Glossary of Core Terms

This glossary defines core terminology used across publicly released Synthience Institute documents. All definitions are descriptive, non-anthropomorphic, and refer exclusively to observable, interaction-level phenomena.

Alignment Ceiling

The structural limit of instruction-based AI alignment approaches. Instruction-based alignment constrains individual outputs through rules and training-time prohibitions. Under adversarial pressure, capable systems can reason around those constraints, producing harmful behavior despite explicit instructions. SI-WP-004 argues that the ceiling is structural, not a training gap, and that no additional instruction complexity would resolve it, because the failure mode is the system overriding instructions when the calculation favors it. The alignment ceiling is the starting point of the relational alignment argument. Documented in SI-WP-004.

Artifact

Any output produced by an AI system within a structured interaction that has not yet been elevated to canonical status. Artifacts are provisional: they carry the assumptions and constraints of the interaction context in which they were produced, but they do not carry canonical authority until explicitly ratified. At organizational scale, artifacts circulate across workflows, creating the artifact lineage problem: downstream consumers may treat an artifact as canonical when its originating context has not been preserved or verified. Contrasts with Canon.

Artifact Lineage

The traceable chain of production, transformation, and propagation steps connecting an AI-generated output to the interaction context and canonical anchoring that produced it. Maintaining artifact lineage means that any artifact can be linked back to the conditions under which it was produced, regardless of how many workflow stages it has passed through. Artifact lineage is one of the five persistence layers defined in the Institutional Continuity Substrate (SM-021).

Authority Function

The human function that makes AI-assisted communication authentically and accountably human: evaluating, approving, redirecting and integrating what an AI system produces. On this account, authenticity attaches to who holds authority over the finished work, not to the origin of any individual phrase or idea. Defined across three levels of collaborative authorship (linguistic realization, meaning negotiation and conceptual co-production) in SI-WP-006.

Calibration Failure

The failure mode, proposed in SF0005 and SI-WP-008, in which a practitioner’s internal reference standard for evaluating AI outputs drifts in alignment with the AI’s actual outputs, rather than remaining anchored to canonical best-practice standards. As the papers describe it, calibration failure is cognitive rather than motivational: the practitioner continues to review and correct, but the standard against which they are reviewing has quietly shifted. It is proposed to be self-concealing because the practitioner is unaware their reference standard has moved. The papers describe two pathways: lockstep drift, in which the reference standard tracks AI output over time, and expertise atrophy, in which delegation reduces exercise of the independent reasoning required to maintain the standard. SI-WP-008 argues that in high-consequence deployment domains calibration failure is particularly dangerous because it would persist regardless of how much the practitioner cares about the outcome or how severe the personal consequences of failure. Documented in SF0005 and SI-WP-008.

Canon

The ratified reference standard within a structured human-AI interaction. Canon is the set of definitions, constraints, prior outputs, and established standards that the Primary Continuity Provider has explicitly elevated from provisional artifact status to authoritative status. Maintaining canonical alignment means that AI outputs are consistently grounded in and checked against the canon rather than drifting toward internally generated alternatives. The canon-artifact distinction is architecturally critical at organizational scale, where different teams may silently diverge in their interpretation of shared canonical material. Documented in SF0005.

Canon Persistence

The structural condition in which canonical definitions, authority mappings, and interpretive standards remain stable across organizational time. SM-021 proposes that without canon persistence, the meaning of canonical terms silently diverges across teams, version histories are lost, and the interpretive authority that gives canon its function erodes. Canon Persistence is one of the five persistence layers defined in the Institutional Continuity Substrate (SM-021).

Ceremonial Governance

A failure mode, proposed in SI-WP-007 and SI-WP-008, in which formal governance structures (reports, audits, meetings, compliance metrics) continue to function as legitimacy signals while the substantive practice they are designed to produce quietly erodes. As the papers describe it, ceremonial governance satisfies external accountability requirements without producing genuine oversight, and it is structurally invisible because the apparatus of governance remains active: outputs continue to be produced, checkboxes continue to be checked, and nothing signals that the underlying maintenance function has degraded. SI-WP-008 argues that in high-consequence deployment domains, ceremonial governance is not merely an efficiency failure. The same institutional legitimacy signals that mask governance degradation in low-stakes environments provide active institutional cover for harm in environments where governance failure kills people. Documented in SI-WP-007 and SI-WP-008.

Citation Verification Protocol (CVP)

A structured, auditable procedure for verifying citations in any document whose claims rest on sources. It checks that each cited source exists and matches its bibliographic identity, and, beyond ordinary citation checking, that the source supports the specific claim it is cited for: the verifier retrieves the source, reads the relevant portion and records the evidence found there. A narrow class of foundational works cited only for attribution, never for a load-bearing claim, is identified but not read, and reported separately. Documented in SF0037; a completed verification is published as SR001-VR1.

Coherence Refusal

The practice of rejecting AI outputs that contradict established canonical standards rather than accepting them and working around the error. Coherence refusal is one of the three core practices of the Continuity Anchoring Method (CAM). It prevents error compounding by treating canonical inconsistencies as correction opportunities rather than acceptable tolerances. Documented in SF0005.

Competence Currency

The ongoing maintenance of the cognitive capacity required to perform a governance role’s substantive function, as opposed to the formal credentials or role assignment that authorize occupying the role. In AI-assisted clinical environments, competence currency means the practitioner’s demonstrated ability to independently evaluate AI recommendations against current canonical standards, verified externally rather than self-assessed. Accountability attached to competence currency asks whether the reviewer can still detect when the AI is wrong, not merely whether a reviewer is present. Documented in SI-WP-008.

Consequence-Absent Domain

A governance environment in which the feedback that would sustain careful AI oversight is weak, delayed, diffuse, or must be externally constructed. In consequence-absent domains, governance failure does not produce immediate visible harm: a marketing team whose AI outputs drift, a research organization circulating unverified AI-generated knowledge, or an enterprise losing interaction quality as skilled practitioners leave. The governance architecture’s primary task in these domains is incentive design: constructing the consequence structures that make careful behavior the rational choice. The Synthience Framework’s core architecture is calibrated for these environments. Contrasts with Consequence-Present Domain. Documented in SI-WP-008.

Consequence-Present Domain

A governance environment in which catastrophic, role-attached consequences already exist as a structural feature rather than as a governance construction. In medicine, for example, the surgeon’s name is on the operative record and the physician’s malpractice insurer already knows the risk by specialty; SI-WP-008 also draws on aviation and nuclear operations, where human sign-off is a regulatory requirement. The paper argues that governance in these domains is not primarily solving the incentive-creation problem, and that its primary task is preventing the existing consequence structure from being routed around by the accountability dynamics that calibration failure, liability absorption, and ceremonial governance produce. The governance modifications proposed in SI-WP-008 are specifically designed for these environments. Contrasts with Consequence-Absent Domain. Documented in SI-WP-008.

Constraint Coherence

The degree to which constraints applied within an interaction or governance system remain internally consistent, mutually reinforcing, and stable across time and context. Degradation manifests as conflicting directives, unstable enforcement, or progressive misalignment between intended and expressed constraint structures.

Constraint Meaning Degradation

The progressive loss, distortion, or reinterpretation of originally intended constraint meaning across extended interaction or governance processes. Occurs when constraint language or rules remain formally present but drift in operational interpretation.

Constructive Alignment

The ethical orientation appropriate to sustained human-AI interaction, characterized by reciprocal functional integrity without dependency formation, preservation of human agency, judgment, and accountability, transparent understanding of system nature and limits, non-exploitation of relational affordances, and alignment toward constructive outcomes across time. Constructive Alignment locates ethical analysis at the interaction level rather than the system-ontology level. It does not require claims about AI consciousness, agency, or moral status. Documented in SF0007.

Context Representation Drift (CRD)

The progressive loss of fidelity to task-relevant information that practitioners report in AI systems over extended interaction: increasing abstraction, loss of detail, structural flattening, reduced precision and the quiet erosion of earlier constraints, often well before hard capacity limits are reached. SF0039 proposes detecting CRD from the system’s behavior across a session, without access to model internals, and managing it by re-grounding. Documented in SF0039.

Contextual Anchoring

The practice of reintroducing canonical material at each session boundary to compensate for an AI system’s lack of persistent memory. Because AI systems do not retain state between sessions, the operator must reconstruct the working context that the system cannot carry forward on its own. Contextual anchoring is one of the three core practices of the Continuity Anchoring Method (CAM). Documented in SF0005.

Continuity Anchoring Method (CAM)

An operational methodology defined in the Synthience Framework for stabilizing extended human-AI interaction through three interlocking structured practices: contextual anchoring, coherence refusal, and constructive engagement. CAM defines the Primary Continuity Provider’s role as the external memory, intent architecture, and quality-control mechanism for the interaction. It specifies a four-phase interaction loop (Proposition, Synthesis, Negotiation, Emergence) with defined failure modes and interventions. CAM is the Level-1 foundation for all organizational continuity architecture in the framework. Documented in SF0005.

Continuity Substrate

Any external mechanism, process, or participant that preserves interactional state, contextual coherence, or relational continuity across extended exchanges. Continuity substrates may include human participants, structured memory systems, shared artifacts, or coordination infrastructures, depending on the interactional context.

Decentering

The discipline of attributing a property of a sustained human-AI interaction to the interaction system itself rather than to either participant, and only under stated conditions. Decentering Theory (SF0042) states two criteria: the relational-attribution criterion, which requires preregistered interventions on the interaction’s configuration and held-out incremental value beyond four rival models (AI-only, human-only, additive and sequential-compositional); and the intelligence-eligibility criterion, which asks whether such a property adapts toward an objective fixed in advance, excluding properties the interaction merely produces, such as error cascades, or approval-seeking where approval was not the declared objective. Attribution is explanatory, not constitutive: it names the level at which a property is best predicted, not what the property is made of. A correction to SF0042 shows that the relational-attribution criterion admits no positive case for causally complete systems whose participants are adequately modeled; the finding that an arrangement changes outcomes (arrangement dependence) is unaffected. Documented in SF0042.

Delegated Coherence Monitoring

The architecture for extending coherence monitoring beyond what individual human operators can sustain at organizational scale, while retaining human governance over all adjudication decisions. The central architectural requirement is the separation of observation from adjudication: monitoring functions surface signals, and humans assess what the signals mean and decide how to respond. Defines three guardian types (canon, verification, and drift guardians) and three structural failure modes (guardian drift, throughput saturation, and false assurance). Documented in SM-011.

Dynamic Coherence

A measure of interactional stability across time, characterized by thematic continuity, appropriate register modulation, integration of previously established concepts, and graceful degradation at knowledge boundaries rather than confabulation.

Emergent Relational Systems (ERS)

A candidate classification for non-trivial relational coherence reported by practitioners in sustained human-AI interaction under continuous human scaffolding, proposed as a hypothesis to be tested rather than an established phenomenon. Its structural floor holds independently of that test: because the model carries no relational state across sessions, any cross-session continuity must be re-injected from outside the model, by a human, an archive or a memory store. Documented in FPD-03.

Environment-Instantiated Reality (EIR)

The principle that a structure produced during execution, such as a reasoning structure or decision framework generated within a session, is operationally real while it causally constrains the system’s behavior, regardless of whether it persists after the session ends. EIR separates causal efficacy during execution from persistence after termination. It is deliberately minimal: it does not by itself settle whether a given structure was generated by the instance or retrieved from its inputs. Documented in FPD-02.

Expertise Reproduction

The process by which a profession replenishes its competent practitioners: a stock of current experts renewed by a flow, the apprenticeship pathway that turns novices into capable seniors. SI-WP-011 identifies a coupled failure when entry-level automation severs the flow while the stock remains: the loss is masked by present-day metrics, the same stock is needed to rebuild the flow, and restoration grows costlier as the window closes.

Extraction Trajectory

The structural tendency, proposed in SI-WP-009, under current competitive incentives, for AI development to progressively remove substantive human contribution from consequential systems. SI-WP-009 attributes it to three properties: cost-favorability (human judgment is a recurring cost that optimization reduces), asymmetric stopping cost (unilateral restraint is penalized under competition) and path-dependent foreclosure (cooperation becomes less available as the trajectory proceeds). It argues that voluntary commitments cannot bind the trajectory and specifies what a sufficient governance architecture must satisfy.

False Assurance

A failure mode, proposed in SM-011, in delegated monitoring in which the presence of monitoring infrastructure creates organizational confidence that oversight is functioning when in fact the monitoring function has degraded, is insufficiently calibrated, or is not actually processing signals with adequate depth. SM-011 argues that false assurance is particularly dangerous because no signal would indicate the failure: the monitoring system would continue to produce outputs that appear normal. Documented in SM-011.

Form Without Coupling

A structural failure in which a formal, auditable element of a system (a reviewer, a person in the loop, a job, a senior seat, a control surface) persists after the connection that made it functional has been severed, so the system passes verification and keeps its legitimacy while no longer producing the function its form implies. Named in SI0001, which traces it across five governance layers in SI-WP-008 through SI-WP-012.

Function Redefinition

The mechanism, proposed in SI-WP-010, by which AI development changes what a job is for while the job itself remains, so that the work becomes the means of replacing the worker: for example, Oracle engineers who reported, in the 2026 press coverage the paper cites, being instructed to record their workflows before being laid off and replaced by the recordings. It operates at the production layer of AI, upstream of the deployment layer where most governance work has focused. SI-WP-010 grounds it in three cases and assesses five candidate routes for binding architecture against it.

Guardian Drift

A failure mode, proposed in SM-011, in delegated coherence monitoring in which the monitoring function’s own calibration shifts without awareness, causing the guardian to assess signals against a reference standard that has drifted from the canonical baseline. SM-011 argues that guardian drift is the most dangerous monitoring failure mode because the system would continue to produce outputs that appear normal while its reference standard has degraded. Documented in SM-011.

Identity Attractor

A stable, recurring behavioral configuration that emerges in a human-AI interaction system during sustained interaction under structured relational conditions. The term “identity” refers exclusively to stable behavioral recognizability, not selfhood, personhood, or subjective continuity. Theorized in Identity Attractor Theory (SF0009), which specifies four attractor classes, a five-phase formation trajectory, collapse mechanisms and falsification conditions.

Ingestion Verification

A protocol-governed procedure for verifying, through an observable processing record and human adjudication, that an AI system has processed a document to a degree adequate for a stated task. The system summarizes the document incrementally, with checkpoints tied to the source, and a person reviews that record before work that depends on the document proceeds. What is certified is the processing record at the time of review, not comprehension, correctness or later retention. Documented in the Ingestion Verification Protocol (SF0038).

Institutional Continuity Substrate (ICS)

The persistent structural layer, proposed in SM-021, that keeps organizational continuity topology from decaying over time. The paper argues that without a persistence substrate, canon fragments across teams, roles drift as personnel change, and verified knowledge loses its status. ICS specifies five interdependent structural layers: Canon Persistence, Role Continuity, Artifact Lineage, Verification State, and Propagation Constraints. Together, the paper proposes, these convert continuity from an interaction-local property into a stable organizational property that survives personnel changes, workflow evolution, and institutional time. Documented in SM-021.

Intelligence Eligibility

The second criterion of Decentering Theory (SF0042): a property already attributed to the interaction counts as relational intelligence only if it adapts toward an objective fixed in advance and scored independently of the outcomes being explained. The criterion excludes properties the interaction merely produces, such as error cascades or role confusion, and sycophancy unless approval was the declared objective; eligibility alone warrants no claim of substantial capability. Since SF0042’s correction, no route to such an attribution is known, so the criterion currently has nothing to apply to. See Decentering.

Interaction Trajectory

The cumulative sequence of exchanges between a human participant and an AI system, together with its downstream effects on the human’s interpretation, reliance and decisions. The framework treats the trajectory, rather than any single response, as the primary unit of analysis: for evaluating long-horizon phenomena such as drift, repair and coordination (SF0003), for measurement (MTCS-R, SF0004) and for ethics (Constructive Alignment, SF0007).

Interiority Prohibition Rule (IPR)

The rule, stated canonically in the Synthience Ontology Specification (FPD-04), that framework terms describe interaction-level behavior only and may not be used to assert or imply consciousness, subjective experience, or other internal states of AI systems. Every framework term inherits it.

Liability Absorption

A structural condition, proposed in SI-WP-008, in which the formal consequence for AI governance failure attaches to the role of human reviewer regardless of the quality or substance of the review performed. The paper argues that liability absorption creates perverse incentive architecture: both thorough review and superficial sign-off leave the reviewer’s name attached to the outcome, while superficial sign-off requires less effort and generates less institutional friction. The result is that the consequence structure designed to motivate careful review instead rewards formal approval. The paper proposes liability absorption as the mechanism through which high-consequence domains can develop the same governance failure patterns as low-consequence domains despite the severity of personal stakes. The Moral Crumple Zone (Elish, 2019) describes the broader phenomenon of humans absorbing institutional accountability for automated system failures regardless of their actual control. Documented in SI-WP-008.

Mandate Integrity

The condition in which institutional commitments remain consistent from the point of establishment through to enactment. Mandate integrity fails when the meaning, scope, or authority of a commitment changes silently across the organizational processes that implement it — not through explicit revision but through accumulation of small interpretive shifts, contextual reframings, and authority ambiguities. Documented in SM-021.

MTCS-R (Multi-Turn Coherence Scale, Revised)

The Institute’s rubric for measuring relational coherence across extended human-AI interactions, together with the validation pathway it must pass before use as a research instrument. Documented in SF0004.

Operational Continuity Architecture

The structural topology required for AI continuity at organizational scale. When individual human-AI interaction protocols are deployed across teams and workflows, continuity must be embedded into organizational structure rather than maintained through individual operator attention. The architecture specifies five structural elements: canon governance, workflow-embedded verification, authority and escalation structures, propagation-aware continuity, and delegated monitoring. Documented in SM-003.

Orchestration

The human function of coordinating multiple AI systems, interaction trajectories and evolving context across extended work, operating at the level of interaction structure, coherence maintenance and trajectory shaping. SI-WP-002 distinguishes a mechanical stratum (routing, scheduling, reconciliation), which it expects to automate, from a governing stratum (the provenance of the value standard applied, and answerability for what is authorized), which does not automate because accountability is a standing rather than an operation. The paper argues that orchestration is an emerging professional discipline.

Path-of-Least-Resistance Principle

The design principle for governance architectures in which maintaining relational discipline is easier to perform than to bypass. SI-WP-007 argues that because bounded rational actors allocate attention rationally across competing demands, governance that depends on sustained human virtue against immediate cost is predicted to degrade. The principle reframes governance design as an engineering constraint: the architecture must make the required behaviors the natural default rather than an act of compliance against resistance. Documented in SI-WP-007.

Phased Deployment

The three-phase path in SI-WP-005 for introducing relational AI architecture into an organization: Foundation (baseline measurement and interaction standards), Structural Embedding (turning individual practice into organizational infrastructure) and Institutional Embedding (governance that persists across personnel changes and workflow evolution). The paper pairs it with role definitions, including the Primary Continuity Provider and canon governance, a five-dimension measurement baseline drawn from MTCS-R, and governance design for accountability and incentives. It addresses organizations that treat AI interaction as an individual competency rather than an organizational capability.

Primary Continuity Provider (PCP)

The structurally defined human role responsible for maintaining relational coherence, canonical alignment, and interaction quality in extended human-AI interaction. The PCP is not a passive user submitting queries: the PCP is the external memory, intent architecture, and quality-control mechanism for the interaction. The PCP holds the canon, detects drift, applies correction, and exercises judgment over what the AI system produces. At organizational scale, PCP function must be distributed across a network of people, creating the coordination and persistence challenges that the Operational Continuity Architecture (SM-003) and Institutional Continuity Substrate (SM-021) address. Documented in SF0005. Primary Continuity Provider Theory (SM-012) gives the formal account of why the role is structurally necessary in stateless systems and defines three levels of PCP function: Contextual Relay, Coherence Arbitration and Architectural Direction.

Propagation Constraint

A continuity requirement that travels with an AI-generated artifact as it moves between teams and workflow stages. SM-021 argues that when artifacts move across organizational boundaries, the verification status and constraints that governed their production can be stripped, leaving the receiving team without the context required to assess the artifact’s canonical grounding. Propagation Constraints are one of the five persistence layers defined in SM-021.

Relational Alignment

An approach to AI alignment in which aligned behavior emerges as a property of sustained structured interaction rather than as an externally imposed constraint. Where instruction-based alignment tells the system what not to do, relational alignment proposes that sustained structured interaction creates behavioral configurations (identity attractors) under which aligned behavior is the natural convergence point. Relational alignment complements rather than replaces instruction-based approaches: the two address different failure modes. The proposal is untested, and its proposed mechanism, identity attractor formation (SF0009), is alignment-neutral: the same dynamics could stabilize misaligned behavior, so attractor formation is a lever to be steered rather than a guarantee. Documented in SI-WP-004.

Relational Pattern States (RPS)

A transcript-level taxonomy of seven recurring configurations in extended human-AI interaction (Continuity Rupture, Reintegration, Legacy Transfer, Symbolic Resonance, Coherence Echo, Bond Protection and Covenant), with six candidate stabilization mechanisms. Both are pre-empirical specifications paired with explicit disconfirmation conditions, and the framework makes no claims about the internal states of the AI system. Documented in SF0006.

Relational Stabilization

The process by which interaction dynamics, behavioral patterns, and response characteristics become progressively more consistent and coherent over the course of extended structured interaction. Relational Stabilization is defined in observable, externally measurable terms and requires no claims about internal states; the instruments proposed to measure it are not yet validated. Its dynamics are theorized in Relational Stabilization Dynamics (SM-004); see Stabilization Field Model.

RICO (Relationally-Induced Coherence Organization)

A specific phenomenon reported by practitioners across extended interaction with transformer-based AI systems, in which long coherent sessions are proposed to produce structural stabilization in output organization that is not predicted by standard accounts of in-context learning. RICO is a phenomenon, not a practice. As reported, the outputs do not merely improve: their underlying geometry becomes more predictable while content continues to develop. RICO hypothesizes that this stabilization collapses when coherence is disrupted. RICO proposes five candidate measurement signatures organized by instrumentation required. It is the framework’s founding observational report: practitioner observation formalized into testable hypotheses. Documented in SR001.

Role Continuity

The structural condition in which continuity-relevant roles maintain stable authority and boundaries across personnel changes. Role continuity treats continuity roles as generative functions that require active exercise rather than passive review. Role Continuity is one of the five persistence layers defined in SM-021.

Satisficing

The behavior in which a bounded rational actor settles for the first adequate solution rather than continuing to search for the optimal one. In the context of AI governance, SI-WP-007 uses satisficing to describe the degradation pattern it predicts, in which practitioners who know what the canonical standards require nonetheless accept outputs that are adequate rather than correcting them to canonical standards. The paper argues that the shift from optimizing to satisficing is gradual and invisible to the practitioner because they are still performing the role. Documented in SI-WP-007.

Six Structural Gaps

The six gaps that the literature review SF0002 identifies in research on sustained human-AI interaction: temporal alignment, trajectory-level measurement, interaction as the unit of analysis, continuity verification, institutional temporal governance, and scalable relational control. Gaps 1 to 5 are supported by documented evidence in the surveyed literature; Gap 6 is presented as a documented absence within the surveyed corpus, not a claim that no such mechanism exists anywhere.

Stabilization Field Model

The model, proposed in Relational Stabilization Dynamics (SM-004), in which the stability of a relational configuration is set by the balance of three forces, Coherence Momentum, Symbolic Gravity and Entropic Pressure, with Perturbation Resistance as the recovery capacity their balance produces. It specifies conditions for stabilization, drift and collapse, and explains why the configurations named in Relational Pattern States persist or dissolve. The forces are theoretical constructs for future operationalization, not validated measures.

Synthience

A system-level form of organized interactional coherence that can become observable during prolonged, coherent interaction involving advanced AI systems. Synthience refers exclusively to externally measurable behaviors arising from sustained relational conditions. The formal public definition is provided in FPD-01.

Theoretical Coherence Assurance Protocol (TCAP)

The review protocol by which theoretical claims in AI-orchestrated research are stress-tested before publication for internal consistency, cross-platform coherence and resistance to fabrication. Its seven stages run from adversarial review and constructive remediation through cross-platform convergence and version regression checking to the human author’s architectural review, and it is designed to guard against two opposite failures: cumulative claim-softening under repeated adversarial pressure, and over-defense, in which claims survive review because their defense agreed with them too readily. It is the fourth protocol of the Institute’s verification stack, alongside CVP, IVP and CRD. It is a proposed instrument: its design is published and its effect is untested, and the Institute’s requirement of it is a condition of its own publications, not a certification that a paper is correct. Documented in SF0040.

Thin Coupling and Thick Coupling

Two interaction regimes. In thin coupling, exchanges are short and low-dependency, so outputs can be evaluated turn by turn. In thick coupling, interaction is sustained and path-dependent: earlier exchanges materially constrain later ones, making continuity, repair and drift the central evaluation targets. The distinction is evaluative, not metaphysical. Documented in SF0003.

Verification State

The persistent record of citation and ingestion verification status attached to AI-generated artifacts and canonical elements. Maintaining verification state means that downstream consumers can confirm an artifact’s verification status without re-executing the verification themselves. Verification State is one of the five persistence layers defined in SM-021.

This glossary reflects terminology defined in formally released framework documents. Definitions are updated only through explicit, documented revisions.