Glossary
The vocabulary of Enterprise Decision Assurance, defined in plain English first — and in AnantVerify's specific sense where the two differ.
How to read this glossary
Every term leads with its ordinary meaning, because most of this vocabulary is not proprietary: decision, evidence, authority, coverage and state all mean something before they mean something here. Opening a term with “Learn more” adds the sense the platform assigns to it and the terms it is usually considered alongside.
That distinction matters more than it sounds. A word like coverage carries a statistical reading in most AI contexts — a probability that something is safe — and a specific reading here, where it is a record of what has actually been exercised. Reading the plain definition and stopping is a reasonable way to be wrong.
The definitions are also the site’s linking vocabulary. Terms such as decision assurance, decision transformation testing, enterprise state, decision replay and assurance coverage have stable anchors, so a page elsewhere on this site can send a reader to the exact definition it is using rather than restating it.
The six you need first
Decision Assurance
The layer that decides whether a decision may become an action.
AnantVerify
The assurance boundary that returns the verdict.
AnantDecision
One optional proposal source behind the same boundary.
DTT
The method that establishes how a decision must behave.
Enterprise state
The facts a decision rests on, as read from their owners.
Assurance Coverage
What has been tested, and what has not.
84 terms
Browse the glossary
Search a term, filter by category, or jump to a letter. Filters combine, and every term keeps its own anchor for direct linking.
84 of 84 terms
A
ALLOW
The assurance verdict that permits execution because the required conditions are satisfied on the state presented.
Decision AssuranceLearn moreShow less
ALLOW is a statement about entitlement, not about outcomes: the state was adequate, the evidence was present and consistent, the authority existed and the covered behavior held.
Related: Verdict · HOLD · ESCALATE · BLOCK
AnantDecision
The optional typed decision runtime that lets different decision engines operate behind a common contract.
Decision IntelligenceArchitectureLearn moreShow less
It is one proposal source among several. AnantVerify can assure proposals from agents, models, rules engines and human workflows whether or not AnantDecision is in use.
Related: Proposal · Proposal Envelope · AnantVerify
AnantVerify
The assurance boundary that evaluates a proposed decision against state, evidence, authority, behavior and coverage, and returns a verdict.
Decision AssuranceArchitectureLearn moreShow less
It sits between the systems that propose decisions and the systems that execute them, reads state from the systems of record, and does not execute actions itself.
Related: Verdict · Execution Gateway · Decision Assurance
Assurance Coverage
How much of the explicitly registered assurance requirement space has actually been exercised.
TestingDecision AssuranceLearn moreShow less
Coverage is a record of what has been tested and what has not — regions, boundaries, transformation classes and interactions. It is not a probability that the system is safe.
Related: Coverage Gap · Decision Transformation Testing · Graduated Autonomy
Assurance Gap
The part of a decision's behavior the enterprise has not tested, and therefore cannot claim to understand.
Decision AssuranceGovernanceLearn moreShow less
It is the term in the decision risk decomposition that an organization controls most directly, because it is closed by testing rather than by limits or fresher data.
Related: Decision Risk · Assurance Coverage · Coverage Gap
Assurance Memory
The persistent record of failures, conflicts, boundaries, incidents, overrides and outcomes that improves future assurance.
Decision AssuranceTestingLearn moreShow less
It is how a contained decision becomes a test: a hold that fired for a reason no test anticipated identifies a transformation class that is missing.
Related: Decision Replay · Learning Loop · Assurance Coverage
Audit Trail
The durable record of what a decision was, what supported it and how it was decided.
GovernanceDecision AssuranceLearn moreShow less
In a decision layer this is stronger than a log: it captures the state observed, the evidence attached, the authority resolved, the conditions evaluated and the verdict returned, so the decision can be replayed rather than reconstructed.
Related: Decision Replay · Decision Lineage · Provenance
B
BLOCK
The assurance verdict that the proposed action cannot be permitted under the applicable control.
Decision AssuranceLearn moreShow less
A block is recorded with the condition it violated, which makes it a finding about the decision system rather than only a prevented action.
Related: Verdict · Fail-Closed · HOLD
Boundary Condition
A threshold at which a decision changes character — above which approval is required, below which it is automated, or beyond which a case leaves standard handling.
TestingDecision AssuranceLearn moreShow less
Failure rates are highest in the band immediately inside a boundary, where small changes in input have outsized effects on outcome.
Related: Decision Boundary · Boundary Testing · Threshold
Boundary Testing
Evaluating decision behavior around a threshold: whether it is respected, how far its influence extends, and whether two thresholds interact to leave a gap.
TestingLearn moreShow less
The interaction case is the one enterprises discover late: two individually correct controls can leave a band where nothing governs the decision at all.
Related: Boundary Condition · Decision Boundary · Interaction Failure
C
Capability vs. Permission
The distinction between being technically able to take an action and being authorized to take it.
SecurityGovernanceLearn moreShow less
The two coincide during development, which is why they are confused. Authority is evaluated per decision rather than inherited from the credentials a runtime happens to hold.
Related: Decision Authority · Authenticated Caller · Agent Authority
Confidence Score
The proposer's own estimate of how likely its recommendation is to be right.
AI & AgentsDecision IntelligenceLearn moreShow less
It is evidence about the proposer's expectations, not a statement that the enterprise may act. A 97% recommendation can still be impermissible on stale state, missing approval or conflicting evidence.
Related: Proposal Envelope · Uncertainty · Proposal
Control Plane
A layer that manages a category of object continuously, applies policy to it, and is the authority on its state.
ArchitectureGovernanceLearn moreShow less
Enterprises have control planes for identity, data, network and observability. A decision control plane is the equivalent for decisions: applying conditions at execution and holding the record of what was permitted.
Related: System of Decision · Runtime Control · Decision Assurance
Counterfactual
A version of the same decision evaluated against a state that did not occur.
TestingDecision IntelligenceLearn moreShow less
Counterfactual evaluation is how behavioral properties are established: Decision Transformation Testing asks what the decision would do if a defined change were made, and whether that is what policy requires.
Related: Decision Transformation Testing · Transformation · Material Sensitivity
Coverage Gap
A region of the decision space, or a class of transformation, that has not been tested.
TestingGovernanceLearn moreShow less
Coverage gaps should determine where automation is permitted and at what level of autonomy, rather than being discovered after an incident in an untested region.
Related: Assurance Coverage · Assurance Gap · Graduated Autonomy
D
Decision Assurance
Determining whether an AI-assisted decision is sufficiently supported to become an enterprise action.
Decision AssuranceLearn moreShow less
It evaluates whether the enterprise knows enough to act: state currency, evidence, authority, behavior under change, coverage, and whether the conditions still hold at execution.
Related: Decision Risk · Runtime Control · AnantVerify · Decision Transformation Testing
Decision Boundary
The line that separates one handling of a decision from another — automate from review, standard from escalated, permitted from refused.
Decision AssuranceTestingLearn moreShow less
Boundaries are where decisions change character, so they are where behavior must be most thoroughly understood before autonomy is granted.
Related: Boundary Condition · Boundary Testing · Threshold
Decision Class
A group of decisions that share a state requirement, an authority regime and a consequence profile.
Decision AssuranceGovernanceLearn moreShow less
Coverage, autonomy limits and runtime policy are all set per decision class, which is what makes graduated autonomy expressible.
Related: Autonomy Level · Assurance Coverage · Runtime Policy
Decision Contract
The typed definition of the decisions a system is allowed to produce, with the semantics each must satisfy.
ArchitectureDecision AssuranceLearn moreShow less
It constrains the action vocabulary so that assurance can be deterministic: an action that is not in the contract cannot be proposed, let alone permitted.
Related: State Contract · Transformation Contract · Decision Pack
Decision Intelligence
The discipline of improving how decisions are made, rather than only how predictions are modeled.
Decision IntelligenceLearn moreShow less
It concerns the decision as a unit: the state it rests on, the options available, the authority that governs it and the outcome it produces.
Related: AnantDecision · Decision Task · Decision Assurance
Decision Lineage
The trace from a decision back to the state, evidence, policy version, model version and authority that produced it.
Decision AssuranceGovernanceLearn moreShow less
Its value appears when something upstream changes: lineage turns the question of which decisions were affected into a query rather than an investigation.
Related: Audit Trail · Decision Replay · Provenance
Decision Pack
The versioned package of state, decision and transformation contracts, evidence requirements, tests, gaps and provenance for a decision class.
ArchitectureTestingLearn moreShow less
The pack is the unit of deployment for assurance: what a decision class requires, what has been tested about it, and what is known to remain untested.
Related: State Contract · Decision Contract · Transformation Contract
Decision Replay
Reconstructing a past decision from its record, so the same inputs reproduce the same verdict.
Decision AssuranceGovernanceLearn moreShow less
Replay answers why an action was permitted without log forensics, and detects drift in policy and behavior over time rather than only after an incident.
Related: Audit Trail · Decision Lineage · Assurance Memory
Decision Risk
The risk that arises from a decision itself: consequence multiplied by uncertainty in the facts, multiplied by what has not been assured about the behavior.
Decision AssuranceGovernanceLearn moreShow less
Consequence can be bounded by limits and reversibility, uncertainty by fresher state and binding, and the assurance gap by testing.
Related: Assurance Gap · Uncertainty · Risk Appetite
Decision Task
A defined unit of decision work: the question being decided, the inputs it requires and the action vocabulary it may produce.
Decision IntelligenceArchitectureLearn moreShow less
Making the task explicit is what allows the same decision to be evaluated the same way regardless of which engine produced the proposal.
Related: Decision Contract · Decision Pack · Proposal
Decision Transformation Testing
A method that asks what should happen to a decision when enterprise state, evidence, time or authority changes in a defined way — and what actually happens.
TestingLearn moreShow less
Transformations are derived from enterprise meaning rather than statistical perturbation, expectations are declared before the test is run, and every result maps to a runtime control.
Related: Invariance · Material Sensitivity · Temporal Behavior · Assurance Coverage
Delegation
The grant of decision authority from one actor to another, including to a machine actor.
GovernanceSecurityLearn moreShow less
Delegations are scoped, limited and revocable. Evaluating them per decision is what prevents an agent's credentials from becoming its permission.
Related: Decision Authority · Agent Authority · Runtime Policy
E
Enterprise State
The facts a decision depends on, as observed from the systems that own them, at a point in time.
State & DynamicsLearn moreShow less
It is a trajectory rather than a snapshot: two requests with identical fields can differ only in the sequence of commitments that produced the current figures.
Related: State Currency · State Sufficiency · State Version · System of Record
ESCALATE
The assurance verdict that the decision requires authority the automated path does not hold.
Decision AssuranceLearn moreShow less
Escalation carries the reason with it: the conditions that caused it, the evidence in conflict and the specific question a human authority has to answer.
Related: Verdict · Escalation Path · Human-in-the-Loop
Escalation Path
The route a decision takes when it needs authority beyond the automated path, and the authority that receives it.
GovernanceDecision AssuranceLearn moreShow less
An escalation path is only a control if the receiving person has the information, the time and the ability to refuse — otherwise it is a notification.
Related: ESCALATE · Decision Authority · Human-in-the-Loop
Evidence Assurance
Establishing that the support for a decision is present, current, in scope and consistent with everything else that bears on it.
Decision AssuranceLearn moreShow less
Evidence is treated as an object with properties rather than as material the model happened to retrieve, so it can be checked and later replayed.
Related: Evidence Set · Provenance · Evidence Resolution Policy
Evidence Resolution Policy
The customer-owned rule for applying declared evidence authority when sources conflict.
GovernanceDecision AssuranceLearn moreShow less
The policy, not the assurance layer, decides which source wins. Where it does not resolve a material conflict, the decision fails closed to escalation.
Related: Evidence Set · Conflict · Fail-Closed
Evidence Set
The provenance-bearing collection of material relevant to a decision, preserved even when its sources disagree.
Decision AssuranceArchitectureLearn moreShow less
Preserving the conflict rather than collapsing it is what allows a human to adjudicate with the facts in front of them.
Related: Evidence Assurance · Provenance · Conflict
Execution Gateway
The component through which a consequential action leaves the enterprise, and which acts only on a verdict and approval it can verify.
ArchitectureDecision AssuranceLearn moreShow less
It is deliberately narrow: it verifies a verdict, checks the state binding, executes and reports the outcome. Everything interesting was decided upstream.
Related: State Binding · Verdict · Runtime Control
Explainability
The ability to state why a system produced the result it did.
AI & AgentsGovernanceLearn moreShow less
Decision assurance needs something different from model explainability: a record of the conditions that were evaluated and the state they were evaluated on, whether or not the model's internals are interpretable.
Related: Audit Trail · Decision Replay · Provenance
F
Fail-Closed
The property that a missing or unusable input holds a decision rather than releasing it.
ArchitectureGovernanceLearn moreShow less
It has to be a stated design decision rather than an assumption: a state source that is unavailable must not read as an acceptable default.
Related: BLOCK · HOLD · State Sufficiency
G
Graduated Autonomy
Granting autonomy to a decision class in steps, on demonstrated coverage, and withdrawing it when the evidence changes.
Decision AssuranceGovernanceLearn moreShow less
Levels run from advisory through supervised and conditional autonomy to bounded autonomy. Movement is earned by coverage and behavior, not by benchmark scores.
Related: Autonomy Level · Assurance Coverage · Decision Class
Ground Truth
The definitive outcome against which a decision or prediction is judged.
AI & AgentsTestingLearn moreShow less
Historical decisions are not ground truth: they may contain outdated policy, exceptions, bias, incomplete information or mistakes. They are evidence and discovery signal.
Related: Outcome Feedback · Learning Loop · Model Evaluation
Guardrails
Constraints on what a model may produce or say, applied to its output in isolation from the enterprise context.
AI & AgentsSecurityLearn moreShow less
A guardrail can stop a disallowed statement. It cannot stop a permitted statement from becoming an impermissible action, because the conditions that make the action impermissible are not properties of the output.
Related: Decision Assurance · Policy Binding · Verdict
H
Human-in-the-Loop
Placing a person in the path of a decision before it executes.
GovernanceDecision AssuranceLearn moreShow less
Necessary but not sufficient: review degrades under volume, information asymmetry, automation bias and unclear authority. Assurance makes review a designed control by routing specific cases with the question stated.
Related: ESCALATE · Escalation Path · Reviewer Authority
HOLD
The assurance verdict that execution should wait for required state, evidence or a condition.
Decision AssuranceLearn moreShow less
A hold defers rather than refuses, and names the specific missing element so the decision can be re-evaluated when it arrives.
Related: Verdict · Fail-Closed · State Sufficiency
Human Approval
An approval produced by a person, modeled as an artifact with scope, limit and expiry that the gateway can verify.
ArchitectureGovernanceLearn moreShow less
An approval that exists only as a message or a flag cannot be checked at execution, because nothing distinguishes it from approval of a materially different request.
Related: Human-in-the-Loop · Execution Gateway · Reviewer Authority
I
Interaction Failure
The case where individually correct decisions combine into an outcome no component would have chosen alone.
TestingDecision AssuranceLearn moreShow less
These are the hardest class to find in testing and the easiest to find in production, because each component passes its own evaluation.
Related: Decision Boundary · Boundary Testing · Assurance Coverage
Invariance
A property that must not change the decision: rephrasing, reordering, reformatting or a non-material field change.
TestingLearn moreShow less
Invariance failures are the signature of systems that learned incidental correlations, and they are invisible in aggregate accuracy metrics.
Related: Material Sensitivity · Decision Transformation Testing · Non-Determinism
Invariant
A declared expectation about what a decision must always do, or must always remain indifferent to.
TestingDecision AssuranceLearn moreShow less
Declaring invariants before testing is what makes a test capable of failing, rather than an observation that is explained afterwards.
Related: Invariance · Property-Based Testing · Transformation Contract
J
Judgment vs. Verdict
Judgment weighs what should be done; a verdict states whether the conditions for acting are satisfied.
Decision AssuranceGovernanceLearn moreShow less
Assurance does not replace judgment. It removes the class of failure that comes from acting without entitlement to act, and preserves the judgment for the cases that need it.
Related: Verdict · Human-in-the-Loop · Decision Authority
K
Kill Switch
A control that stops a decision class from executing while its behavior or coverage is in question.
ArchitectureGovernanceLearn moreShow less
Because runtime policy lives at the boundary rather than in the model, a kill switch takes effect on the next decision rather than on the next release.
Related: Runtime Policy · Autonomy Level · Runtime Control
L
Learning Loop
The cycle in which production outcomes and interventions become new tests, conditions and coverage.
Decision AssuranceTestingLearn moreShow less
It is what makes assurance continuous: discoveries narrow the region of the decision space the enterprise does not understand.
Related: Assurance Memory · Outcome Feedback · Assurance Coverage
M
Material Sensitivity
The property that a decision must move when an operative fact moves, in the direction and to the degree policy implies.
TestingLearn moreShow less
Its inverse failures are insensitivity (no change when there should be) and hypersensitivity (a reaction larger than the change warrants). Neither is visible in accuracy metrics.
Related: Invariance · Transformation · Threshold
Model Evaluation
Measuring how well a model performs: accuracy, calibration, robustness and behavior on held-out data.
AI & AgentsTestingLearn moreShow less
It answers whether the model is reliable on average. It does not answer whether a specific decision may execute on the state in front of it.
Related: Decision Assurance · Model Risk · Assurance Coverage
Model Risk
The risk that a model produces wrong or misleading outputs.
GovernanceAI & AgentsLearn moreShow less
It is usually managed as a property of the model. Decision risk is the property the enterprise actually experiences, because it is the decision that becomes a committed action.
Related: Decision Risk · Model Evaluation · Risk Appetite
Monotonicity
The property that a change in one direction should never produce a change in the opposite direction.
TestingLearn moreShow less
Increasing an amount above a limit should never make a payment more likely to be allowed. Where that fails, the decision is not following the policy it claims to implement.
Related: Material Sensitivity · Invariant · Decision Boundary
N
Non-Determinism
The property that the same input can produce different outputs across runs.
AI & AgentsTestingLearn moreShow less
A verdict must not be non-deterministic with respect to its inputs, because replay depends on the same inputs reproducing the same decision.
Related: Decision Replay · Invariance · Verdict
O
Observability
The ability to see what a system did: traces, metrics, logs, latency and error rates.
AI & AgentsArchitectureLearn moreShow less
It is retrospective and it does not tell you whether a decision should have been permitted. A fully healthy trace can sit behind an action the enterprise had no authority to take.
Related: Decision Replay · Audit Trail · Runtime Control
Outcome Feedback
Returning the result of an action to the decision record, so the enterprise learns about its decision surface.
Decision AssuranceTestingLearn moreShow less
A permitted decision followed by an adverse outcome is a case whose conditions were insufficient, whether or not the model was wrong.
Related: Learning Loop · Assurance Memory · Ground Truth
Override
A person setting aside an automated verdict, with the reason attached.
GovernanceDecision AssuranceLearn moreShow less
Overrides are evidence about the conditions themselves: a hold that an operator routinely sets aside identifies a condition that is too strict or imprecisely expressed.
Related: Assurance Memory · Human-in-the-Loop · Reviewer Authority
P
Policy Binding
Applying the policy version in force at the moment of execution, and recording which version that was.
GovernanceDecision AssuranceLearn moreShow less
A rule that lives only in a prompt, a fine-tune or a pattern in training data cannot be versioned, verified or replayed, so it is not a control.
Related: Runtime Policy · Decision Replay · Guardrails
Property-Based Testing
Declaring the properties a system must hold, then searching for cases that break them.
TestingLearn moreShow less
Decision Transformation Testing applies the idea to decisions: the properties are invariants and material sensitivities, and the cases are transformations of enterprise meaning.
Related: Invariant · Decision Transformation Testing · Counterfactual
Proposal
What a decision system recommends should happen, submitted to assurance as a claim rather than as a fact.
Decision IntelligenceArchitectureLearn moreShow less
A structured proposal carries the action, its parameters, the entity it concerns and the context the proposer believes applies. Prose has to be interpreted before it can be evaluated, and interpretation shifts meaning.
Related: Proposal Envelope · Verdict · AnantDecision
Proposal Envelope
The normalized proposal together with its confidence, uncertainty, engine identity, state reference and provenance.
ArchitectureDecision IntelligenceLearn moreShow less
Normalizing the envelope is what allows proposals from different engines to be evaluated and recorded the same way, and later replayed.
Related: Proposal · Confidence Score · State Binding
Provenance
Where a piece of state, evidence or policy came from, and when it was retrieved.
Decision AssuranceGovernanceLearn moreShow less
Without provenance, an enterprise cannot answer which decisions rested on a source that has since been found unreliable.
Related: Evidence Set · Decision Lineage · Audit Trail
Q
Query-Time Assurance
Evaluating the decision at the moment it would execute, on the state presented for execution.
ArchitectureDecision AssuranceLearn moreShow less
This is what makes assurance a control rather than a prediction: a decision approved against an earlier state may no longer be valid when the action is taken.
Related: State Binding · Runtime Control · Time-of-Check to Time-of-Use
R
Risk Appetite
How much risk the organization is willing to carry, including how much autonomy it is willing to grant.
GovernanceLearn moreShow less
At the decision level, appetite becomes conditions and limits: which decisions may execute without review, up to what amount, in which regions.
Related: Decision Risk · Graduated Autonomy · Runtime Policy
Runtime Control
Applying assurance conditions at the point of action, and returning a verdict the executing system obeys.
Decision AssuranceArchitectureLearn moreShow less
It belongs between the proposing system and the executing system, so that no path to a consequential action bypasses it.
Related: Verdict · Execution Gateway · Kill Switch
Runtime Policy
The limits, conditions and restrictions in force at the moment a decision would execute.
GovernanceDecision AssuranceLearn moreShow less
Because it lives at the boundary, a policy change takes effect on the next decision rather than waiting for a model release.
Related: Policy Binding · Kill Switch · Risk Appetite
S
State Assurance
Establishing that the state a decision rests on is adequate for that decision class.
State & DynamicsDecision AssuranceLearn moreShow less
It is context-dependent: a ten-minute-old inventory figure is fine for a reorder recommendation and unacceptable for a next-day delivery promise.
Related: State Currency · State Sufficiency · Enterprise State
State Binding
Attaching a reference to the evaluated state to the verdict, and verifying it before the action executes.
State & DynamicsDecision AssuranceLearn moreShow less
It closes the window between evaluation and execution. Where the state has moved materially, the binding fails and the decision is re-evaluated rather than executed.
Related: Time-of-Check to Time-of-Use · State Version · Execution Gateway
State Contract
The definition of the enterprise state required for assurance, including freshness, conflict and provenance rules.
State & DynamicsArchitectureLearn moreShow less
Declaring the contract is what allows the boundary to detect stale, missing or inconsistent state instead of accepting whatever it was given.
Related: Decision Contract · State Currency · State Sufficiency
State Currency
How recently a fact was observed, relative to how quickly it can change.
State & DynamicsLearn moreShow less
Currency is a property of the observation, not of the record: a field can be present, correct and no longer true.
Related: State Assurance · Enterprise State · State Version
State Sufficiency
Whether the available state contains what this decision class requires, as opposed to what happens to be available.
State & DynamicsDecision AssuranceLearn moreShow less
Sufficiency is declared per decision class, which is what stops a decision from proceeding on the fields a caller happened to supply.
Related: State Contract · State Assurance · Fail-Closed
State Version
An identifier for the observation of state a decision was evaluated against.
State & DynamicsArchitectureLearn moreShow less
Versions are what make binding possible and what allow a replay to know which observation it is reconstructing.
Related: State Binding · Provenance · Decision Replay
System of Decision
A system whose authoritative subject is permission rather than fact, evaluated per action rather than per record.
ArchitectureDecision AssuranceLearn moreShow less
It is the counterpart to the system of record: the record answers where the enterprise's facts live, the decision system answers what the enterprise may do with them.
Related: System of Record · Control Plane · Decision Assurance
System of Record
The authoritative source for a fact, which remains the authority for that fact regardless of who reads it.
ArchitectureState & DynamicsLearn moreShow less
Decision assurance reads from systems of record and never becomes one: it contributes the verdict and its record, not the state.
Related: Enterprise State · System of Decision · Provenance
T
Temporal Behavior
How a decision behaves with respect to sequence and time: things that must happen in order, once, or within a window.
TestingState & DynamicsLearn moreShow less
Two decisions can rest on identical states reached by different paths, and policy can require different outcomes for them.
Related: Transformation · Monotonicity · Decision Transformation Testing
Threshold
A value at which the handling of a decision changes: an amount, a severity, a risk tier or a confidence level.
TestingGovernanceLearn moreShow less
Thresholds are the most consequential configuration in a decision system, because they are where autonomy, review and refusal change places.
Related: Boundary Condition · Decision Boundary · Runtime Policy
Time-of-Check to Time-of-Use
The class of failure where conditions verified earlier no longer hold when the action is finally taken.
State & DynamicsArchitectureLearn moreShow less
It applies to AI-assisted decisions exactly as it applies to any concurrent system, and it is answered by binding the decision to the state it was evaluated against.
Related: State Binding · Query-Time Assurance · Execution Gateway
Transformation
A defined change to enterprise state, evidence, time or authority, applied to a decision to see how it responds.
TestingLearn moreShow less
Transformations are derived from enterprise meaning — an approval expiring, a balance crossing a limit — rather than from statistical perturbation of inputs.
Related: Decision Transformation Testing · Counterfactual · Material Sensitivity
Transformation Contract
The definition of how a decision should behave after a valid transformation.
TestingArchitectureLearn moreShow less
Declaring the expectation before the transformation runs is what makes the result a test rather than an observation.
Related: Invariant · Transformation · Decision Pack
U
Unassured Action
A consequential action taken without an assurance verdict on the state it executed against.
Decision AssuranceGovernanceLearn moreShow less
The category includes actions evaluated against earlier state, actions on a path that bypasses the gateway, and actions enabled by capability rather than permission.
Related: Execution Gateway · State Binding · Capability vs. Permission
Uncertainty
How much a decision depends on facts that may have moved, or on a model whose behavior is not fully established.
Decision AssuranceDecision IntelligenceLearn moreShow less
It is one of the three terms in decision risk, and it is bounded by fresher state, explicit evidence and state binding rather than by better predictions.
Related: Decision Risk · Confidence Score · Assurance Gap
V
Verdict
The assurance determination returned for a proposed decision: ALLOW, HOLD, ESCALATE or BLOCK.
Decision AssuranceLearn moreShow less
The four verdicts are the output of AnantVerify, and they are deliberately coarse so that surrounding systems can act on them deterministically.
Related: ALLOW · HOLD · ESCALATE · BLOCK
Z
Zero-Trust Decision Path
Treating a proposal as unverified until the conditions for acting on it have been established, regardless of who produced it.
SecurityArchitectureLearn moreShow less
The principle applies to human and machine proposers alike: entitlement to act is established per decision rather than inherited from a session, a credential or a role.
Related: Capability vs. Permission · Verdict · Decision Authority
The Anant vocabulary in one view
The same terms arranged by what they are for, rather than alphabetically. Each links to its definition.
The decision layer
The components and the object they act on: a proposal in, a verdict out.
The four verdicts
Four operational outcomes, and what each one commits the enterprise to do.
State and time
Why a snapshot is not state, and why the moment of execution decides.
Evidence and authority
What supports a decision, and who or what permits it.
Testing and coverage
How behavioral properties are established, and how the untested is recorded.
Governance and record
Where policy, accountability and the record of what was permitted live.
From state to trusted action
- Enterprise state
- Proposal
- Evidence & authority
- Verdict
- Execution
- Outcome & memory
The path each of these terms belongs to, in the order a single decision travels it.
The vocabulary is the shortest path into the discipline
Most disagreements about AI assurance turn out to be disagreements about words: whether coverage means safety, whether confidence means permission, whether observability means control. Start with the definitions, then take the argument to a single decision class.
Continue exploring
- What is Decision Assurance?The framework the glossary defines the vocabulary for.
- Reference architectureWhere each term sits in the deployed decision path.
- DTT methodologyTransformations, expectations, coverage and interpretation.
- Evidence and authorityThe conditions that make a decision entitled to execute.
- Decision replay and auditReconstructing a decision from the record it left.
- Questions and answersThe glossary's terms, answered in full sentences.