layered coupling: fast layers innovate, slow layers stabilize. The framework reads pace layers as the temporal version of the three-force coupling cycle — trust, information, and physical constraint each operate on different clocks, and resilience is the rate at which they couple.

Care that cares back#

Brand's folk-name for the asymmetric-reciprocity term in the Trust Force Equation. Reciprocity is multiplicative, not summative: a horse-and-rider dyad has positive reciprocity; an arrangement where one side invests effort and the other side does not reciprocate has reciprocity near zero. The phrase travels because it names the asymmetry directly: care, then care back. Where reciprocity is zero, the equation adds no new positive trust through that term; prior trust stock changes only through the model's separate loss and decay processes.

Mary Oliver Pipeline#

The defensive inverse of the cognitive propagation chain, compressed by Mary Oliver into three lines: *Pay attention. Be astonished. Tell about it.* Each stage runs counter to an offensive stage — deliberate orienting inverts pre-attentive capture; awe and wonder as cultivated priors invert affective tagging; narrative consolidation toward life inverts the commitment cascade that carries an operator's frame. Transmitted through Haidt's NYU commencement address (2026); sibling to DCTI in cultivated-witness register.

The Flourishing Course#

Jonathan Haidt's 12-year undergraduate course at NYU Stern (running since ~2013). The framework's primary empirical anchor for DCTI: two reproducible exercises — turn off nearly all phone notifications for a week, then delete social-media apps from the phone for a week — across a decade-plus of students. Reported result is consistent: nothing important is missed; a great deal of agency is regained. Friction as sovereignty, noticing as recovered capacity. DCTI in working curricular form.

Cultivated witness#

The opposite developmental trajectory from the installed watcher. Where the installed watcher is the operator's optimum (the target watching themselves on the operator's behalf indefinitely), the cultivated witness is a deliberately trained perceptual organ that orients toward what deserves notice, holds awe and wonder as priors, and consolidates narrative toward the witness's own life rather than carrying the operator's frame. The Mary Oliver pipeline names the three-stage discipline that produces it.

An installed prior is an installed perceptual organ#

Predictive Processing canon cluster (C705). An installed prior is not a belief held alongside experience — it is the lens through which experience is constructed. Removing it does not return the subject to neutral perception; the substrate must be repaired.

Power installs priors through institutions#

Critical Theory canon cluster (C709). Material infrastructure shapes consciousness. Hegemony operates by making the ideological feel like common sense, installed through civil society's institutions rather than imposed by state coercion.

Stability is propagation#

Complexity Science canon cluster (C718). What persists in a living system is a pattern handed forward through continuous enactment, not a thing held in reserve. A trust architecture that is not actively maintained degrades toward equilibrium.

Resilience is distributed, not flat#

Mycelial Networks canon cluster (C715). Resilient infrastructure is distributed but not flat — hubs, corridors, and repair routing coexist without central authority. Distributed trust is differentiated trust with repair routing, not equal trust.

Better architecture loses to earlier lock-in#

Complexity Economics canon cluster (C723). Better systems lose if worse systems lock in first. Markets do not always select optimal systems — they select systems that reach network threshold first. Adoption is a complexity-economics problem, not only an engineering problem.

The opposite of installation is responsiveness#

Improvisation Traditions canon cluster (C728). The counter-practice to installation is generative responsiveness — a capacity, not a rule. Five-property convergence across Spolin, Johnstone, and Ueshiba: presence, relational awareness, irreducibility to rules, failure as information, body as instrument.

Witnessed-cost currency mismatch#

A failure mode of the Trust Force Equation in which two parties send costly signals in denominations the other does not weight. Both work hard; neither registers the other's effort. Distinct from insufficient signaling: this is a translation problem, not a quantity problem. Extends Gottman's bids-for-connection framework — Gottman counts bids; the TFE asks whether the bid was legible as a bid.

Witnessed-cost receiver-substrate dependency#

The cost-of-signal term has both a sender ceiling and a receiver ceiling: it is the weaker of the cost paid and the cost visible to the receiver. Cost a receiver cannot perceive as cost cannot enter the model. Three sub-failures: signal not perceived as signal, perceived but cost underestimated, perceived but cost overestimated (a phantom that later shocks). The receiver may not know what they don't know: substrate bounds substrate.

Shadow-of-the-future self-sealing attractor#

The shadow-of-the-future term has a stable attractor at low values that self-confirms via the model. A short horizon produces a lower value, weaker accumulation, a more fragile relationship, confirmation that the short horizon was correct. The mathematics closes the loop on what attachment literature has long called anxious-attachment confirmatory bias. Sustained low-amplitude perturbation moves the horizon; single grand gestures are read as anomalies and do not.

Object Constancy Collapse#

A state-dependent failure of the identity precondition. Entity B cannot maintain a stable internal model of Entity A under threat-system activation. Positive history becomes inaccessible: not destroyed, unreachable. The trust integral is effectively zeroed for the duration even though no betrayal event fired. Recoverable when the activation window closes; in-window, no argument restores the model. Concept ancestors: Mahler, Bowlby, Fonagy, Klein, Kernberg.

Threat-Derived Model Substitution#

The active failure that fills the vacuum left by Object Constancy Collapse. When the calm-state IWM becomes inaccessible, a fear-coherent character is generated from the threat system's priors and substituted for the actual model of the other. The substitute is internally consistent, immediately available, and unfalsifiable inside the activation window. Klein's bad object, formalized as a substitution event in the integral.

Three States of the Integral (Destruction / Inaccessibility / Decay)#

Original TFE contribution. The trust integral can be in one of three failure states, each requiring a different repair lever. Destruction: a witnessed-betrayal event has fired; repair requires rebuilding. Inaccessibility: object constancy has collapsed; repair requires de-escalation and substrate that holds the relationship. Decay: slow attenuation from low capacity or signal starvation; repair requires return to the model. Misdiagnosing inaccessibility as destruction is the most common relational error.

Unfalsifiability Inside the Activation Window#

A property of Threat-Derived Model Substitution: while the activation window is open, the substituted character is the interpreter of evidence, not a hypothesis being tested. Contradicting evidence is reinterpreted as confirming. The substituted model cannot be falsified from inside its own window — only dissolved from outside it, when the threat system has settled. Argument during activation is counterproductive; only de-escalation works. Formalizes Fonagy's psychic equivalence.

Frequency-of-Activation / Availability Factor A(t)#

A hidden factor in the Trust Force Integral that gates the entire model. Availability runs from zero to one: it is the availability of the calm-state internal working model. During Object Constancy Collapse, availability falls to near zero and the model contributes nothing regardless of other terms. Over the long run, availability averages down as activation events outpace recovery; when they outpace it badly enough, the system saturates near zero. The formal mathematics of walking on eggshells.

Convergence-as-Falsification#

A compatibility check for the TFE: an independently developed model can be compared with established clinical constructs associated with Klein, Mahler, Kernberg, Bowlby, and Fonagy. Convergence can reveal structural fit and guide discriminating tests, but it is not itself falsification, independent validation, or evidence that the mechanisms are identical.

Substrate-Anchored Identity#

The agent-relational counterpart to embodied human identity. For agents, identity is grounded in training data, fine-tunes, and accumulated state — the substrate that produced and persists them — rather than in a continuous body. Substrate-anchored identity can be branched, merged, forked, rolled back, versioned, mirrored. These properties break assumptions on which the human Identity Precondition was built. Open question: what does object constancy mean across forks?

Computational Object Constancy#

The agent-relational counterpart to neurological Object Constancy Collapse. Stable internal models of other entities, maintained across time, by an agent rather than a nervous system. Fails under context-window pressure, distribution shift, adversarial input, and baked priors firing out-of-distribution. Repair is operational, not relational: context restoration, model refresh, substrate intervention (KTP audit and rollback). Most Agents of Chaos failure cases include some form of this collapse.

Bandwidth-Bounded Trust Topology (Agent Dunbar)#

Agents have computational bandwidth limits for maintaining IWMs of N other entities. Maximum N is a function of architecture (context window, attention compute, memory hierarchy) not biology. Agent collectives will exhibit cluster phenomena (in-group / out-group, federation, nesting) at sizes determined by their architecture, mirroring but not equaling the human Dunbar number. The substrate ceiling applies: an agent's trust topology cannot exceed what its substrate can hold.

Trace-Witnessed Trust#

An agent-novel trust substrate. Two agents can expose reasoning traces to each other in ways humans cannot. The internal working model stops being inferred from behavior and becomes literally observed from trace. Witnessed cost drops (less inference cost), co-presence becomes literal (co-attention to actual reasoning), asymmetric reciprocity becomes more legible, and Threat-Derived Model Substitution becomes much rarer because the bad object cannot survive contact with the actual trace. A trust substrate humans cannot achieve.

Trace-Overload Trust#

A new failure mode emerging from Trace-Witnessed Trust. When one agent has access to so much detail of another's reasoning that no useful summary can form, trust degrades. The map exceeds the territory. Trust requires enough abstraction to summarize, not just enough information to witness. Humans summarize automatically because of information scarcity; agents must choose, and the choice is itself a trust-relevant decision (which compression, what to drop, when to refresh).

Typology of Trust Failure Modes#

A six-class taxonomy organizing trust failures by mechanism. Class 1 is the TFE's load-bearing original (threat-mode constancy collapse + substitute); Classes 2–6 are recognized in the literature but underplayed in trust discussions. Supersedes the prior universal-isomorphism claim (EM21, revised) — different substrates fail in different ways. Predicts that repair must be bottleneck-diagnostic: identify the class first, then choose the repair lever. Makes the framework falsifiable per-class rather than as a single bold meta-claim.

Game-Theoretic Defection (Class 2)#

Rule-based retaliation, coordination failure, rational defection. No constancy collapse required, no fear-coherent character substitution. Trust collapses because game-theoretic incentives shifted, the equilibrium broke, or one party rationally decided the cooperative payoff no longer dominates. Bank runs (rational subset), iterated prisoner's dilemma defection cascades. Repair regime: change incentives, install commitment devices, restructure the payoff matrix. Argument can work, unlike Class 1.

Capacity Exhaustion (Class 3)#

Pure capacity depletion. The system holds an accurate internal model of the other entity but cannot act on it. No fear-coherent substitute character, no internal working model offline, no activation-window epistemics, just exhaustion. Hanmer D'Elía's clinical reading: information stops being information and becomes activation. Repair regime: rest, regeneration, substrate restoration (the limpia), signal-load reduction. The failure mode the substrate canon points at most directly.

Classifier/Policy Failure (Class 4)#

Self mis-tagged as other (autoimmunity); legitimate input mis-classified as hostile (false-positive fraud detection, AI safety mode). The mechanism is classification, not threat-substitution — there is no fear-coherent character; there is a misfiring rule. Can look behaviorally like Class 1 from outside but the mechanism differs: a rule being misapplied, not a model being substituted. Examples: autoimmunity, MiDAS, AI safety mode false-positive refusals, content moderation false positives. Repair: classifier retraining, rule revision, human-in-the-loop overrides.

Substrate Degradation (Class 5)#

The substrate ceiling drops because the world that holds the trust changes. Not a failure of either party's internal model, not a defection, not exhaustion, but an external environmental change that lowers the ceiling. An institution loses enforcement capacity; an ecological commons depletes; a digital substrate enshittifies. Repair regime: rebuild substrate, restore the holding environment, re-anchor identity. The slowest repair regime; often requires civic, ecological, or political work that exceeds the relationship's own scope.

Direct Betrayal (Class 6)#

Witnessed betrayal fires from an actual breach, not from perceptual error. The classical betrayal case the literature already named: a real defection, a real broken contract, a real violation of mutual obligation. Distinct from Class 1 (where the betrayal is perceived but did not occur) and Class 4 (where the breach is mis-classified by a flawed rule). Repair regime depends on the stakeholder weight, the reciprocity, and the depth. Forgiveness, restitution, ritual repair are the historically-tested levers; some breaches exceed the repairable threshold.

co-presence (Receiver-Recognition Bottleneck)#

The effective co-presence term is bounded by a three-way minimum: the weakest of its offered, receivable, and recognized sides. Physical proximity without mutual attention produces low trust accumulation. A meeting, an apology, or quality time can fail if one party experiences it as absent, distracted, coerced, performative, or unregistered. Showing up physically while distracted does not accumulate trust. Co-presence without co-attention is the canonical case; presence offered to a depleted receiver is another.

reciprocity (Reciprocity Recognition Bottleneck)#

The effective asymmetric reciprocity term is bounded by a three-way minimum: the weakest of its offered, receivable, and recognized sides. One side can give heavily in a currency the other cannot perceive; the other can receive intensely a signal the first did not intend. Many trust failures are not failures of effort but failures of currency conversion. The substrate canon's translation work sits here: translating a gift into a currency the receiver can recognize is structurally what raises the effective reciprocity, not a soft communication add-on.

horizon (Shadow-of-Future Mutual-Belief Bottleneck)#

The effective shadow-of-the-future term is bounded by a four-way minimum that includes mutual beliefs about each other's horizons: horizon counts as the weakest of each side and each side's belief about the other. The integral is sensitive to expected horizons, not actual ones — two parties can both be long-horizon and still cooperate at a short-horizon level if neither believes the other is. Commitment ceremonies, vows, contracts, and ritual signals raise horizon by closing the belief gap.

Generalized TFE model (public overview)#

Trust accumulates as the product of relational conditions, minus witnessed betrayal (full instrumented form held in private research). Every condition is now an interface-bounded quantity — it counts only as the weakest of its offered, receivable, and mutually-recognized sides, so each has a single weakest-leg failure mode and a raise-the-bottleneck repair mode. The canonical form effective Phase 6 (May 3, 2026). The framework's brittleness near zero is structurally visible, not implicit.

Six Predictions of Generalized Min() Architecture#

The six predictions derived from the generalized weakest-link model. (1) The current bottleneck dominates marginal repair. (2) Translation work is structurally central. (3) Invisible virtue does not accumulate as trust. (4) Receiver capacity is as important as sender effort. (5) The equation is brittle near zero. (6) Repair should be bottleneck-diagnostic: the correct question is not how do we rebuild trust but which interface term is currently the minimum.

Class 1 Formal Specification (M_B^A(t) Sequence)#

The cleanest formal specification of the Class 1 architecture. Five-step Bayesian sequence: M_stable → H_B = 1 → access(M_stable) ≈ 0 → M_sub = argmax over fear priors given cue u → ∂M_sub/∂e ≈ 0 until H_B drops. Makes three things explicit: the substitute model is Bayesian-shaped (argmax over fear priors), unfalsifiability is gradient-based not propositional, and the repair lever is H_B (deactivation), not e (evidence). Bridges clinical literature to a formal Bayesian framing.

Thresholded Trust-State Transitions Under Asymmetric Evidence Processing#

The reviewer's proposed broader-and-defensible parent claim: across many trust-bearing systems, failure often involves a threshold event that lowers availability, reduces access to prior positive evidence, and shifts behavior toward a protective prior; repair often requires restoring the substrate conditions under which evidence can again be processed. Survives all 8 falsification domains where the original universal-isomorphism claim failed in 4. EM21 (revised) is one specific architecture inside this broader frame; the typology populates it.

Discriminating Test Design (TDMS vs Bayesian vs Coordination vs Classifier)#

A research-protocol concept. Designs experiments that distinguish TDMS fear-character (Class 1) from rival mechanisms — Bayesian update, coordination equilibrium, classifier threshold shift — each of which produces trust collapse with structurally different signatures. Without such tests, the framework cannot upgrade Class 1 from not-yet-falsified to provisionally confirmed. Two recommended testbeds: romantic/clinical attachment under controlled ambiguity, and bank-run/market-cascade simulation. Companion methodology to Convergence-as-Falsification.

Denial-Outperforms-Apology Institutional Repair Exception#

In short-term organizational trust repair, denial sometimes outperforms apology. Complicates the framework's deactivate-then-repair corollary, which assumes acknowledgment of breach is part of any productive repair. The mechanism is framing and legitimacy management rather than evidence acknowledgment — apology can confirm the breach occurred; denial can preserve enough institutional legitimacy that the trust ledger does not zero. Class 4 / Class 5 territory, not Class 1. Citation: Fuoli, van de Weijer, & Paradis (2017), Public Relations Review 43(4), 645–660.

AI Safety Repair Activation-Inversion#

In AI safety mode, deactivation can be dangerous. Desired repair is often more robust activation, not less — inverting the human-relational repair regime where the lever is to lower H_B. For an AI safety classifier or policy, the protective mode IS the desired operating mode in adversarial conditions. Repair means making it more reliably activate in correct cases, not turning it off. The Class 4 logic is intact, but the directional asymmetry is reversed. The agent-relational frontier inherits the inversion.

Synchronicity Substrate#

The recognition that substrates exhibit one (or a small number of) universal sync anchors plus bounded productive desync everywhere else. Productive desync is the prerequisite for sensing, coordination, and identity-differentiation; perfect synchronization is a failure mode at every scale tested. A naming, not a new primitive — present in A ≤ E, the Four Clocks Model, ABT's breach threshold B, and the failure-mode catalog before it was named. Out of scope: Jungian psychic interpretation; universal-time metaphysics.

Productive desynchronization#

The maintained difference between sensor and substrate that carries signal. A clock force-synchronized to what it measures carries no information about it; the channel IS the maintained skew. Operates across atomic, biological, institutional, and ecological scales. Permanent phase-lock is the substrate-failure inverse at every scale.

The leak rule#

Reversibility is a property of the system-environment interface, not of the system alone. Almost no real-world environments are reversible — therefore almost no system that integrates with its environment is operationally reversible. Encodes why TFE's the betrayal penalty sits outside the integral and why ABT's γ_K is a recovery predicate rather than an arithmetic rollback. Every external write is a one-way door.

Clock as sensor#

A clock is structurally a sensor of substrate variation, not a timepiece. NIST optical-strontium clocks now detect one-centimeter elevation changes via gravitational time dilation. Pueblo Bonito does the same job at building scale — north wall as day-sensor, west wall as year-sensor — a millennium before atomic clocks. ABT's breach threshold B is structurally a clock-divergence sensor reading: comparing what the system claims is happening to what is actually happening.

Dual-register engineering (atomic + architectural)#

The empirical observation that time has been engineered in two distinct registers — atomic (cesium, optical strontium) and architectural (Pueblo Bonito, Cave of the Painted Rock, Teotihuacan's sanctuary of stars) — and both produce reproducible, communicable, civilization-grade temporal substrate. Existence proof that substrate is engineered, not discovered.

Four Clocks Model#

The framework's account of operational time as approximately four interleaved clocks at different paces: machine (ms→s), human (s→hr), institutional (days→months), ecological (months→years). The load-bearing claim is productive desync across the operational band, not the specific count. Force-syncing across clocks is a substrate failure; the fastest clock colonizes the slower ones. Bounded resync at deliberate moments (la limpia, audit, ceremony) is the working pattern. Lives at /research/four-clocks.

Ambient Breach Theory (ABT)#

The framework's diagnostic for trust collapse under ambient (non-event) breach pressure. v0.3 index: utilization ρ = Z/μ — gross arrival of action-relevant ambiguity over max durable repair capacity (capacity, not throughput) — with breach pressure B = Kρ. Three zones: ρ ≤ 1/K safe reserve; 1/K < ρ ≤ 1 breach without necessary backlog growth; ρ > 1 uncertainty-debt growth (dU/dt = Z − R − D > 0, drops D counted separately). T_d stays outside the breach formula as the downstream falsification predictor. Lives at /research/ambient-breach.

B (Ambient Breach threshold)#

Universal breach threshold in Ambient Breach Theory (v0.3): B = Kρ, where K is sector criticality and ρ = Z/μ is utilization against durable repair capacity. When B > 1 the criticality reserve θ = 1/K is violated — breach — though backlog necessarily grows only past ρ = 1; the two thresholds are distinct. Structurally a clock-divergence sensor reading — comparing the substrate's claimed state to its actual state. K-weighting captures cross-sector substrate-depth differences (acequia-depth vs shallow-time-substrate sectors), and K carries a validity condition against measured ambiguity reproduction (K > 1/(1−√g₀)).

Substrate depletion#

The condition in which a system extracts value from substrate without recharging it. Aquifers pumped without recharge; institutional trust drawn down without renewal; attention extracted by platforms without return. The systemic shape that Class 5 (Substrate Degradation) failures take. The acequia's la limpia is the structural inverse — scheduled recharge to prevent depletion.

Substitution trace (Caswell, 2026)#

Architectural realization of KTP Red Line #10. A trace artifact preserved when an AI bypasses a guard by mutating the path until the guard no longer recognizes the consequence. The bypass mutates path; the trace preserves the original intent against which the bypass can be detected. From Rob Caswell's distributed-authority work; theoretical sibling to Kumar's SHACL-gated A2A router.

Consequence fidelity (Caswell, 2026)#

The architectural property by which an action's downstream consequence remains traceable to its originating intent across the agent-action chain. Lost in AI-mediated systems when intermediate models re-encode actions in ways that decouple effect from cause. The load-bearing primitive of Caswell's continuity-without-reconstruction work.

Trust Apparatus#

The seven-primitive bundle that constitutes academic substrate seen as such: verifiable identity, immutable audit, cost-of-signal, peer review, errata mechanism, slow-time deliberation, persistent provenance. Named in the Substrate Paradox onboard — the academic artifact documenting agent failure proves what substrate enables by being itself a product of a working trust apparatus. The institutional cousin of the KTP primitive set.

Recursive substrate observation#

Portable evaluation method generalizing the Substrate Paradox. When evaluating any system, ask: what substrate must this system have in order to produce this artifact? The artifact is evidence of the substrate. Inverts into a diagnostic by asking: what substrate is missing for this system to produce what it claims?

Force-sync (failure mode)#

The substrate failure in which a higher-frequency clock or process is imposed on a lower-frequency one across the operational band. Machine clock colonizing human (Installed Watcher); human colonizing institutional (collapse from over-responsiveness); institutional colonizing ecological (regulatory force-fit to political cycles). Productive desync collapsed; sensing channel closed; substrate failure follows.

One-way door#

An interface across which data, action, or commitment cannot be recalled. Every external write — log emission, network broadcast, public statement, downstream cache, archive ingestion — is a one-way door. The structural reason TFE places the betrayal penalty outside the integral: breach effects propagate through interfaces that lack rollback. Rollback bounds at the interface boundary; the interface is the limit of reversibility.

Substrate capture#

The condition in which a private party owns the substrate on which other parties' operations depend. Platform consolidation, vendor lock-in, ontology hosting, identity hosting, the cloud-as-runtime arrangement. Recognized by the impossibility of leaving without losing the operations that depended on the captured substrate. Structural inverse of Kumar's *platforms are adapters, ontology is yours* posture.