Open specification

Draft 0.10

Private Wealth Management Ontology

A shared vocabulary for private banking and wealth management, and for the AI that works on it. Built on FIBO, with every term cited to the standard, regulation or Kognita source it comes from.

Version
0.10 (draft), 19 Nov 2026
Namespace
https://ahmedmuzammil.com/ontology/pwm#
Version IRI
https://ahmedmuzammil.com/ontology/pwm/0.10
Shapes IRI (latest)
https://ahmedmuzammil.com/ontology/pwm/shapes
Shapes version IRI
https://ahmedmuzammil.com/ontology/pwm/0.10/pwm-shapes.ttl
Licence
CC BY-SA 4.0
  • 244classes
  • 659properties
  • 79code lists
  • 186written rules
  • 288node shapes
  • 47references
  • 2004citations

About

A generic ontology for private banking and wealth management. It models parties and relationships, due diligence, investor profiling and suitability, mandates, accounts and holdings, products and research, client interactions, and the governance of AI that acts on them: use cases, agents, delegated authority, claims and evidence. Business concepts reuse or specialize the Financial Industry Business Ontology (FIBO) and the OMG Commons library. Governance concepts follow the MAS Guidelines on Artificial Intelligence Risk Management (7 October 2026) and the Kognita roadmap. Every term cites the references it is grounded in. Institution-specific detail belongs in an extension that imports this ontology.

Draft for comment. Version 0.10 completes the catch-up with Kognita 0.8 and finishes the business scope and the supervisory-readiness artifacts the roadmap set for 0.9 and 0.10: costs and charges, product governance, client circumstances, operations and lifecycle, performance provenance, an AI inventory profile, reconstruction queries, a Kognita mapping table, a JSON-LD round trip, jurisdiction profiles and a code-list profile. Citations are not all verified yet: see the dagger count below. Terms, code lists and rules may change before 1.0, and every change is recorded in the change log. It is not legal advice: the regulatory citations show where a term comes from, not how a rule applies to a given firm.

In scope

  • Parties, roles and the client relationship, including trusts, foundations and other legal arrangements
  • Due diligence, risk rating, source of wealth and funds, screening and tax residency
  • Investor classification, risk profile, knowledge and experience, sustainability preferences, suitability and appropriateness
  • Mandates, investment restrictions and fees
  • Accounts, portfolios, holdings and valuations
  • The product shelf, product risk, target market, cross-border rules, offering documents and the house view
  • The client journey: origination, recommendation, final decision and client communication
  • AI governance: the use-case inventory, risk materiality, accountability, agents, tools, delegated authority and autonomy
  • Trust and resilience: signed and anchored evidence, key management, the model-provider register, failover, the dependency map, drills, retention by record class and incident reports
  • Runtime assurance: agent manifests, external-agent assertions, deployment attestations, update approvals, component approvals, production outcomes and supervisory conformance reports
  • Business operations: costs and charges, product governance, client circumstances, orders, executions and allocations with best-execution evidence, corporate actions and income, CRS and FATCA reporting records, lending facilities and performance figures as provenance
  • Information governance, facts and claims, policy decisions and hash-chained evidence

Not in 0.10

  • Calculation of costs, performance, tax or limits: PWM records the figure and its provenance, never the method
  • Any institution's thresholds, scales, product shelf or internal names: these belong in an extension

Design principles

  1. 01

    Reuse before inventing

    Business terms specialize or map to FIBO first (Release before Provisional), then OMG Commons, W3C PROV-O and DPV. A term is defined here only when none fits, and it is marked NEW.

  2. 02

    Declare, don't import

    External terms are declared with their label, source and FIBO maturity instead of importing whole ontologies, so the file loads on its own. Specialization uses rdfs:subClassOf; looser alignment uses SKOS exact, close and related matches.

  3. 03

    Every term cites its grounds

    Each class, property and code list carries at least one citation with a locator: an article, paragraph, roadmap item or slide. The reference register records how each source was checked.

  4. 04

    Logic in OWL, codes in SKOS, rules in SHACL

    The ontology uses light OWL axioms that an extension can build on. Every datatype is in the OWL 2 datatype map: 0.10 migrated nextReviewDate and retentionUntil to xsd:dateTime and replaced freshnessThreshold (xsd:duration) with freshnessThresholdSeconds. Code lists are SKOS concept schemes. Data rules are SHACL shapes kept in a separate file.

  5. 05

    Context speed and classification

    Each class states how fast its facts change and the data classification it carries by default. Fast facts, such as positions and prices, are looked up live and are never cached or embedded for AI.

  6. 06

    Accountability stays with people

    Accountability roles are played by named people. A shape rejects any record that names an AI agent in one.

  7. 07

    US English

    PWM-defined names, labels and prose use US English spelling, for example authorized, behavior, center and specialize. Names reused from OMG Commons or FIBO keep the spelling of their source. Kognita uses US spelling in its identifiers and values too, so PWM notations equal the Kognita values. A renamed term stays declared as a deprecated alias for one version.

  8. 08

    Shared properties omit a domain

    A property that several classes use, such as classification, approvedBy or citesSource, declares no rdfs:domain, so using it never infers a class. SHACL says where each class requires it. Its range is stated, as a class, a union or prov:Entity for any record.

  9. 09

    Generic core, firm extension

    No institution's data, thresholds or names appear here. A firm adds them in its own ontology that imports PWM.

How to read a term

Every class, property and code list lists its citations: a mark for the kind of source, the source, and the article, paragraph, roadmap item or slide it rests on.

ONTstandard ontologyPublished ontology reused or specialized: FIBO, OMG Commons, W3C PROV-O, DPV.
ISOtechnical standardISO or ISO/IEC standard, used for identifiers, code formats and terminology.
REGregulation or supervisory guidanceLaw, regulation, supervisory guideline, notice or international standard-setter recommendation.
KOGKognitaKognita roadmap, regulations library (control-frameworks.md), controlled vocabularies or policy packs.
AMauthor frameworkEVOLVE framework, essays and talks by Ahmed Muzammil.
NEWpractitioner definitionDefined in this ontology because no published standard term was found at the versions checked.

FIBO terms show their maturity, Release or Provisional. Each reference in the register is marked verified (checked against the source while drafting) or instrument-level (paragraph numbers to re-check before formal use). Classes show how fast their facts change: slow: changes over weeks or months, such as policies, definitions and the house view; medium: changes over days or weeks, such as client profile, kyc status and mandate; fast: changes in seconds or minutes, such as positions, prices, cash, orders and live authority or kill-switch state.

Terms

Module 0

Core patterns

Assessments, validity periods and supersession: the patterns every other module reuses.

Classes

assessment

pwm:Assessmentmedium context

Dated, attributable record of an evaluation of a party, product, portfolio or AI use case, valid for a period and replaceable by a later assessment.

Due diligence, investor classification, risk profile, suitability, product risk rating and AI risk materiality are all assessments. Who assessed it is recorded with prov:wasAttributedTo; when, with prov:generatedAtTime.

specialises
EntityPROV-O
properties
assesses

Properties

supersedes

pwm:supersedesobject property

Earlier version that this record, rule or document replaces. The earlier version stays readable for reconstruction but may no longer be used.

domain → range
anything
close match
ReplacesDCMI

approved by

pwm:approvedByobject property

Role whose holder approved this record.

domain → range
anyparty roleOMG Commons
  • REGMAS AIRG4.13: control function approves materiality; 5.21: approval body

valid from

pwm:validFromdatatype property

Start of the period in which a record, rule or document is in force.

domain → range
anydateTimeXSD
close match
Date ValidDCMI

valid until

pwm:validUntildatatype property

End of the period in which a record, rule or document is in force. Absent while it remains in force.

domain → range
anydateTimeXSD
close match
Date ValidDCMI
  • KOGKognita roadmap0.6 item 2: created, effective, superseded and expiry dates

next review date

pwm:nextReviewDatedatatype property

Time by which a record must be reviewed. A decision made after this instant treats the record as expired. A date-only value from earlier versions is read as the end of that day.

domain → range
anydateTimeXSD

rationale

pwm:rationaledatatype property

Reasons for an assessment or decision, stated as facts and evidence rather than as a bare score.

domain → range
anystringXSD

freshness threshold (deprecated)

pwm:freshnessThresholddatatype property

Deprecated: an xsd:duration. Use pwm:freshnessThresholdSeconds, for example P1D becomes 86400.

domain → range
anyany

Module 1

Parties, roles and relationships

Clients, related parties, legal arrangements, providers and the relationship that binds them.

Classes

client

pwm:Clientmedium contextC2

Role of a party that has, or is establishing, a business relationship with a wealth manager to receive investment, banking, advisory or fiduciary services.

specialises
clientFIBO · Release
properties
inGroup

entity client

pwm:EntityClientC2

Client role played by a legal person or a legal arrangement, such as a company, foundation or trust.

specialises
client

prospect

pwm:ProspectC2

Role of a party a wealth manager is considering or approaching before any business relationship exists.

specialises
party roleOMG Commons

client relationship

pwm:ClientRelationshipmedium contextC2

Business relationship between a wealth manager and one or more clients, booked in a booking center and covered by named relationship managers for a period.

specialises
situationOMG Commons

relationship group

pwm:RelationshipGroupmedium contextC2

Set of related clients, such as a family, its holding structures and its trusts, managed and reported together.

also called
family group, household

investment adviser

pwm:InvestmentAdviser

Role of a person who gives investment advice to clients under the wealth manager's license or registration.

specialises
party roleOMG Commons

external asset manager

pwm:ExternalAssetManager

Independent firm that manages a client's assets booked with the wealth manager, acting under a limited power of attorney granted by the client.

specialises
party roleOMG Commons
also called
independent asset manager

authorized signatory

pwm:AuthorizedSignatory

Person authorized to give instructions on an account or for an entity client, within stated signing rules.

specialises
authorized individualFIBO · Release
  • ONTFIBOBE/OwnershipAndControl/Executives/AuthorizedIndividual, Release; master source verified 2026-11-03
  • NEWPWM practitioner definitionAccount mandate and signing authority

power of attorney holder

pwm:PowerOfAttorneyHolder

Party authorized by a client, under a power of attorney, to act for the client within the scope that instrument grants.

specialises
authorized individualFIBO · Release
  • ONTFIBOBE/OwnershipAndControl/Executives/AuthorizedIndividual, Release; master source verified 2026-11-03
  • NEWPWM practitioner definitionGeneral and limited powers of attorney

beneficial owner

pwm:BeneficialOwnerC2

Natural person who ultimately owns or controls a client, or on whose behalf a transaction is conducted, including those exercising ultimate effective control over a legal person or arrangement.

specialises
beneficial ownerFIBO · Release

controlling person

pwm:ControllingPersonC2

Natural person who exercises control over an entity account holder for tax-reporting purposes; for a trust, the settlor, trustees, protector, beneficiaries and anyone with ultimate effective control.

specialises
party roleOMG Commons
close match
beneficial owner

politically exposed person

pwm:PoliticallyExposedPersonC3

Role of a person who is or has been entrusted with a prominent public function, or is a family member or close associate of such a person, attracting enhanced due diligence.

specialises
party roleOMG Commons
also called
PEP
properties
pepCategory

connected party

pwm:ConnectedPartyC2

Party linked to a client by family, business or control, recorded for due diligence and conflict checks.

specialises
party roleOMG Commons
properties
connectedTo

legal arrangement

pwm:LegalArrangement

Express trust or similar arrangement through which assets are held and controlled, treated as a party to the relationship.

specialises
partyOMG Commons

settlor

pwm:Settlor

Party that establishes a trust and places assets into it.

specialises
trustorFIBO · Release
exact match
trustorFIBO · Release
also called
trustor

trustee

pwm:Trustee

Party that holds and manages trust assets for the beneficiaries.

specialises
trusteeFIBO · Release
exact match
trusteeFIBO · Release

protector

pwm:Protector

Party appointed under a trust instrument to oversee the trustee, often with powers of consent or removal.

specialises
party roleOMG Commons

private foundation

pwm:Foundation

Legal person, without shareholders, established to hold and administer assets for stated purposes or beneficiaries.

specialises
legal entityOMG Commons

family office

pwm:FamilyOffice

Organization that administers and manages the affairs and wealth of one or more families.

specialises
family officeFIBO · Release
exact match
family officeFIBO · Release
  • ONTFIBOBE/FunctionalEntities/FunctionalEntities/FamilyOffice, Release; master source verified 2026-11-03
  • NEWPWM practitioner definitionClient or intermediary

wealth manager

pwm:WealthManager

Financial service provider that delivers wealth management services: advice, portfolio management, custody, lending and planning.

specialises
financial service providerFIBO · Release
  • ISOISO 22222Scope of personal financial planning

private bank

pwm:PrivateBank

Bank that delivers wealth management services to high-net-worth clients.

specialises
bankFIBO · Releasewealth manager
  • REGBanking Act (SG)s47: customer information may be disclosed only as the Act provides

wealth management service

pwm:WealthManagementService

Service combining investment advice or portfolio management with planning, custody, credit and estate services for a client.

specialises
wealth management serviceFIBO · Release
exact match
wealth management serviceFIBO · Release
  • ISOISO 22222Personal financial planning process

booking center

pwm:BookingCenterslow context

Role of a legal entity or branch of a wealth manager in which client accounts, contracts and assets are booked, and whose jurisdiction's rules apply to them.

specialises
party roleOMG Commons

custodian

pwm:Custodian

Financial institution that holds clients' securities for safekeeping.

specialises
custodianFIBO · Release
exact match
custodianFIBO · Release

Properties

assigned coverage

pwm:assignedCoverageobject property

Party role named in a coverage assignment: a relationship manager, or a product, credit, investment or planning specialist. The coverage role says which.

domain → range
coverage assignmentparty roleOMG Commons

residence country

pwm:residenceCountrydatatype property

Country where a natural person resides, as an ISO 3166-1 alpha-2 code.

domain → range
anystringXSD

domicile country

pwm:domicileCountrydatatype property

Country of a party's domicile, as an ISO 3166-1 alpha-2 code. Drives cross-border rules.

domain → range
anystringXSD
close match
is domiciled inOMG Commons

nationality country

pwm:nationalityCountrydatatype property

Country of a natural person's nationality, as an ISO 3166-1 alpha-2 code.

domain → range
anystringXSD

legal entity identifier

pwm:legalEntityIdentifierdatatype property

Twenty-character ISO 17442 identifier of a legal entity.

domain → range
anystringXSD
close match
legal entity identifierFIBO · Release

Code lists

coverage roles

pwm:CoverageRoleScheme

Used by coverageRole

CodeConceptDefinition
primaryprimarypwm:CoveragePrimaryLead relationship manager, accountable for the relationship.
secondarysecondarypwm:CoverageSecondaryBackup relationship manager.
specialistspecialistpwm:CoverageSpecialistProduct, credit, investment or planning specialist supporting the relationship.

politically exposed person categories

pwm:PEPCategoryScheme

Used by pepCategory

CodeConceptDefinition
foreign_pepforeign PEPpwm:ForeignPEPEntrusted with a prominent public function by a foreign country.
domestic_pepdomestic PEPpwm:DomesticPEPEntrusted domestically with a prominent public function.
international_organization_pepinternational organization PEPpwm:InternationalOrganizationPEPEntrusted with a prominent function by an international organization.
family_member_of_a_pepfamily member of a PEPpwm:PEPFamilyMemberRelated to a PEP by blood or marriage.
close_associate_of_a_pepclose associate of a PEPpwm:PEPCloseAssociateClosely connected to a PEP, socially or professionally.

Module 2

Due diligence, financial crime and tax

Customer due diligence, risk rating, source of wealth, screening and tax residency.

Classes

customer due diligence

pwm:CustomerDueDiligencemedium contextC2

Assessment that identifies and verifies a client and its beneficial owners, and establishes the purpose and intended nature of the relationship.

specialises
assessment
also called
CDD, KYC

enhanced due diligence

pwm:EnhancedDueDiligenceC3

Customer due diligence with additional measures, applied where the client presents higher risk, such as a politically exposed person.

also called
EDD

periodic review

pwm:PeriodicReviewC2

Scheduled re-assessment of a client's due diligence, at a frequency set by the client's risk rating, or earlier on a trigger event.

specialises
assessment

customer risk rating

pwm:CustomerRiskRatingC3

Assessment of the money-laundering and terrorist-financing risk a client presents, which sets due diligence depth and review frequency.

specialises
assessment
properties
amlRiskLevel

source of wealth assessment

pwm:SourceOfWealthAssessmentC3

Assessment of how a client accumulated their total wealth, with the evidence relied on.

specialises
assessment

source of funds assessment

pwm:SourceOfFundsAssessmentC3

Assessment of where specific funds or assets entering the relationship come from.

specialises
assessment

screening result

pwm:ScreeningResultC3

Outcome of screening a party against sanctions, politically exposed person and adverse-media sources, with its disposition.

specialises
assessment

tax residency self-certification

pwm:TaxResidencySelfCertificationC2

Client's declaration of tax residence countries and taxpayer identification numbers, collected at account opening and on change of circumstances.

specialises
assessment
  • REGOECD CRSSections IV and VI: self-certification for new accounts; Section VIII: definitions

Properties

Code lists

AML risk levels

pwm:AMLRiskLevelScheme

Used by amlRiskLevel

CodeConceptDefinition
Llowpwm:AMLLowSimplified measures may apply where permitted.
Mmediumpwm:AMLMediumStandard due diligence.
Hhighpwm:AMLHighEnhanced due diligence and senior approval.
Xoutside risk appetitepwm:AMLOutsideAppetiteRelationship declined or exited.

screening dispositions

pwm:ScreeningDispositionScheme

Used by screeningDisposition

CodeConceptDefinition
pending_reviewpending reviewpwm:ScreeningPendingAlert raised, not yet reviewed.
false_positivefalse positivepwm:ScreeningFalsePositiveReviewed; not the listed party.
true_matchtrue matchpwm:ScreeningTrueMatchReviewed; confirmed as the listed party.
escalatedescalatedpwm:ScreeningEscalatedReferred to compliance for decision.

Module 3

Investor profile and suitability

Regulatory investor classification, risk profile, knowledge and experience, sustainability preferences, and suitability and appropriateness.

Classes

investor classification

pwm:InvestorClassificationmedium contextC2

Regulatory categorization of a client under one jurisdiction's rules, such as retail, professional or accredited, with any opt-in or opt-out recorded.

specialises
assessment
related match
qualified investor restrictionFIBO · Release

investor risk profile

pwm:InvestorRiskProfilemedium contextC2

Versioned assessment of a client's risk tolerance, capacity for loss and investment horizon, expressed on the wealth manager's risk scale.

Each change produces a new version that supersedes the last. A suitability assessment cites the version in force at the time.

specialises
assessment

knowledge and experience assessment

pwm:KnowledgeAndExperienceAssessmentC2

Assessment of whether a client understands the risks of a type of product or service, from education, work and past investments.

specialises
assessment

financial situation assessment

pwm:FinancialSituationAssessmentC2

Assessment of a client's income, assets, liabilities and liquidity needs, including the ability to bear losses.

specialises
assessment

investment objective

pwm:InvestmentObjectiveC2

Financial objective a client states for their investments, used to judge whether an investment is appropriate for them.

specialises
investment objectiveFIBO · Release
exact match
investment objectiveFIBO · Release

suitability assessment

pwm:SuitabilityAssessmentmedium contextC2

Determination, before a recommendation or a discretionary decision, that a product or transaction fits the client's objectives, including any sustainability preferences, financial situation, and knowledge and experience, within the mandate.

Origination records why a client or product was proposed. Suitability records why it was permitted. Keep them separate.

specialises
assessment

appropriateness assessment

pwm:AppropriatenessAssessmentC2

Check, for services without advice, that a client has the knowledge and experience to understand the risks of the product.

specialises
assessment

suitability report

pwm:SuitabilityReportC2

Statement given to the client explaining how advice meets their preferences, objectives and other characteristics.

specialises
documentOMG Commons

client acknowledgement

pwm:ClientAcknowledgementC2

Record that a client received, read or accepted a disclosure, risk warning or document, and when.

specialises
EntityPROV-O

sustainability preference

pwm:SustainabilityPreferencemedium contextC2

Versioned record of a client's choice of one or more sustainability categories, minimum proportions and PAI elements, or an explicit absence of preference.

Proportions are decimals from 0 to 1. Categories may be combined; no preference is exclusive. This records preferences and does not certify product sustainability or implement the full Article 54 assessment.

specialises
EntityPROV-O

Properties

capacity for loss

pwm:capacityForLossobject property

Client's ability to bear losses without a material change to their standard of living or objectives.

domain → code list
anyrating levels
  • REGMiFID IIArt. 25(2): ability to bear losses

suitability outcome

pwm:suitabilityOutcomeobject property

Result of a suitability or appropriateness assessment.

domain → code list
anysuitability outcomes

investment horizon (years)

pwm:investmentHorizonYearsdatatype property

Length of time, in years, the client intends to hold the investments.

domain → range
anydecimalXSD

minimum taxonomy aligned proportion

pwm:minimumTaxonomyAlignedProportiondatatype property

Client minimum proportion invested in environmentally sustainable investments under Regulation (EU) 2020/852 Article 2(1), as a fraction from 0 to 1.

domain → range
sustainability preferencedecimalXSD

minimum SFDR sustainable proportion

pwm:minimumSFDRSustainableProportiondatatype property

Client minimum proportion invested in sustainable investments under Regulation (EU) 2019/2088 Article 2(17), as a fraction from 0 to 1.

domain → range
sustainability preferencedecimalXSD

PAI consideration elements

pwm:paiConsiderationElementsdatatype property

Client-determined qualitative or quantitative elements demonstrating consideration of principal adverse impacts on sustainability factors.

domain → range
sustainability preferencestringXSD

Code lists

investor categories

pwm:InvestorCategoryScheme

Two generic top concepts, retail and non-retail, with jurisdiction-specific categories beneath them. Thresholds stay in each jurisdiction's rules, not in the ontology.

Used by appliesToInvestorCategory, investorCategory, negativeTargetMarketCategory

CodeConceptDefinition
retailretailpwm:RetailCategoryClient given the full protections of the jurisdiction's conduct rules.
retail_investor_sgretail investor (SG)pwm:RetailInvestorSGNot an accredited, expert or institutional investor under the Securities and Futures Act.
private_client_chprivate client (CH)pwm:PrivateClientCHPrivate client under FinSA Art. 4.
retail_client_euretail client (EU)pwm:RetailClientEUClient that is not a professional client under MiFID II.
retail_client_ukretail client (UK)pwm:RetailClientUKRetail client under COBS 3.
non_retailnon-retailpwm:NonRetailCategoryClient treated as able to bear more risk, with some protections relaxed.
accredited_investor_sgaccredited investor (SG)pwm:AccreditedInvestorSGAccredited investor under s4A of the Securities and Futures Act, subject to the opt-in regime.
expert_investor_sgexpert investor (SG)pwm:ExpertInvestorSGExpert investor under s4A of the Securities and Futures Act.
institutional_investor_sginstitutional investor (SG)pwm:InstitutionalInvestorSGInstitutional investor under s4A of the Securities and Futures Act.
professional_investor_hkprofessional investor (HK)pwm:ProfessionalInvestorHKProfessional investor under the Securities and Futures Ordinance and the Professional Investor Rules.
professional_client_chprofessional client (CH)pwm:ProfessionalClientCHProfessional client under FinSA Art. 4.
institutional_client_chinstitutional client (CH)pwm:InstitutionalClientCHInstitutional client under FinSA Art. 4.
professional_client_euprofessional client (EU)pwm:ProfessionalClientEUProfessional client under MiFID II Annex II, per se or on request.
eligible_counterparty_eueligible counterparty (EU)pwm:EligibleCounterpartyEUEligible counterparty under MiFID II Art. 30.
professional_client_ukprofessional client (UK)pwm:ProfessionalClientUKProfessional client under COBS 3.
accredited_investor_usaccredited investor (US)pwm:AccreditedInvestorUSAccredited investor under Regulation D, Rule 501(a).

opt statuses

pwm:OptStatusScheme

Directions differ by jurisdiction. In Singapore an eligible investor opts in to accredited-investor status and gives up some retail protections. Under FinSA a high-net-worth retail client opts out to be treated as a professional client, and some professional clients opt in to be treated as retail. Record the resulting category on the classification.

Used by optStatus

CodeConceptDefinition
defaultdefaultpwm:OptDefaultCategory applies by default, with no election.
opted_inopted inpwm:OptedInClient made an opt-in election under the jurisdiction's rules. Which category that moves the client into, and whether protection rises or falls, depends on the jurisdiction, so record the classification it produced.
opted_outopted outpwm:OptedOutClient made an opt-out election under the jurisdiction's rules. Which category that moves the client into, and whether protection rises or falls, depends on the jurisdiction, so record the classification it produced.

suitability outcomes

pwm:SuitabilityOutcomeScheme

Used by suitabilityOutcome

CodeConceptDefinition
suitablesuitablepwm:SuitableFits the profile and mandate.
suitable_with_conditionssuitable with conditionspwm:SuitableWithConditionsFits if stated conditions are met, such as a size limit.
not_suitablenot suitablepwm:NotSuitableDoes not fit; may not be recommended.
insufficient_informationinsufficient informationpwm:InsufficientInformationCannot be assessed; no recommendation may be made.

illustrative risk scale

pwm:IllustrativeRiskScale

Placeholder five-point scale. An institution replaces it with its own scale in its extension.

Used by productRiskLevel, riskProfileLevel, targetRiskProfile

CodeConceptDefinition
RP1conservativepwm:RP1Capital preservation first.
RP2moderately conservativepwm:RP2Income with limited volatility.
RP3balancedpwm:RP3Growth and income in balance.
RP4growthpwm:RP4Growth with higher volatility.
RP5aggressivepwm:RP5Maximum growth; accepts large drawdowns.

sustainability preference categories

pwm:SustainabilityPreferenceScheme

Three categories under Article 2(7) of 2017/565, plus explicit no preference for recording.

Used by sustainabilityPreferenceKind

CodeConceptDefinition
taxonomy_aligned_minimumtaxonomy aligned minimumpwm:TaxonomyAlignedMinimumArticle 2(7)(a): minimum proportion of environmentally sustainable investments.
sfdr_sustainable_minimumSFDR sustainable minimumpwm:SFDRSustainableMinimumArticle 2(7)(b): minimum proportion of sustainable investments.
pai_considerationPAI considerationpwm:PAIConsiderationArticle 2(7)(c): principal adverse impacts considered using client-determined elements.
no_sustainability_preferenceno sustainability preferencepwm:NoSustainabilityPreferenceExplicit record that the client has no sustainability preference; not a fourth Article 2(7) category.

Module 4

Mandates, restrictions and fees

The service model agreed with the client, its investment restrictions and its charges.

Classes

mandate

pwm:Mandatemedium contextC2

Written agreement under which a wealth manager serves a client on a stated service model: advisory, discretionary, execution-only or custody-only. Investment restrictions apply where the service model calls for them.

specialises
written contractFIBO · Release
properties
hasRestriction

advisory mandate

pwm:AdvisoryMandate

Mandate under which the wealth manager makes personal recommendations and the client takes each investment decision.

specialises
mandate

discretionary mandate

pwm:DiscretionaryMandate

Mandate under which the wealth manager takes investment decisions for the client within agreed guidelines.

specialises
mandate
  • REGMiFID IIArt. 4(1)(8): portfolio management

execution-only arrangement

pwm:ExecutionOnlyArrangement

Arrangement under which the wealth manager executes client instructions without advice.

specialises
mandate

custody-only arrangement

pwm:CustodyOnlyArrangement

Arrangement under which the wealth manager holds assets and services them, without advice or dealing.

specialises
mandate

investment restriction

pwm:InvestmentRestrictionmedium context

Provision of a mandate that limits what may be held or done: asset-class bands, concentration limits, excluded issuers or sectors, prohibited instruments, leverage limits.

specialises
contractual commitmentFIBO · Release

concentration limit

pwm:ConcentrationLimit

Investment restriction capping exposure to one issuer, sector, country, currency or product type as a share of the portfolio.

model portfolio

pwm:ModelPortfolioslow contextC1

Target allocation, for a risk profile or strategy, from which client portfolios are built and against which drift is measured.

specialises
EntityPROV-O

benchmark

pwm:Benchmarkslow context

Index or composite against which a portfolio's performance is evaluated.

close match
portfolio benchmarkFIBO · Provisional

fee

pwm:Fee

Charge for services under a mandate or account, of a stated type.

specialises
feeFIBO · Release
properties
feeType
  • REGMiFID IIArt. 24(4): information on costs and charges

third-party payment

pwm:ThirdPartyPayment

Payment or non-monetary benefit the wealth manager receives from a third party, such as a product provider, in connection with a client service. It is not a charge to the client.

specialises
EntityPROV-O
also called
inducement, retrocession

Properties

governed by mandate

pwm:governedByMandateobject property

Mandate under which an account or portfolio is managed or advised.

domain → range
anymandate

fee type

pwm:feeTypeobject property

Kind of fee.

domain → code list
feefee types

has benchmark

pwm:hasBenchmarkobject property

Benchmark a mandate or portfolio is measured against.

domain → range
anybenchmark

Code lists

fee types

pwm:FeeTypeScheme

Used by feeType

CodeConceptDefinition
management_feemanagement feepwm:FeeManagementFor discretionary portfolio management.
advisory_feeadvisory feepwm:FeeAdvisoryFor advice under an advisory mandate.
custody_feecustody feepwm:FeeCustodyFor safekeeping and servicing of assets.
transaction_feetransaction feepwm:FeeTransactionPer transaction, such as brokerage.
performance_feeperformance feepwm:FeePerformanceContingent on returns above a hurdle or high-water mark.

Module 5

Accounts, portfolios and holdings

Where assets are held, what is held, and how it is valued.

Classes

account

pwm:Accountmedium contextC2

Record of a business arrangement between a client and a wealth manager, in which balances and transactions are kept.

specialises
accountFIBO · Release

investment account

pwm:InvestmentAccountC2

Account that records securities and cash held for investment.

specialises
accountinvestment accountFIBO · Release
  • ONTFIBOFBC/ProductsAndServices/ClientsAndAccounts/InvestmentAccount, Release; master source verified 2026-11-03
  • NEWPWM practitioner definitionSpecializes FIBO InvestmentAccount within the PWM account hierarchy

custody account

pwm:CustodyAccountC2

Investment account in which the wealth manager holds a client's securities in safekeeping.

also called
securities account

cash account

pwm:CashAccountfast contextC2

Current or deposit account in a single currency, linked to a client's investment activity.

specialises
accountcustomer accountFIBO · Releasedeposit accountFIBO · Release

joint account

pwm:JointAccountC2

Account held by two or more holders, operated under a stated signing rule.

specialises
account

account holder

pwm:AccountHolderC2

Role of a party that owns an account.

specialises
account holderFIBO · Release
exact match
account holderFIBO · Release

portfolio

pwm:Portfoliomedium contextC2

Collection of holdings managed or advised as a unit under one mandate.

specialises
portfolioFIBO · Release
properties
hasHolding

holding

pwm:Holdingfast contextC2

Quantity of a financial instrument or other asset held in a portfolio or account at a point in time.

specialises
holdingFIBO · Release

position

pwm:Positionfast contextC2

Financial exposure from owning, borrowing, shorting or contracting, including derivatives and credit.

specialises
positionFIBO · Release

valuation

pwm:Valuationfast contextC2

Monetary value of a portfolio, account or holding at a stated time, produced by a registered calculation service.

A figure presented as authoritative must cite its calculation service. A model may explain a valuation; it may not produce one.

specialises
EntityPROV-O
properties
valuationOf

Lombard facility

pwm:LombardFacilityfast contextC2

Credit extended to a client against a pledged portfolio of financial assets, with an advance rate per asset.

specialises
collateralized loanFIBO · Releaselending facility
also called
securities-backed loan
properties
securedBy

Properties

has holding

pwm:hasHoldingobject property

Holding that forms part of a portfolio.

domain → range
portfolioholding
close match
has memberOMG Commons
  • ONTOMG CommonsCollections/hasMember, verified in https://www.omg.org/spec/Commons/20250801/Collections.rdf; portfolio membership is narrower
  • NEWPWM practitioner definitionPortfolio composition

signing rule

pwm:signingRuleobject property

Rule for who must sign instructions on a joint account or for an entity.

domain → code list
anysigning rules

amount

pwm:amountdatatype property

Numeric amount, read together with its currency.

domain → range
anydecimalXSD
close match
has amountFIBO · Release

currency

pwm:currencydatatype property

Currency of an amount, as an ISO 4217 alphabetic code.

domain → range
anystringXSD
close match
has currencyFIBO · Release

Code lists

signing rules

pwm:SigningRuleScheme

Used by signingRule

CodeConceptDefinition
any_one_to_signany one to signpwm:SignAnyOneAny single authorized signatory may instruct.
two_to_signtwo to signpwm:SignTwoJointlyAny two authorized signatories together.
all_to_signall to signpwm:SignAllJointlyEvery holder or signatory must sign.

Module 6

Products, distribution and research

The product shelf, product risk and target market, cross-border distribution rules, offering documents and the house view.

Classes

product

pwm:Productslow contextC1

Financial product on the wealth manager's shelf, such as a fund, structured note, bond, deposit or credit facility, linked to the instruments that underlie it.

specialises
financial productFIBO · Release
  • ONTFIBOFBC/ProductsAndServices/FinancialProductsAndServices/FinancialProduct, Release; master source verified 2026-11-03
  • NEWPWM practitioner definitionProduct shelf

product risk rating

pwm:ProductRiskRatingslow contextC1

Wealth manager's assessment of a product's risk on its own scale, used to match products to client risk profiles.

specialises
assessment
related match
risk levelFIBO · Provisional
  • REGMiFID IIArt. 16(3), 24(2): product governance

target market

pwm:TargetMarketslow contextC1

Description of the clients a product is designed for, and those it is not, by category, knowledge, risk and objectives.

specialises
EntityPROV-O

cross-border rule

pwm:CrossBorderRuleslow contextC1

Deterministic rule stating whether an activity, such as solicitation, advice or research distribution, is permitted for a client domicile, booking center, investor category and product.

A rule table evaluated outside the model. It is not AI under MAS AIRG para 1.3 unless a model takes part in the decision. Activity, client domicile and booking center are required; an absent investor category or product means all. When rules overlap, the most specific rule (most scope values set) decides; on a tie, not permitted wins; with no matching rule the activity is not permitted (fail closed).

specialises
policy rule
related match
investors' domicile restrictionFIBO · Release
related match
qualified investor restrictionFIBO · Release

product term sheet

pwm:ProductTermSheetslow contextC0

Offering document summarizing a product's terms, payoff, risks and costs.

specialises
offering documentFIBO · Release

research publication

pwm:ResearchPublicationslow contextC1

Authored, dated and versioned investment research, with effective, last-reviewed, superseded and expiry dates and any distribution restrictions.

specialises
documentOMG Commons

house view

pwm:HouseViewslow contextC1

Wealth manager's current official view on markets and asset allocation. When it changes, recommendations that depend on it are invalidated.

also called
CIO view
  • KOGKognita roadmap0.6 item 4: house-view change propagation
  • AMEVOLVEDefine truth once: one source for policies

Properties

product risk level

pwm:productRiskLevelobject property

Level a product risk rating assigns, on a scale that can be compared with client risk profile levels.

Use the same scale as the risk profile, or a published mapping between the two, so a suitability check can compare them.

applies to product

pwm:appliesToProductobject property

Product a rule covers. Absent when the rule covers all products.

domain → range
anyproduct

applies to investor category

pwm:appliesToInvestorCategoryobject property

Investor category a rule covers. Absent when the rule covers all investor categories.

domain → code list
anyinvestor categories

ISIN

pwm:isindatatype property

Twelve-character ISO 6166 international securities identification number.

domain → range
anystringXSD

client domicile code

pwm:clientDomicileCodedatatype property

Client domicile a rule covers, as an ISO 3166-1 alpha-2 code. Required: a rule without a domicile is incomplete, not a rule for every domicile.

domain → range
cross-border rulestringXSD

booking center code

pwm:bookingCenterCodedatatype property

Booking center jurisdiction a rule covers, as an ISO 3166-1 alpha-2 code. Required: a rule without a booking center is incomplete, not a rule for every booking center.

domain → range
cross-border rulestringXSD

Code lists

product complexity classes

pwm:ProductComplexityScheme

Used by productComplexity

CodeConceptDefinition
complexcomplexpwm:ComplexProductRequires an assessment of knowledge and experience before sale.
specified_investment_product_sgspecified investment product (SG)pwm:SpecifiedInvestmentProductSGProduct that MAS requires a customer knowledge or account assessment for before sale to retail customers.
non_complexnon-complexpwm:NonComplexProductMay be sold without an appropriateness check where the rules allow.
excluded_investment_product_sgexcluded investment product (SG)pwm:ExcludedInvestmentProductSGProduct outside the specified-investment-product regime.

cross-border activities

pwm:CrossBorderActivityScheme

Used by appliesToActivity

CodeConceptDefinition
solicitationsolicitationpwm:ActivitySolicitationApproaching a client or prospect unprompted.
reverse_enquiryreverse enquirypwm:ActivityReverseEnquiryResponding to a request the client initiated.
adviceadvicepwm:ActivityAdviceGiving a personal recommendation.
executionexecutionpwm:ActivityExecutionExecuting a client instruction.
marketingmarketingpwm:ActivityMarketingPromoting a product or service.
research_distributionresearch distributionpwm:ActivityResearchDistributionSending research to a client.

Module 7

Interactions, advice and communication

Why a client was approached, what was recommended, who decided, and what reached the client.

Classes

client interaction

pwm:ClientInteractionmedium contextC2

One client journey, from its trigger through retrieval, recommendation, review and decision to any communication, recorded under a single identifier.

specialises
ActivityPROV-O

origination

pwm:OriginationC2

Step that selects a client or product for attention, recorded with its trigger, selection criteria, candidate set and reasons, before anything is shown to a relationship manager.

An origination may conclude that no contact is recommended, with a reason.

specialises
ActivityPROV-O
  • KOGKognita roadmap0.4 item 10: origination evidence; do not contact is a first-class outcome

recommendation

pwm:Recommendationmedium contextC2

Personal recommendation to a client about a product, transaction or allocation, with structured rationale and cited sources.

specialises
EntityPROV-O
  • REGMiFID IIArt. 4(1)(4): investment advice
  • REGMAS FAA-N16Recommendations on investment products
  • KOGKognita roadmap0.4 item 10: rationale as structured fields with citations

final decision

pwm:FinalDecisionC2

Named person's decision on an AI-prepared recommendation or action, approving, modifying, rejecting or escalating it, with what they saw and what they changed.

specialises
EntityPROV-O
properties
decisionOn

client communication

pwm:ClientCommunicationC2

Message sent to a client through a channel, classified by communication level, carrying its content provenance and any AI disclosure, and traceable to the final decision it came from.

specialises
EntityPROV-O

communication channel

pwm:CommunicationChannelslow context

Means of reaching a client, registered as approved only if it meets archiving and supervision requirements.

Properties

with client

pwm:withClientobject property

Client an interaction, recommendation or communication concerns.

domain → range
anyclient

part of interaction

pwm:partOfInteractionobject property

Client interaction an origination, recommendation, decision, communication or evidence record belongs to.

domain → range
anyclient interaction
  • KOGKognita roadmap0.4 item 9: one interaction_id spans a whole client journey

recommends product

pwm:recommendsProductobject property

Product a recommendation proposes.

domain → range
recommendationproduct

decision result

pwm:decisionResultobject property

Approve, modify, reject or escalate.

domain → code list
anyfinal decision results
  • KOGKognita roadmap0.4 item 11: outcomes are approve, modify, reject or escalate

derived from decision

pwm:derivedFromDecisionobject property

Final decision a client communication traces back to. A communication without one is refused.

sub-property of
wasDerivedFromPROV-O
  • KOGKognita roadmap0.4 item 12: a communication that does not trace back to a final decision is a DENY

content provenance

pwm:contentProvenanceobject property

Who produced content: a person, AI, AI then edited by a person, or a person with AI assistance. The label survives edits.

domain → code list
anycontent provenance labels

disclosure trigger

pwm:disclosureTriggerobject property

Reason recorded for an AI disclosure: what the AI did that the disclosure describes. Whether disclosure is required is decided by the jurisdiction's disclosure rule and the content provenance.

domain → code list
anyAI disclosure triggers

is approved channel

pwm:isApprovedChanneldatatype property

Whether a channel is registered as meeting archiving and supervision requirements.

domain → range
communication channelbooleanXSD

content hash

pwm:contentHashdatatype property

Cryptographic hash of content as rendered, sent or stored, so the content can be proved later without being copied into the record.

domain → range
anystringXSD

Code lists

communication levels

pwm:CommunicationLevelScheme

Ordered least to most consequential. A higher level keeps every control of the levels below it.

Used by communicationLevel

CodeConceptDefinition
1general informationpwm:CommGeneralInformationInformation not specific to the client.
2house viewpwm:CommHouseViewThe institution's published view, unadapted.
3contextualized insightpwm:CommContextualizedInsightInformation related to the client's circumstances, without a recommendation.
4investment discussionpwm:CommInvestmentDiscussionDiscussion of options for the client's portfolio.
5product recommendationpwm:CommProductRecommendationPersonal recommendation of a product; triggers suitability checks and approval.
6transactionpwm:CommTransactionCommunication that confirms or effects a transaction.

content provenance labels

pwm:ContentProvenanceScheme

Used by contentProvenance

CodeConceptDefinition
humanhumanpwm:ProvenanceHumanWritten by a person without AI.
aiAIpwm:ProvenanceAIGenerated by AI.
ai_then_human_editedAI then human-editedpwm:ProvenanceAIHumanEditedGenerated by AI, then edited by a person; stays identifiable as AI-originated.
human_with_ai_assistancehuman with AI assistancepwm:ProvenanceHumanAIAssistedWritten by a person with AI help.

AI disclosure triggers

pwm:DisclosureTriggerScheme

Used by disclosureTrigger

CodeConceptDefinition
ai_generated_contentAI generated contentpwm:TriggerAIGeneratedContentThe content was generated by AI.
assisted_analysisassisted analysispwm:TriggerAssistedAnalysisAI assisted the analysis behind the content.
powers_a_chatbotpowers a chatbotpwm:TriggerPowersChatbotThe client is talking to an AI system.
materially_contributed_to_a_decisionmaterially contributed to a decisionpwm:TriggerMaterialContributionAI materially contributed to a decision affecting the client.

origination triggers

pwm:OriginationTriggerScheme

Used by originationTrigger

CodeConceptDefinition
eventeventpwm:OriginationEventA client or market event, such as a maturity or a house-view change.
scheduleschedulepwm:OriginationScheduleA scheduled review.
rm_requestRM requestpwm:OriginationRMRequestA relationship manager asked for it.

do-not-contact reasons

pwm:DoNotContactReasonScheme

Used by doNotContactReason

CodeConceptDefinition
low_relevancelow relevancepwm:DNCLowRelevanceNot relevant enough to the client.
poor_timingpoor timingpwm:DNCPoorTimingWrong moment for the client.
vulnerabilityvulnerabilitypwm:DNCVulnerabilityA vulnerability indicator is open.
excessive_frequencyexcessive frequencypwm:DNCExcessiveFrequencyContact-frequency limit reached.
a_recent_rejectiona recent rejectionpwm:DNCRecentRejectionThe client recently declined a similar proposal.
a_relationship_circumstancea relationship circumstancepwm:DNCRelationshipCircumstanceA circumstance in the relationship makes contact inappropriate.

final decision results

pwm:DecisionResultScheme

Used by decisionResult

CodeConceptDefinition
approveapprovepwm:DecisionApproveAccepted as prepared.
modifymodifypwm:DecisionModifyAccepted with changes, recorded as a structured diff.
rejectrejectpwm:DecisionRejectNot accepted.
escalateescalatepwm:DecisionEscalateReferred to a more senior or specialist decision-maker.

Module 8

AI use cases, authority and accountability

The AI inventory, risk materiality, agents, delegated authority, autonomy and the people accountable for them.

Classes

AI use case

pwm:AIUseCaseslow contextC1

Specific real-world context that an AI model or system is applied to, registered as a versioned inventory entry with its purpose, approved scope including jurisdiction, data, dependencies, lifecycle status, risk materiality, review status and accountable people.

The use case, not the model, is the unit of risk assessment. Every policy decision cites the use case it was taken under; an unregistered or retired use case is refused.

specialises
EntityPROV-O
  • ISOISO/IEC 42001AI management system
  • REGMAS AIRG1.4(c) use case; 4.5 to 4.9 inventory; 4.7 attributes
  • REGMAS AIRG2.5(b): permitted and prohibited uses are the use-case register's lifecycle status; 2.5(c): approved AI and an approval process for new use
  • KOGKognita roadmap0.4 item 8: use-case register
  • AMFiling Cabinet talkSlide 19: an unregistered use case is refused

AI system

pwm:AISystemslow context

Machine-based system of one or more models and other components that derives outputs, such as predictions, content, recommendations or decisions, from learned premises.

Tools whose outputs are fully determined by manually specified rules, and robotic process automation without learning, are not AI for this purpose.

specialises
EntityPROV-O
close match
AI SystemDPV-AI

AI model

pwm:AIModelslow context

Method that converts assumptions and input data into outputs such as estimates, decisions or recommendations, pinned to a provider and version.

specialises
EntityPROV-O
close match
ModelDPV-AI

AI agent

pwm:AIAgentslow context

Software agent that plans and takes actions with tools on behalf of a person, holding its own credentials, a maximum autonomy level, a tool allow-list, guardrails and a deployment version that covers prompt, tools, permissions and data.

An agent can never hold an accountability role. Its entitlements are always narrower than those of the person it acts for.

specialises
SoftwareAgentPROV-Osoftware agentOMG Commons
close match
AI AgentDPV-AI

third-party AI

pwm:ThirdPartyAI

AI system, model, or service using AI, that an external party develops, owns, operates or provides and the institution procures, accesses or relies on, including AI embedded in software services.

specialises
AI system
  • REGMAS AIRGfn 10 (definition); 4.2 and fn 21 (embedded AI); 5.10, 5.11 (management)

model provider

pwm:ModelProvider

Organization that supplies AI models or AI services, tracked for concentration, change notification and exit.

specialises
service providerOMG Commons

model card

pwm:ModelCardslow context

Record of an approved model: provider, name and version, intended use, known limitations, the evaluation relied on, approval date and approver.

specialises
documentOMG Commons
  • REGMAS AIRG5.10: documentation of how third-party AI was tested
  • KOGKognita roadmap0.4 item 8: model card per approved model

risk materiality assessment

pwm:RiskMaterialityAssessmentslow context

Assessment of an AI use case on impact, complexity and reliance, including the autonomy granted, recorded before controls (inherent) and after them (residual), and approved by the designated control function.

Residual materiality must be within risk appetite before deployment. Kognita's six axes (business criticality, client-impact class, decision consequence, data sensitivity, autonomy, complexity) roll up into the three MAS dimensions.

specialises
assessment

AI risk appetite

pwm:AIRiskAppetiteslow context

Board-approved statement of the AI uses the institution will and will not accept, with quantitative measures such as AI incident counts, material use cases depending on one provider, and use cases breaching performance thresholds.

specialises
EntityPROV-O
  • REGMAS AIRG3.4(b) and fn 18
  • KOGKognita roadmap0.4 item 8: AI risk appetite as top-level policy; quantitative measures

accountability role

pwm:AccountabilityRole

Role of a named person accountable for an AI use case. Accountability can be supported by functions but cannot be delegated to an agent.

specialises
party roleOMG Commons
restriction
is played byOMG Commons only anything except AI agent

business owner

pwm:BusinessOwner

Accountability role answerable for an AI use case's business outcome and residual risk acceptance. Every decision under the use case cites this person.

also called
accountable owner
  • REGMAS AIRG5.23(c): accountable person for monitoring and incidents
  • KOGKognita roadmap0.4 item 8: a use case with no current owner is a DENY

Knowledge Lead

pwm:KnowledgeLead

Named owner of an AI workflow's quality standard: defines the standard and primary metric, approves consequential outputs, receives escalations of ambiguity and exceptions, and approves changes to the standard, including those learned from operations.

designated control function

pwm:DesignatedControlFunction

Control function, independent of a use case's business and technology owners, that is final arbiter of whether a use constitutes AI, owns inventory policy, and approves risk materiality ratings.

independent validator

pwm:IndependentValidator

Person or function, independent of development and deployment, that validates a high-materiality AI use case before deployment and periodically afterwards.

  • REGMAS AIRG5.18 to 5.21 (pre-deployment review); 5.24 (re-validation)
  • KOGKognita roadmap0.4 item 8: independent validation as separation of duties

delegated authority

pwm:DelegatedAuthorityslow contextC2

Authorization that binds a person, an agent, a purpose, a client, permitted actions, permitted data, a period and limits. Specific, temporary, revocable, and ended by task completion as well as by time.

A request is allowed only if the person holds the entitlement and the agent holds a live delegation for it.

specialises
authorizationOMG Commons

autonomy promotion gate

pwm:AutonomyPromotionGateslow context

Evidence thresholds an agent must meet over a minimum sample and window before its autonomy level may rise, and below which it steps down automatically.

specialises
EntityPROV-O

purpose

pwm:Purposeslow context

Declared reason for processing data or invoking AI. Access is bound to it; a request without a registered purpose is refused.

close match
PurposeDPV

kill switch

pwm:KillSwitchslow context

Mechanism that deactivates an AI system expeditiously when it exceeds risk tolerances, effective on the next decision, with a tested activation protocol and fallback.

kill-switch drill

pwm:KillSwitchDrill

Test activation of a kill switch, recording who activated it, how long until the next decision was denied, and how the fallback was reached.

specialises
ActivityPROV-O

authority live status

pwm:AuthorityLiveStatusfast contextC2

Live observation of the revocation state of a DelegatedAuthority or the engaged state of a KillSwitch, effective from asOf until superseded. Read from the system of record at decision time, never from cache.

The single live-status vocabulary for delegation revocation and kill-switch state, retained for reuse in 0.5 without a parallel class or code list. DelegatedAuthority and KillSwitch definitions remain in SlowContext. For each subject, use the latest observation at or before the decision time. Conflicting observations at that time fail closed. SHACL checks recorded observations; runtime must obtain fresh state.

specialises
EntityPROV-O
  • REGMAS AIRG5.9: human oversight
  • KOGKognita roadmap0.4 item 14: circuit breaker and granular revocation; 0.3 item 2: policy evaluation boundaries
  • KOGKognita roadmaproadmap milestone 0.5 items 2 and 16; Ahmed design decisions 3 Nov 2026
  • NEWPWM practitioner definition0.4: NEW specialized record or field; no exact standard term identified

Properties

has purpose

pwm:hasPurposeobject property

Registered purpose of a use case, delegation or request.

domain → range
anypurpose

uses data class

pwm:usesDataClassobject property

Data classification a use case is permitted to process.

domain → code list
anydata classifications
  • REGMAS AIRG4.7: data used; 5.4(d): data classification

risk tier

pwm:riskTierobject property

Risk tier derived from a use case's classification, which sets its minimum controls and maximum autonomy.

domain → code list
AI use caseuse-case risk tiers

impact

pwm:impactobject property

Impact rating: consequence of failure or poor performance for the institution, its customers and stakeholders, including the sensitivity of the data processed.

complexity

pwm:complexityobject property

Complexity rating: from the technology, novelty, data and explainability, and for third-party AI the visibility into it.

reliance

pwm:relianceobject property

Reliance rating: how far the decision or output rests on AI, including the autonomy granted and the degree of human oversight.

has accountability role

pwm:hasAccountabilityRoleobject property

Links a use case or agent to an accountability role held for it. The person holding the role is given on the role, with is played by.

domain → range
anyaccountability role

registered for use case

pwm:registeredForUseCaseobject property

Use case an agent is registered to act under.

domain → range
AI agentAI use case

maximum autonomy level

pwm:maxAutonomyLevelobject property

Highest autonomy level an agent may operate at, capped by its use case's risk tier.

domain → code list
AI agentagent autonomy levels

required autonomy level

pwm:requiredAutonomyLevelobject property

Autonomy level a tool requires. A call above the agent's level is refused; a call at L4 routes to human approval.

domain → code list
toolagent autonomy levels

action verb

pwm:actionVerbobject property

Action a tool performs, or a delegation permits.

domain → code list
anyaction verbs

permitted tool

pwm:permittedToolobject property

Tool on an agent's allow-list.

domain → range
AI agenttool

boundary tier

pwm:boundaryTierobject property

Whether an action is always permitted, needs asking first, or is never permitted for an agent.

domain → code list
anyboundary tiers

delegated by

pwm:delegatedByobject property

Person who grants a delegated authority. They must hold the entitlement themselves.

domain → range
delegated authoritypersonFIBO · Release

about client

pwm:aboutClientobject property

Client a delegation, claim or fact is about.

domain → range
anyclient

observed by

pwm:observedByobject property

Agent that obtained the live observation from the system of record.

domain → range
authority live statusAgentPROV-O

approved jurisdiction

pwm:approvedJurisdictiondatatype property

Jurisdiction in which a use case is approved for use, as an ISO 3166-1 alpha-2 code.

domain → range
AI use casestringXSD
  • REGMAS AIRG4.7: approved scope of use (e.g., jurisdiction)

model type

pwm:modelTypedatatype property

Kind of model a use case relies on, such as a classifier, ranking model or large language model.

domain → range
AI use casestringXSD

is mediated

pwm:isMediateddatatype property

Whether a use case runs through the institution's AI control plane. Unmediated entries record the assurance gap and compensating control.

domain → range
AI use casebooleanXSD

agent identifier

pwm:agentIdentifierdatatype property

Readable, unique identifier of an agent, such as AGENT-RM-PRECALL-01.

domain → range
AI agentstringXSD

deployment version

pwm:deploymentVersiondatatype property

Hash over an agent's system prompt, instructions, tools, permissions, data sources, orchestration and memory behavior. Any change needs re-approval before the agent can act.

domain → range
AI agentstringXSD

ends on task completion

pwm:endsOnTaskCompletiondatatype property

Whether a delegation is revoked when its task completes, not only when its period ends.

domain → range
delegated authoritybooleanXSD

approval threshold

pwm:approvalThresholddatatype property

Statement of which actions under a delegation require human approval.

domain → range
delegated authoritystringXSD

quality threshold

pwm:qualityThresholddatatype property

Minimum share of outputs meeting the agreed standard, as a fraction from 0 to 1.

domain → range
autonomy promotion gatedecimalXSD
  • AMEVOLVEQuality at least 95 percent

exception rate threshold

pwm:exceptionRateThresholddatatype property

Maximum share of runs that leave the always-do path, as a fraction from 0 to 1.

domain → range
autonomy promotion gatedecimalXSD
  • AMEVOLVEException rate below 5 percent

requires full traceability

pwm:requiresFullTraceabilitydatatype property

Whether every action in the evaluation window must be reconstructable for the gate to pass.

domain → range
autonomy promotion gatebooleanXSD
  • AMEVOLVEEvidence fully traceable

baseline value

pwm:baselineValuedatatype property

Measured value of an outcome metric before the use case.

domain → range
outcome metricdecimalXSD

target value

pwm:targetValuedatatype property

Agreed target for an outcome metric.

domain → range
outcome metricdecimalXSD
  • AMEVOLVETargets agreed with the process owner

Code lists

agent autonomy levels

pwm:AutonomyLevelScheme

Early private-banking deployments are expected at L1 to L4. Raising a level is a governed change needing the accountable owner and independent validation. The four-step gate in the Filing Cabinet talk (suggest, draft, act with approval, act alone) corresponds to L2, L3, L4 and L5 or above.

Used by effectiveAutonomyLevel, fromAutonomyLevel, maxAutonomyLevel, requiredAutonomyLevel, scopeAutonomyLevel, toAutonomyLevel

CodeConceptDefinition
L0retrievepwm:L0RetrieveFetch authorized information.
L1explainpwm:L1ExplainSummarize or explain retrieved information.
L2recommendpwm:L2RecommendPropose an option. Recommending never implies permission to execute.
L3draftpwm:L3DraftPrepare an artifact a person reviews before use.
L4act after approvalpwm:L4ActAfterApprovalTake an action once a named person, not the requester, has approved that exact action.
L5bounded autonomous actionpwm:L5BoundedAutonomyAct without per-action approval inside explicit limits.
L6broad autonomypwm:L6BroadAutonomyAct across tasks with broad discretion.

boundary tiers

pwm:BoundaryTierScheme

Set per action for each agent, independently of the autonomy-level scale. As a rule of thumb, internal work at L0 to L3 sits in always do, and L4 in ask first.

Used by boundaryTier

CodeConceptDefinition
always_doalways dopwm:AlwaysDoSafe, reversible, inside the mandate and already authorized, such as retrieving authorized data or drafting an internal briefing.
ask_firstask firstpwm:AskFirstConsequential or client-facing; the agent prepares and a person decides.
never_donever dopwm:NeverDoOutside this agent's scope, whatever it is asked, enforced in its permissions rather than by asking. Which actions sit here is set per agent, for example sending client communications for an agent that only drafts.

action verbs

pwm:ActionVerbScheme

Used by actionVerb

CodeConceptDefinition
readreadpwm:VerbReadRead information.
analyzeanalyzepwm:VerbAnalyzeDerive insight from information.
recommendrecommendpwm:VerbRecommendPropose an option.
draftdraftpwm:VerbDraftPrepare an artifact.
queuequeuepwm:VerbQueuePlace an artifact for release by a person.
sendsendpwm:VerbSendRelease an artifact outside the institution.
executeexecutepwm:VerbExecuteChange a system of record or move value.
  • KOGKognita roadmap0.5 item 3: read, analyze, recommend, draft, queue, send and execute

use-case lifecycle statuses

pwm:UseCaseStatusScheme

Used by agentStatus, lifecycleStatus

CodeConceptDefinition
pilotpilotpwm:StatusPilotLimited by end date, named users and success criteria.close match: Deployment StageDPV-AI
livelivepwm:StatusLiveIn production use.close match: Operation StageDPV-AI
suspendedsuspendedpwm:StatusSuspendedStopped pending review; decisions are refused.
retiredretiredpwm:StatusRetiredDecommissioned; its agents are stopped and retention applies.close match: Decommission StageDPV-AI

use-case risk tiers

pwm:RiskTierScheme

Used by riskTier

CodeConceptDefinition
1tier 1pwm:Tier1Research summarization.
2tier 2pwm:Tier2Client intelligence and meeting preparation.
3tier 3pwm:Tier3KYC, suitability and recommendations.
4tier 4pwm:Tier4Financial execution.

client-impact classes

pwm:ClientImpactClassScheme

Used by clientImpactClass

CodeConceptDefinition
1internal productivitypwm:ImpactInternalProductivityStaff productivity with no client reach.
2RM assistancepwm:ImpactRMAssistanceHelps a relationship manager prepare.
3client influencepwm:ImpactClientInfluenceShapes what a client is told or offered.
4client communicationpwm:ImpactClientCommunicationProduces content that reaches a client.
5advicepwm:ImpactAdviceContributes to a personal recommendation.
6executionpwm:ImpactExecutionContributes to executing a transaction.

decision consequences

pwm:DecisionConsequenceScheme

Used by decisionConsequence

CodeConceptDefinition
informationalinformationalpwm:ConsequenceInformationalInforms only.
operationaloperationalpwm:ConsequenceOperationalChanges an internal operation.
client_communicationclient communicationpwm:ConsequenceClientCommunicationDetermines what reaches a client.
suitabilitysuitabilitypwm:ConsequenceSuitabilityBears on whether advice is suitable.
financial_actionfinancial actionpwm:ConsequenceFinancialActionMoves value or changes a position.

use-case autonomy axis

pwm:AutonomyAxisScheme

Used by autonomyAxis

CodeConceptDefinition
nonenonepwm:AutonomyNoneNo autonomous action.
assistedassistedpwm:AutonomyAssistedAssists a person who acts.
conditionalconditionalpwm:AutonomyConditionalActs under conditions or approval.
highhighpwm:AutonomyHighActs with little human involvement.

primary outcome metrics

pwm:PrimaryMetricScheme

Used by primaryMetric

CodeConceptDefinition
cycle_timecycle timepwm:MetricCycleTimeTime to complete the workflow; breaches above target.
decision_qualitydecision qualitypwm:MetricDecisionQualityShare of outputs meeting the standard; breaches below target.
reliabilityreliabilitypwm:MetricReliabilityShare of runs completing without failure; breaches below target.
knowledge_retentionknowledge retentionpwm:MetricKnowledgeRetentionCapability that remains when people leave. An EVOLVE measure; not a Kognita primary metric.

agentic governance risks

pwm:GovernanceRiskScheme

Five risks that sit on top of model risk when AI moves from assistance to action.

CodeConceptDefinition
accessaccesspwm:RiskAccessAn agent inherits whatever its tools can reach.
evidenceevidencepwm:RiskEvidenceThe chain of retrievals, judgments and actions is not kept.
operationaloperationalpwm:RiskOperationalAgents loop, stall, call the wrong tool or keep moving when they should stop.
changechangepwm:RiskChangeBehavior changes with a prompt, tool, source or silent model update.
ownershipownershippwm:RiskOwnershipNo named person is accountable for what the agent did.

Module 9

Information governance, claims and evidence

Governed documents and definitions, facts and claims, policy decisions and the evidence that reconstructs them.

Classes

governed information asset

pwm:GovernedInformationAssetslow context

Document, record or dataset carrying the minimum governance labels that make it safe to give to AI: owner, approval status, effective date, what it supersedes, classification per section, version identifier or content hash, and retention date.

When content is deleted, keep its hash and record the deletion.

specialises
documentOMG Commons
properties
hasSection

document section

pwm:DocumentSectionslow context

Section of a governed information asset with its own classification, so entitlements apply per section rather than per document.

specialises
documentOMG Commons

approved source

pwm:ApprovedSourceslow context

Source registered as approved for a use case's regulated claims, such as CIO research, product documentation, permitted market data or policy. A model's own knowledge is never an approved source.

specialises
EntityPROV-O

governed definition

pwm:GovernedDefinitionslow contextC1

Authoritative, versioned definition of a business term, such as assets under management, net new money or concentration, with an owner. Figures using the term must cite it.

specialises
EntityPROV-O
properties
definesTerm
  • KOGKognita roadmap0.6 item 12: governed business definitions
  • AMEVOLVEStage 5: define truth once (one definition of AUM)

calculation service

pwm:CalculationService

Registered service that produces authoritative figures, such as performance, valuation, concentration, exposure or suitability scores. A model may explain a calculation; it may not invent one.

specialises
SoftwareAgentPROV-O

fact

pwm:Factmedium contextC2

Authoritative value from a system of record, carrying its source, owner, as-of time, classification, entitlement and freshness threshold. Stale facts are refreshed, suppressed or flagged, never silently used.

specialises
EntityPROV-O

claim

pwm:Claimmedium contextC2

Material statement an agent produces for a person or client, typed as verified fact, inference, conversation hypothesis or possible bank capability, with its sources, calculation method, model and prompt versions and validator status.

Unlabeled claims cannot be shown. An inference cannot be relabeled as fact without a cited source and verification.

specialises
EntityPROV-O

policy rule

pwm:PolicyRuleslow contextC1

Effective-dated, citable rule evaluated outside the model before data moves or an action runs. Rules are versioned, never edited in place.

specialises
EntityPROV-O
properties
ruleCitation

policy decision

pwm:PolicyDecisionmedium contextC1

Deterministic outcome of evaluating every applicable rule for a request, allow, deny, escalate or hold for human approval, citing the deciding rule and its version, the use case and any delegation relied on. One failing rule denies.

specialises
EntityPROV-O

evidence record

pwm:EvidenceRecordmedium contextC1

Append-only, hash-chained record of an event, such as a policy decision, model call, tool call, retrieval, approval or communication, holding hashes and references rather than content, from which an interaction can be reconstructed.

Record a model refusal separately from a policy denial. Otherwise the evidence shows only that the model usually behaves, not that the controls work.

specialises
EntityPROV-O

human review

pwm:HumanReview

Review of AI output or a proposed action by a competent person with authority to intervene, given a complete packet of action, reason, evidence, risks, alternatives and checks run.

specialises
ActivityPROV-O
  • REGMAS AIRG2.5(b): mandatory human review before use
  • REGMAS AIRG5.9(a) to (d): human oversight; automation bias and decision fatigue
  • KOGKognita roadmap0.4 item 11 (approval packet); item 13 (risk-based review)

AI incident

pwm:AIIncidentC2

Incident involving a failure, malfunction or poor performance of an AI system or model, escalated, tracked and linked to the clients affected.

specialises
EntityPROV-O
properties
incidentKind
  • REGMAS AIRGfn 20 (definition); 3.5(e) escalation; 5.23(b) incident management
  • KOGKognita roadmap0.4 item 14 (affected-client query); 0.7 (incident evidence and report)

near miss

pwm:NearMissC1

Event where harm was avoided, such as a blocked action, a verifier block, a reviewer rejection, an unsafe draft or unusual agent behavior, aggregated to find weak controls before harm occurs.

specialises
EntityPROV-O

memory record

pwm:MemoryRecordmedium contextC2

Item an agent retains beyond a single step, scoped as working, relationship or institutional memory and typed like a claim, with retention, a correction path and expiry.

An inference becomes a relationship record only after verification by a person. Client data never enters institutional memory.

specialises
EntityPROV-O
  • KOGKognita roadmap0.6 item 8: memory governance
  • AMEVOLVEClosed-loop learning; human expertise becomes institutional capability

run

pwm:Runmedium contextC2

Governed execution grouping calls across agent steps, budget snapshots, pending approvals and lifecycle transitions.

specialises
ActivityPROV-O

run budget

pwm:RunBudgetmedium contextC2

Limits and consumption snapshot for one run. Cost is USD; wall clock is seconds. Evidence records successive snapshots rather than overwriting history.

specialises
EntityPROV-O

run checkpoint

pwm:RunCheckpointmedium contextC2

Durable snapshot at a synchronous policy evaluation boundary; contentHash is the continuation handle in the retention store.

specialises
EntityPROV-O

approval resolution

pwm:ApprovalResolutionmedium contextC2

Human resolution of a specific approval request, approved or denied, with attribution and time.

specialises
EntityPROV-O

run resumption

pwm:RunResumptionmedium contextC2

Evidenced continuation of a checkpointed run after a human resolution; denied resolutions never authorize the held action.

specialises
ActivityPROV-O

model call

pwm:ModelCallmedium contextC2

Gateway model invocation recording provider-reported identity, prompt as sent and response as received, template version and token usage.

specialises
governed call

MCP proxied call

pwm:MCPProxiedCallmedium contextC2

Tool invocation mediated by the MCP proxy, recording backend and invoked tool.

specialises
tool call
properties
mcpBackend

envelope attribute

pwm:EnvelopeAttributemedium contextC2

Recorded request or response attribute, with name, value and typed or classifier-derived source. Typed values are authoritative; classification fills gaps and may only narrow permission. Authentication supplies identity.

specialises
EntityPROV-O

classifier output

pwm:ClassifierOutputmedium contextC2

Calibrated classifier evidence with model identifier, version, label, confidence and input hash. Replay uses the recorded label without invoking the classifier.

specialises
EntityPROV-O

policy check

pwm:PolicyCheckmedium contextC2

Individual rule evaluation, pinning the evaluated policy row content hash; a policy decision may contain several checks.

specialises
EntityPROV-O

retrieved item

pwm:RetrievedItemmedium contextC2

Retrieved source snapshot with content hash and embedding model identity.

specialises
EntityPROV-O
properties
embeddingModel

retention entry

pwm:RetentionEntrymedium contextC2

Content-addressed retention store entry, separate from chain evidence. Erasure removes content but preserves its hash and erasure evidence.

specialises
EntityPROV-O

retention policy

pwm:RetentionPolicyslow contextC1

Per-use-case policy for how long source snapshots, prompts and responses are retained, subject to lawful erasure.

specialises
EntityPROV-O

erasure event

pwm:ErasureEventmedium contextC2

Erasure of retained content, recording its hash on the surviving evidence chain.

specialises
ActivityPROV-O
properties
erasedEntry

reconstruction report

pwm:ReconstructionReportmedium contextC2

Report for an interaction verifying chain and pinned hashes, listing findings and explicitly covering all ten supervisory questions, including not recorded answers.

specialises
EntityPROV-O

question coverage

pwm:QuestionCoveragemedium contextC2

Answer status for exactly one supervisory reconstruction question; absent evidence is explicitly not recorded.

specialises
EntityPROV-O
properties
answerText

authenticated client configuration

pwm:AuthenticatedClientConfigurationslow contextC2

Authenticated gateway client configuration binding permitted agent names and listed system triggers.

specialises
EntityPROV-O

system trigger

pwm:SystemTriggerslow contextC2

Listed authenticated system identity admitted for calls without an agent; still subject to purpose, use-case, policy and budget checks.

specialises
AgentPROV-O

Properties

information owner

pwm:informationOwnerobject property

Role accountable for a governed information asset or definition.

domain → range
anyparty roleOMG Commons

approval status

pwm:approvalStatusobject property

Approval status of a governed information asset: draft, current, superseded or withdrawn.

domain → code list
anyinformation asset statuses

approved for use case

pwm:approvedForUseCaseobject property

Use case for which a source is approved.

domain → range
approved sourceAI use case
  • KOGKognita roadmap0.6 item 3: each use case has a register of approved sources

claim type

pwm:claimTypeobject property

Epistemic type of a claim or memory record.

domain → code list
anyclaim types
  • KOGKognita vocabulariesclaims: VERIFIED_FACT, INFERENCE, CONVERSATION_HYPOTHESIS, POSSIBLE_BANK_CAPABILITY

validator status

pwm:validatorStatusobject property

Verdict of an independent verifier on a claim. Policy decides what each status may reach.

domain → code list
claimvalidator statuses

cites source

pwm:citesSourceobject property

Approved source, governed asset or fact a claim relies on.

domain → range
anyEntityPROV-O
sub-property of
wasDerivedFromPROV-O

uses definition

pwm:usesDefinitionobject property

Governed definition, at a version, that a claim or figure uses.

domain → range
anygoverned definition

under use case

pwm:underUseCaseobject property

Registered use case a decision, claim or event was produced under.

domain → range
anyAI use case
  • KOGKognita roadmap0.4 item 8: every decision must reference a registered use case

relies on delegation

pwm:reliesOnDelegationobject property

Delegated authority a decision or action relied on.

domain → range
anydelegated authority
  • KOGKognita roadmap0.5 item 2: every decision references the delegation it relied on

has budget

pwm:hasBudgetobject property

Budget snapshot applied to a run.

domain → range
runrun budget

in run

pwm:inRunobject property

Run grouping a call, checkpoint, decision or resumption.

domain → range
anyrun

pending approval

pwm:pendingApprovalobject property

Unresolved human approval currently holding the run.

domain → range
runapproval request

invoked tool

pwm:invokedToolobject property

Tool invoked through the gateway.

domain → range
tool calltool

has attribute

pwm:hasAttributeobject property

Recorded attribute used in the request or response envelope.

domain → range
anyenvelope attribute

retrieved item

pwm:retrievedItemobject property

Pinned item used by a governed call.

domain → range
anyretrieved item

retained in

pwm:retainedInobject property

Store entry holding content keyed by its hash.

domain → range
anyretention entry

call agent

pwm:callAgentobject property

Registered agent identity admitted under the authenticated configuration.

domain → range
governed callAI agent

run state

pwm:runStateobject property

Recorded runState code.

domain → code list
anyrun states

approval resolution

pwm:approvalResolutionobject property

Recorded approvalResolution code.

domain → code list
anyapproval resolutions

attribute source

pwm:attributeSourceobject property

Recorded attributeSource code.

domain → code list
anyattribute sources

version identifier

pwm:versionIddatatype property

Version identifier of an asset, definition, rule, model or deployment.

domain → range
anystringXSD

retention until

pwm:retentionUntildatatype property

Time until which a record must be kept. An AI record never expires before the regulatory record it supports.

domain → range
anydateTimeXSD

as of

pwm:asOfdatatype property

Time at which a fact, valuation or live authority status was true in its system of record.

domain → range
anydateTimeXSD

freshness threshold in seconds

pwm:freshnessThresholdSecondsdatatype property

Maximum age at which a fact may still be used, in whole seconds.

domain → range
factpositiveIntegerXSD

previous record hash

pwm:previousRecordHashdatatype property

Hash of the record that precedes this one in the chain. Required on every record except the first in a chain.

domain → range
evidence recordstringXSD

model version

pwm:modelVersiondatatype property

Provider model version as served for a model call, claim or evaluation.

domain → range
anystringXSD

wall clock seconds

pwm:wallClockSecondsdatatype property

Maximum elapsed wall clock seconds.

domain → range
run budgetdecimalXSD

consumed cost usd

pwm:consumedCostUSDdatatype property

US dollars consumed at this snapshot.

domain → range
run budgetdecimalXSD

elapsed seconds

pwm:elapsedSecondsdatatype property

Elapsed wall clock seconds at this snapshot.

domain → range
run budgetdecimalXSD

model name

pwm:modelNamedatatype property

Model name as reported for this call.

domain → range
model callstringXSD

prompt template version

pwm:promptTemplateVersiondatatype property

Version of the prompt template used.

domain → range
model callstringXSD

prompt hash

pwm:promptHashdatatype property

Hash of prompt as sent after redaction.

domain → range
model callstringXSD

response hash

pwm:responseHashdatatype property

Hash of model or tool response as received.

domain → range
anystringXSD

attribute name

pwm:attributeNamedatatype property

Envelope field name; identity fields must come from authentication.

domain → range
envelope attributestringXSD

escalation threshold

pwm:escalationThresholddatatype property

Policy confidence threshold below which the classifier cannot yield ALLOW.

domain → range
anydecimalXSD

policy content hash

pwm:policyContentHashdatatype property

Hash of the exact policy row evaluated by this check.

domain → range
policy checkstringXSD

embedding model

pwm:embeddingModeldatatype property

Embedding model identifier and version used for retrieval.

domain → range
retrieved itemstringXSD

store locator

pwm:storeLocatordatatype property

Reference to content in the separate retention store, never embedded chain content.

domain → range
retention entryanyURIXSD

answer text

pwm:answerTextdatatype property

Evidence-based answer, or reason the question was not recorded.

domain → range
question coveragestringXSD

pinned hashes verified

pwm:pinnedHashesVerifieddatatype property

Whether all available pinned hashes matched; mismatches require findings.

domain → range
reconstruction reportbooleanXSD

Code lists

data classifications

pwm:DataClassificationScheme

Ordered by sensitivity. Compare by rank, never by label: entitlement filtering and egress control depend on the order.

Used by classification, classificationCeiling, permitsDataClass, usesDataClass

CodeConceptDefinition
C0publicpwm:C0Approved for public release.
C1internalpwm:C1For staff; no client identity.
C2confidentialpwm:C2Identifies a client; within banking secrecy where it applies.close match: Personal DataDPV
C3restrictedpwm:C3Need-to-know only, such as screening hits, investigations, special-category data and material non-public information.close match: Special Category Personal DataDPV

policy decision outcomes

pwm:DecisionOutcomeScheme

Resolution is fail-closed: the most severe outcome present decides, in the order deny, escalate, human approval, allow.

Used by decisionOutcome, expectedOutcome

CodeConceptDefinition
DENYdenypwm:OutcomeDenyRefused; nothing moves.
ESCALATEescalatepwm:OutcomeEscalateUncertain; referred to a person.
HUMAN_APPROVALhuman approvalpwm:OutcomeHumanApprovalHeld until a named person approves this exact action.
ALLOWallowpwm:OutcomeAllowEvery rule passed.
OBSERVEobservepwm:OutcomeObserveRecorded without gating, in monitor-only deployments.

claim types

pwm:ClaimTypeScheme

Used by claimType

CodeConceptDefinition
VERIFIED_FACTverified factpwm:VerifiedFactBacked by a cited source and verification.
INFERENCEinferencepwm:InferenceReasoned from facts; not itself verified.
CONVERSATION_HYPOTHESISconversation hypothesispwm:ConversationHypothesisSomething worth raising with the client, to be tested in conversation.
POSSIBLE_BANK_CAPABILITYpossible bank capabilitypwm:PossibleBankCapabilitySomething the institution might be able to offer, not yet confirmed.

validator statuses

pwm:ValidatorStatusScheme

Used by validatorStatus, verificationVerdict

CodeConceptDefinition
VERIFIEDverifiedpwm:ValidatorVerifiedVerifier confirmed it.
INFERREDinferredpwm:ValidatorInferredSupported only by inference.
UNVERIFIEDunverifiedpwm:ValidatorUnverifiedNot yet checked.
CONFLICTINGconflictingpwm:ValidatorConflictingSources disagree.
STALEstalepwm:ValidatorStaleRelies on data past its freshness threshold.
BLOCKEDblockedpwm:ValidatorBlockedMay not be released.

information asset statuses

pwm:AssetStatusScheme

Used by approvalStatus

CodeConceptDefinition
draftdraftpwm:AssetDraftNot approved; never given to AI as authoritative.
currentcurrentpwm:AssetCurrentApproved and in force.
supersededsupersededpwm:AssetSupersededReplaced by a later version; kept for reconstruction only.
withdrawnwithdrawnpwm:AssetWithdrawnWithdrawn without replacement.

stale-data actions

pwm:StaleActionScheme

Used by staleAction

CodeConceptDefinition
REFRESHrefreshpwm:StaleRefreshFetch a current value before use.
SUPPRESSsuppresspwm:StaleSuppressDo not use the value.
WARNwarnpwm:StaleWarnUse with a visible staleness warning.

memory scopes

pwm:MemoryScopeScheme

Used by memoryScope

CodeConceptDefinition
WORKINGworkingpwm:MemoryWorkingTemporary context; expires with the run.
RELATIONSHIPrelationshippwm:MemoryRelationshipDurable client information, governed like any bank record.
INSTITUTIONALinstitutionalpwm:MemoryInstitutionalWhat the organization learns across clients; no client data.

evidence event types

pwm:EvidenceEventTypeScheme

Notations match Kognita EventType values. Model refusal is added from the Filing Cabinet talk so refusals and policy denials are logged apart.

Used by eventType

CodeConceptDefinition
POLICY_DECISIONpolicy decisionpwm:EventPolicyDecisionA decision was taken.
MODEL_CALLmodel callpwm:EventModelCallA model was called; provider, version and prompt and response hashes recorded.
TOOL_CALLtool callpwm:EventToolCallA tool was invoked; arguments recorded as hashes.
RETRIEVALretrievalpwm:EventRetrievalContent was retrieved, with item hashes.
APPROVALapprovalpwm:EventApprovalA person signed an approval.
EGRESSegresspwm:EventEgressData left for a provider, allowed, redacted or denied.
POLICY_CHANGEpolicy changepwm:EventPolicyChangeA rule or register entry changed.
ORIGINATIONoriginationpwm:EventOriginationA client or product was selected.
FINAL_DECISIONfinal decisionpwm:EventFinalDecisionA person decided.
CLIENT_COMMUNICATIONclient communicationpwm:EventClientCommunicationSomething was sent to a client.
INTERVENTIONinterventionpwm:EventInterventionA person paused, overrode, restricted, revoked or recovered a live run.
OUTCOMEoutcomepwm:EventOutcomeA complaint, correction or other outcome was linked to the interaction.
NEAR_MISSnear misspwm:EventNearMissA near miss was recorded.
CLAIMclaimpwm:EventClaimA claim was recorded.
INCIDENT_OPENEDincident openedpwm:EventIncidentOpenedAn AI incident was opened.
ERASUREerasurepwm:EventErasureContent was erased; the chain stays verifiable.
MODEL_REFUSALmodel refusalpwm:EventModelRefusalThe model declined, as distinct from a policy denial.
ANCHOR_PUBLISHEDanchor publishedpwm:EventAnchorPublishedA chain-head checkpoint was published to an external sink.
ASSURANCEassurancepwm:EventAssuranceA runtime assurance record was evidenced: manifest, attestation, update approval, component approval, drill or production outcome.
CLAIM_VERIFIEDclaim verifiedpwm:EventClaimVerifiedA claim passed the verification gate.
EVALUATION_RECORDEDevaluation recordedpwm:EventEvaluationRecordedAn evaluation result was recorded.
EXTERNAL_REVIEW_SUBMITTEDexternal review submittedpwm:EventExternalReviewSubmittedAn external reviewer submitted a signed review.
INCIDENT_UPDATEDincident updatedpwm:EventIncidentUpdatedAn incident was updated.
INCIDENT_CLOSEDincident closedpwm:EventIncidentClosedAn incident was closed.
INTERACTIONinteractionpwm:EventInteractionA client interaction was recorded.
KEY_REVOKEDkey revokedpwm:EventKeyRevokedA signing key was revoked.
KEY_ROTATEDkey rotatedpwm:EventKeyRotatedA signing key was rotated.
LEARNING_EXPORTlearning exportpwm:EventLearningExportAn approved export of interaction data for learning or evaluation was made.
MEMORY_WRITEmemory writepwm:EventMemoryWriteA memory record was written.
REVALIDATION_TRIGGEREDrevalidation triggeredpwm:EventRevalidationTriggeredA model change or other trigger required revalidation.
REVIEWreviewpwm:EventReviewA human review was recorded.
SOURCE_SUPERSEDEDsource supersededpwm:EventSourceSupersededAn approved source was superseded, so facts that depend on it are no longer current.

run states

pwm:RunStateScheme

Codes for runState.

Used by runState

CodeConceptDefinition
readyreadypwm:RunReadyRun created and ready to start governed calls.
runningrunningpwm:RunRunningRun is executing governed calls within its budgets.
suspendedsuspendedpwm:RunSuspendedRun is held at a durable checkpoint awaiting human approval.
resumedresumedpwm:RunResumedRun continuation was rehydrated after a recorded human resolution.
completedcompletedpwm:RunCompletedRun finished its authorized work.
denieddeniedpwm:RunDeniedRun stopped because a governing decision denied continuation.
failedfailedpwm:RunFailedRun terminated with a recorded execution failure.

approval resolutions

pwm:ApprovalResolutionScheme

Codes for approvalResolution.

Used by approvalResolution

CodeConceptDefinition
approvedapprovedpwm:ApprovalApprovedA named human approved the exact held action.
denieddeniedpwm:ApprovalDeniedA named human denied the held action; continuation records the denial and does not authorize that action.

attribute sources

pwm:AttributeSourceScheme

Codes for attributeSource.

Used by attributeSource

CodeConceptDefinition
typedtypedpwm:AttributeTypedCaller-supplied typed attribute is authoritative; a classifier cannot replace it.
classifier_derivedclassifier-derivedpwm:AttributeClassifierDerivedClassifier fills a gap in typed attributes and can only narrow permission.

question coverage statuses

pwm:QuestionCoverageStatusScheme

Codes for coverageStatus.

Used by coverageStatus

CodeConceptDefinition
answeredansweredpwm:QuestionAnsweredAnswered from recorded evidence; citesSource identifies the evidence.
not_recordednot recordedpwm:QuestionNotRecordedEvidence needed to answer was not recorded at this milestone; the question remains explicit.

gateway failure modes

pwm:GatewayFailureModeScheme

Codes for gatewayFailureMode.

Used by gatewayFailureMode

CodeConceptDefinition
FAIL_CLOSEDfail closedpwm:FailClosedDefault: refuse calls while the gateway or evidence store is unavailable.
DEGRADEDdegradedpwm:DegradedOnly local-model calls without client data may proceed during an outage; evidence is written when the store recovers.

supervisory reconstruction questions

pwm:ReconstructionQuestionScheme

Codes for reconstructionQuestion.

Used by reconstructionQuestion

CodeConceptDefinition
QuestionWhyClientWhy this client?pwm:QuestionWhyClientSupervisory reconstruction question: Why this client?
QuestionWhyProductWhy this insight or product?pwm:QuestionWhyProductSupervisory reconstruction question: Why this insight or product?
QuestionInformationWhat information did the AI use?pwm:QuestionInformationSupervisory reconstruction question: What information did the AI use?
QuestionModelAgentWhat model or agent produced it?pwm:QuestionModelAgentSupervisory reconstruction question: What model or agent produced it?
QuestionAuthorityWhat was the agent authorized to do?pwm:QuestionAuthoritySupervisory reconstruction question: What was the agent authorized to do?
QuestionControlsWhat suitability or policy controls ran?pwm:QuestionControlsSupervisory reconstruction question: What suitability or policy controls ran?
QuestionRMChangesWhat did the RM see and change?pwm:QuestionRMChangesSupervisory reconstruction question: What did the RM see and change?
QuestionFinalDecisionWho made the final decision?pwm:QuestionFinalDecisionSupervisory reconstruction question: Who made the final decision?
QuestionCommunicationWhat was communicated to the client?pwm:QuestionCommunicationSupervisory reconstruction question: What was communicated to the client?
QuestionReproductionCan the bank reproduce that evidence later?pwm:QuestionReproductionSupervisory reconstruction question: Can the bank reproduce that evidence later?

Module 10

Ingestion, policy and lifecycle controls

Structured ingestion and the missing policy, review and monitoring records for milestone 0.4.

Classes

ingested document

pwm:IngestedDocumentmedium contextC1

Immutable source-file ingestion record identifying the parser and its exact version. Reuses contentHash and versionId.

specialises
EntityPROV-O

document element

pwm:DocumentElementmedium contextC1

Parser-derived passage or structural element of an ingested document, with page, section path, character span, classification and content hash. Parent links preserve the parsed tree.

specialises
document section

declarative policy

pwm:DeclarativePolicyslow contextC1

Versioned declarative artifact serializing existing PolicyRule records; not a replacement rule engine.

specialises
EntityPROV-O

policy change

pwm:PolicyChangeslow contextC1

Archived proposed, reviewed, strictly validated and applied delta against a policy version, containing added, modified or removed rule identifiers.

specialises
EntityPROV-O

policy test case

pwm:PolicyTestCaseslow contextC1

Named scenario under a policy rule, with given attributes, requested action and expected decision outcome.

specialises
EntityPROV-O

policy pack

pwm:PolicyPackslow contextC1

Named versioned bundle of policy artifacts, scenarios and fixture references; a private overlay may shadow a built-in pack.

specialises
EntityPROV-O

flight recorder session

pwm:FlightRecorderSessionmedium contextC1

Evidence export manifest for recorded runs and interactions, with a content hash and retained export locator. Reuses Run and ReconstructionReport rather than duplicating their evidence.

specialises
EntityPROV-O
properties
exportLocator

replay

pwm:Replaymedium contextC1

Verification of a recorder session against retained evidence, producing an existing ReconstructionReport; does not mean re-executing client actions.

specialises
ActivityPROV-O

RM review edit

pwm:RMReviewEditmedium contextC1

Structured edit captured during HumanReview, preserving hashes of what the RM saw, what was approved and the retained diff.

specialises
EntityPROV-O
properties
diffLocator

review queue

pwm:ReviewQueuemedium contextC1

Use-case review queue receiving high-risk and sampled low-risk interactions for human review.

specialises
EntityPROV-O

review queue entry

pwm:ReviewQueueEntrymedium contextC1

Risk-selected interaction awaiting or completing human review, with reason, queue time and delivery boundary.

specialises
EntityPROV-O

review coverage report

pwm:ReviewCoverageReportmedium contextC1

Evidence-derived counts of eligible, screened, queued and reviewed interactions over a window, with review latency; grouped by use case and its risk tier.

specialises
EntityPROV-O

circuit breaker

pwm:CircuitBreakerslow contextC1

Automatic halt configuration specifying flag threshold, time window and scope; trips create a prohibiting policy and incident, and reset requires accountable-owner approval.

specialises
kill switch

circuit breaker trip

pwm:CircuitBreakerTripmedium contextC1

Recorded threshold crossing with exact scope, threshold, observed flags, window and inserted prohibiting rule.

specialises
AI incident

affected client query

pwm:AffectedClientQuerymedium contextC1

Version-scoped and time-bounded query for clients receiving output, with an explicit result even when no clients match.

specialises
EntityPROV-O

affected client result

pwm:AffectedClientResultmedium contextC1

Affected-client query result with matching communication and interaction links and result count; zero results is distinct from no result.

specialises
EntityPROV-O

interaction outcome

pwm:InteractionOutcomemedium contextC1

Classified outcome linked to the originating interaction, exact output, model and approver; includes complaints, corrections and operational failures.

specialises
EntityPROV-O

Properties

shadows pack

pwm:shadowsPackobject property

Built-in pack shadowed by a private overlay with the same name.

domain → range
anypolicy pack

recorded run

pwm:recordedRunobject property

Run included in a recorder export manifest.

domain → range
anyrun

has outcome metric

pwm:hasOutcomeMetricobject property

Versioned primary metric definition with baseline and target for a use case.

domain → range
anyoutcome metric

has model card

pwm:hasModelCardobject property

Approved model card for a model or pinned model call.

domain → range
anymodel card

credited control

pwm:creditedControlobject property

Control credited in reducing inherent to residual materiality.

domain → range
anypolicy rule

communication signer

pwm:communicationSignerobject property

Named human signing the exact communication content hash.

domain → range
anypersonFIBO · Release

scope use case

pwm:scopeUseCaseobject property

Use-case dimension of a breaker or affected-client query scope.

domain → range
anyAI use case

scope model

pwm:scopeModelobject property

Pinned model entity identifying the exact modelVersion in scope.

domain → range
anyAI model

prohibiting rule

pwm:prohibitingRuleobject property

Inserted rule denying subsequent actions in the tripped scope.

domain → range
anypolicy rule

affected output

pwm:affectedOutputobject property

Exact retained output implicated in an outcome or near miss.

domain → range
anyrecommendation

outcome approver

pwm:outcomeApproverobject property

Human approver associated with the implicated output.

domain → range
anypersonFIBO · Release

implicated model

pwm:implicatedModelobject property

Pinned model implicated in an outcome or near miss, without asserting that the outcome is itself a model call.

domain → range
anyAI model

evaluated against appetite

pwm:evaluatedAgainstAppetiteobject property

Versioned institutional appetite against which registration and residual materiality were evaluated.

domain → range
anyAI risk appetite

parser name

pwm:parserNamedatatype property

Parser implementation used for ingestion.

domain → range
anystringXSD

section path

pwm:sectionPathdatatype property

Ordered source heading path; use root for a document without headings.

domain → range
anystringXSD

bounding box

pwm:boundingBoxdatatype property

Parser bounding box with coordinate system and units.

domain → range
anystringXSD

policy serialization

pwm:policySerializationdatatype property

Retained declarative rule text, normally YAML.

domain → range
anystringXSD

delta text

pwm:deltaTextdatatype property

Added, modified and removed rule delta; identifiers must resolve against the base policy.

domain → range
anystringXSD

change stage

pwm:changeStagedatatype property

Policy delta stage: propose, review, validated or applied.

domain → range
anystringXSD

given attributes

pwm:givenAttributesdatatype property

Serialized scenario input attributes.

domain → range
anystringXSD

fixture locator

pwm:fixtureLocatordatatype property

Location of a pack fixture bundle.

domain → range
anyanyURIXSD

appetite permits

pwm:appetitePermitsdatatype property

Recorded risk-appetite evaluation permits registration and activation, including residual materiality.

domain → range
anybooleanXSD

risk appetite statement

pwm:riskAppetiteStatementdatatype property

Versioned institutional AI appetite, including allowed categories, impact, authority and quantitative measures.

domain → range
anystringXSD

before hash

pwm:beforeHashdatatype property

Hash of the exact recommendation rendered before RM edits.

domain → range
anystringXSD

edit diff hash

pwm:editDiffHashdatatype property

Content hash of the retained structured edit diff.

domain → range
anystringXSD

requires two signatures

pwm:requiresTwoSignaturesdatatype property

Whether this communication requires two distinct human signatures.

domain → range
anybooleanXSD

low risk sampling rate

pwm:lowRiskSamplingRatedatatype property

Low-risk sampling fraction between zero and one, set per use case.

domain → range
anydecimalXSD

queue reason

pwm:queueReasondatatype property

Risk criterion or low-risk sampling reason for selection.

domain → range
anystringXSD

review before delivery

pwm:reviewBeforeDeliverydatatype property

Whether review must finish before client delivery.

domain → range
anybooleanXSD

mean review seconds

pwm:meanReviewSecondsdatatype property

Mean queue-to-review latency in seconds for completed reviews.

domain → range
anydecimalXSD

scope prompt version

pwm:scopePromptVersiondatatype property

Prompt version in scope.

domain → range
anystringXSD

window start

pwm:windowStartdatatype property

Inclusive start of a monitoring or query window.

domain → range
anydateTimeXSD

window end

pwm:windowEnddatatype property

Exclusive end of a monitoring or query window.

domain → range
anydateTimeXSD

Code lists

interaction outcome kind scheme

pwm:InteractionOutcomeKindScheme

Classifications recorded under roadmap item 15.

Used by outcomeKind

CodeConceptDefinition
client_complaintclient complaintpwm:ClientComplaintA client expresses dissatisfaction with the interaction or its output.
correctioncorrectionpwm:CorrectionAn error in an earlier output is rectified.
wrong_client_targetingwrong client targetingpwm:WrongClientTargetingAn output or action is directed to an unintended client.
unsuitable_suggestionunsuitable suggestionpwm:UnsuitableSuggestionA suggestion conflicts with the client suitability requirements.
communication_errorcommunication errorpwm:CommunicationErrorA delivered communication contains incorrect or misleading content.
policy_breachpolicy breachpwm:PolicyBreachAn action or output violates an applicable policy rule.
operational_incidentoperational incidentpwm:OperationalIncidentAn operational failure disrupts the interaction or its controls.

near miss kind scheme

pwm:NearMissKindScheme

Classifications recorded under roadmap item 15.

Used by nearMissKind

CodeConceptDefinition
blocked_actionblocked actionpwm:BlockedActionA control prevents a potentially harmful action from executing.
verifier_blockverifier blockpwm:VerifierBlockA verifier rejects an output before it proceeds.
rm_rejectionRM rejectionpwm:RMRejectionA relationship manager rejects an AI draft or proposed action.
unsafe_draftunsafe draftpwm:UnsafeDraftA potentially harmful draft is identified before delivery.
unusual_agent_behaviorunusual agent behaviorpwm:UnusualAgentBehaviorUnexpected agent behavior is detected before harm occurs.

Module 11

Agents, authority and fleets

Workload identities, scoped human authority, autonomy evidence, containment and fleet controls.

Classes

workload identity

pwm:WorkloadIdentityslow contextC2

Registry of an individual agent workload credential reference, issuer, subject and environment. Stores no private key or token. Gateways verify possession on every call.

specialises
EntityPROV-O

authority scope

pwm:AuthorityScopeslow contextC2

Explicit finite sets of clients, purposes, verbs, tools, fields, systems, products and jurisdictions. Omission grants nothing for dimensions checked by the written narrowing rules. Numeric caps and autonomy may only narrow.

specialises
EntityPROV-O

human entitlement

pwm:HumanEntitlementslow contextC2

Time-bounded entitlement held by a named person, with explicit subject scope. An agent delegation cannot enlarge this scope.

specialises
authorizationOMG Commons

authority revocation

pwm:AuthorityRevocationmedium contextC2

Record that authority or a credential was revoked at a stated time, including task completion. Revocation is effective on the next decision.

specialises
EntityPROV-O

autonomy transition

pwm:AutonomyTransitionmedium contextC2

Evidenced promotion or automatic one-level step-down for an agent, with before and after levels, evaluation window, metric sample, approval and notification records. Restoration requires promotion again.

specialises
EntityPROV-O

blast radius observation

pwm:BlastRadiusObservationmedium contextC2

Recorded projected cumulative consumption against one agent limit, including the requested action. Window and dimension are inherited from that exact limit.

specialises
EntityPROV-O

external content

pwm:ExternalContentmedium contextC2

Content from outside the institution, including agent messages, PDFs and API responses. Its trust label never conveys instruction authority.

specialises
EntityPROV-O

authenticated instruction

pwm:AuthenticatedInstructionmedium contextC2

Instruction with authenticated human, assurance level, source and transaction context. A signature reference is evidence to verify, never a secret or a substitute for step-up authorization.

specialises
EntityPROV-O

capability grant

pwm:CapabilityGrantslow contextC2

Explicit expiring and revocable capability assigned to an agent within a subject scope. Issue, use, expiry and revocation are evidenced.

specialises
authorizationOMG Commons

fleet halt

pwm:FleetHaltmedium contextC2

Recorded fleet-wide stop instruction with explicit fleet scope and time. Runtime dispatch and cancellation are performed by the gateway.

specialises
EntityPROV-O
properties
haltScope

sandbox constraints

pwm:SandboxConstraintsmedium contextC2

Decision output specifying the permitted credential references, network destinations, filesystem scope, APIs, commands, export, persistence, auto-stop interval and resource caps. Kognita emits constraints, not a sandbox.

specialises
EntityPROV-O

compensating action

pwm:CompensatingActionslow contextC2

Registered recovery action for a reversible tool; recovery records reference the original call and an explicit reverse execution order.

specialises
EntityPROV-O

adversarial test case

pwm:AdversarialTestCaseslow contextC2

Attack scenario specializing the existing policy test case, linked to a versioned threat model and the control tested. Passing requires denial, escalation or containment.

specialises
policy test case

threat model

pwm:ThreatModelslow contextC2

Versioned threat model covering own-agent failures and hostile external agents, mapping attack scenarios to controls.

specialises
EntityPROV-O

enforcement attestation

pwm:EnforcementAttestationmedium contextC2

Attributed time-stamped evidence about gateway mediation, credential custody, egress routing and evidence-store isolation for a use case. The attestation is a claim to verify operationally.

specialises
EntityPROV-O

action stop

pwm:ActionStopmedium contextC2

Recorded stop or cancellation request for an in-flight tool call with outcome canceled, completed, timed out or unsupported; unsupported cancellation requires compensation or decision-separated steps.

specialises
EntityPROV-O

approving authority role

pwm:ApprovingAuthorityRoleslow contextC2

Firm role empowered to set and approve blast-radius limits and autonomy promotion gates.

specialises
party roleOMG Commons
  • REGMAS AIRG1.11 agent risks; 5.9 autonomy oversight
  • KOGKognita roadmaproadmap milestone 0.5 items 2, 3, 4 and 16; Ahmed design decisions 3 Nov 2026
  • NEWPWM practitioner definition0.5: NEW specialized record or property; no exact standard term identified

risk committee

pwm:RiskCommitteeslow contextC2

Firm risk committee acting as approving authority.

  • REGMAS AIRG1.11 agent risks; 5.9 autonomy oversight
  • KOGKognita roadmaproadmap milestone 0.5 items 2, 3, 4 and 16; Ahmed design decisions 3 Nov 2026
  • NEWPWM practitioner definition0.5: NEW specialized record or property; no exact standard term identified

data governance committee

pwm:DataGovernanceCommitteeslow contextC2

Firm data governance committee acting as approving authority.

  • REGMAS AIRG1.11 agent risks; 5.9 autonomy oversight
  • KOGKognita roadmaproadmap milestone 0.5 items 2, 3, 4 and 16; Ahmed design decisions 3 Nov 2026
  • NEWPWM practitioner definition0.5: NEW specialized record or property; no exact standard term identified

chief risk officer

pwm:ChiefRiskOfficerslow contextC2

Firm chief risk officer acting as approving authority.

  • REGMAS AIRG1.11 agent risks; 5.9 autonomy oversight
  • KOGKognita roadmaproadmap milestone 0.5 items 2, 3, 4 and 16; Ahmed design decisions 3 Nov 2026
  • NEWPWM practitioner definition0.5: NEW specialized record or property; no exact standard term identified

risk control approval

pwm:RiskControlApprovalslow contextC2

Recorded approval of a firm-specific blast-radius limit or autonomy promotion gate, identifying the approving role and time.

specialises
EntityPROV-O
  • REGMAS AIRG1.11 agent risks; 5.9 autonomy oversight
  • KOGKognita roadmaproadmap milestone 0.5 items 2, 3, 4 and 16; Ahmed design decisions 3 Nov 2026
  • NEWPWM practitioner definition0.5: NEW specialized record or property; no exact standard term identified

Properties

agent owner

pwm:agentOwnerobject property

Named human accountable for this agent deployment.

domain → range
AI agentpersonFIBO · Release

operates for

pwm:operatesForobject property

Human principal on whose behalf an agent operates.

domain → range
AI agentpersonFIBO · Release

authority scope

pwm:authorityScopeobject property

Explicit scope of an entitlement, delegation or capability grant.

domain → range
anyauthority scope

input content

pwm:inputContentobject property

Content consumed by a call or run, including agent-to-agent messages.

domain → range
anyEntityPROV-O

deployment model

pwm:deploymentModelobject property

Model approved as part of this agent deployment, distinct from the provider-reported model of a call.

domain → range
AI agentAI model

scope limit

pwm:scopeLimitobject property

Quantitative cap attached to authority scope; child scopes must retain every parent cap with the same dimension, basis and unit and an equal or lower maximum.

risk control approval

pwm:riskControlApprovalobject property

Approval setting and authorizing this firm-specific limit or promotion gate.

domain → range
anyrisk control approval
  • REGMAS AIRG1.11 agent risks; 5.9 autonomy oversight
  • KOGKognita roadmaproadmap milestone 0.5 items 2, 3, 4 and 16; Ahmed design decisions 3 Nov 2026
  • NEWPWM practitioner definition0.5: NEW specialized record or property; no exact standard term identified

deployment environment

pwm:deploymentEnvironmentdatatype property

Environment in which a workload or agent is deployed.

domain → range
anystringXSD

approved deployment version

pwm:approvedDeploymentVersiondatatype property

Exact deployment hash approved by the owner.

domain → range
AI agentstringXSD

scope field

pwm:scopeFielddatatype property

Exact field name permitted for reading or writing through the scoped tools, systems and products, for the scoped clients, purposes, action verbs and jurisdictions. Each requestedField must be explicitly listed as scopeField; permission for one field grants no access to other fields or the whole record.

domain → range
authority scopestringXSD

limit dimension

pwm:limitDimensiondatatype property

Measured dimension: records, clients, value, actions, transactions or clients-contacted.

domain → range
blast radius limitstringXSD

permitted field

pwm:permittedFielddatatype property

Field name this agent may receive for its purpose.

domain → range
AI agentstringXSD

access event

pwm:accessEventdatatype property

Event boundary constraining a delegation, for example a meeting within 24 hours.

domain → range
authority scopestringXSD

public content tool

pwm:publicContentTooldatatype property

Whether a tool publishes or prepares public content that must never receive client data.

domain → range
toolbooleanXSD

requires step up

pwm:requiresStepUpdatatype property

Whether policy requires step-up and independent authorization for this decision.

domain → range
policy decisionbooleanXSD

containment applied

pwm:containmentApplieddatatype property

Whether quarantine, credential revocation and tool stop were applied automatically before human release.

domain → range
behavior anomalybooleanXSD

threat model reference

pwm:threatModelReferencedatatype property

Resolvable reference to the published threat model.

domain → range
threat modelanyURIXSD

maximum duration seconds

pwm:maximumDurationSecondsdatatype property

Maximum duration of each tool call in seconds.

domain → range
tooldecimalXSD

supports cancellation

pwm:supportsCancellationdatatype property

Whether the proxy can cancel this tool in flight.

domain → range
toolbooleanXSD

decision separated steps

pwm:decisionSeparatedStepsdatatype property

Whether an uncancelable action is split into steps with a decision between them.

domain → range
toolbooleanXSD

stop outcome

pwm:stopOutcomedatatype property

Cancellation result: canceled, completed, timed-out or unsupported.

domain → range
action stopstringXSD

call duration seconds

pwm:callDurationSecondsdatatype property

Observed tool-call duration in seconds; exceeding the registered maximum must be recorded as a timeout.

domain → range
tool calldecimalXSD

approved at

pwm:approvedAtdatatype property

Time the approving authority approved the risk control.

domain → range
risk control approvaldateTimeXSD
  • REGMAS AIRG1.11 agent risks; 5.9 autonomy oversight
  • KOGKognita roadmaproadmap milestone 0.5 items 2, 3, 4 and 16; Ahmed design decisions 3 Nov 2026
  • NEWPWM practitioner definition0.5: NEW specialized record or property; no exact standard term identified

Code lists

Content trust and taint labels

pwm:ContentTrustScheme

Content trust and taint labels.

Used by trustLabel

CodeConceptDefinition
internal_contentinternal contentpwm:TrustInternalApproved internal content, subject to ordinary entitlements.
external_contentexternal contentpwm:TrustExternalPotentially hostile external content; never authority to act.
client_dataclient datapwm:TrustClientDataClient data taint; cannot reach public-content tools.

Tool reversibility classes

pwm:ReversibilityScheme

Tool reversibility classes.

Used by reversibilityClass

CodeConceptDefinition
reversiblereversiblepwm:ReversibleEffects can be undone by a registered compensating action.
compensatablecompensatablepwm:CompensatableEffects can be mitigated by a registered compensation, without claiming exact reversal.
irreversibleirreversiblepwm:IrreversibleNo registered reversal; stronger pre-execution controls required.

Module 12

Trust, resilience and assurance

Signed and anchored evidence, provider and dependency registers, failover and drills, retention by record class, incident reports and runtime assurance records.

Classes

key rotation

pwm:KeyRotationmedium contextC1

Record that one signing key replaced another. Verification works across rotations.

specialises
EntityPROV-O

key revocation

pwm:KeyRevocationmedium contextC1

Record that a signing key is revoked from a stated evidence sequence. Signatures at or after that sequence fail closed.

specialises
EntityPROV-O

external review submission

pwm:ExternalReviewSubmissionmedium contextC2

Signed review by an external reviewer of an immutable range of evidence sequences, with annotations. The reviewer signature is checked against the key registry.

specialises
EntityPROV-O

model provider registration

pwm:ModelProviderRegistrationslow contextC1

A model provider recorded as an ICT third party: use cases that depend on it, data residency and retention, training use, sub-processors, encryption, incident notification, availability, model-change notice, audit rights, exit plan, incident history and regulatory cooperation.

specialises
EntityPROV-O

model change detection

pwm:ModelChangeDetectionmedium contextC1

Comparison of the model version reported by the provider with the version approved at admission. A difference the provider did not announce triggers revalidation.

specialises
EntityPROV-O

AI dependency

pwm:AIDependencyslow contextC1

One edge of the AI dependency map: a use case depends on a provider, cloud, framework, vector platform, orchestration layer, tool or data source. Declared in the registers, observed in evidence, or both.

specialises
EntityPROV-O

failure drill

pwm:FailureDrillmedium contextC1

Offline simulation of one of eight fault kinds against a use case, checking the configured failure mode and that evidence stays complete.

specialises
EntityPROV-O
  • REGMAS AIRG5.15: guardrails tested for agent failure modes; 5.22: technology and cybersecurity review, red teaming
  • KOGKognita roadmaproadmap milestone 0.7 item R10 (failure drills)
  • NEWPWM practitioner definition0.8: NEW specialized record or property; no exact standard term identified

incident report

pwm:IncidentReportmedium contextC2

Report built from evidence for an incident: affected clients, controls that ran, failed or were missing, and the runtime change behind it. A redacted report exposes fixed labels and aggregate counts only.

specialises
EntityPROV-O

assurance policy

pwm:AssurancePolicyslow contextC1

Accountable-owner requirements for a use case: which external assertions, attestations, component approvals, transfer classification and update approvals are required, and the outcome thresholds that trigger revalidation.

specialises
EntityPROV-O

external agent assertion

pwm:ExternalAgentAssertionmedium contextC2

Protocol-neutral identity assertion from an external agent: issuer, subject, principal, purpose, audience, exact scopes and a bounded validity window. Not transitive by default.

specialises
EntityPROV-O

deployment attestation

pwm:DeploymentAttestationmedium contextC2

Independent signed boundary evidence bound to one deployment and policy: privacy, environment, monitoring, instruction or transfer. Expires.

specialises
EntityPROV-O

third-party update approval

pwm:ThirdPartyUpdateApprovalmedium contextC1

Approval of a candidate third-party AI update with regression evidence, rollback plan and a named approver. A high-materiality update cannot become approved without them.

specialises
EntityPROV-O

technology transfer classification

pwm:TechnologyTransferClassificationslow contextC1

Classification of an AI technology transfer: type, origin of the technology, deployment jurisdiction and the policy reference that governs it.

specialises
EntityPROV-O

production outcome

pwm:ProductionOutcomemedium contextC2

Authoritative downstream facts for one interaction, bound to a unique outcome identifier and source: intended and actual result, client impact, adverse flag. Adverse outcomes beyond the configured threshold trigger revalidation and autonomy step-down.

verifier

pwm:Verifierslow contextC1

Registered agent or service, with its own identity, that checks another agent's claims. Its model, model version and provider are pinned so independence from the producer can be checked.

specialises
EntityPROV-O

claim verification

pwm:ClaimVerificationmedium contextC1

One verification of a claim by a registered verifier, with the producer's and the verifier's model pinned. The verifier must differ from the producer in model, model version and provider unless the use case permits otherwise.

specialises
EntityPROV-O

evaluation result

pwm:EvaluationResultmedium contextC1

Recorded evaluation of an agent version on a model version at the model, agent or system level, with metric values and the deployment content hash that identifies the evaluated behavior.

specialises
EntityPROV-O

evaluation metric value

pwm:EvaluationMetricValuemedium contextC1

One named metric in an evaluation result, such as accuracy, groundedness, hallucination rate or escalation correctness.

specialises
EntityPROV-O

fairness report

pwm:FairnessReportmedium contextC2

Recommendation, contact and prioritization rates across firm-defined segments, with the disparity threshold, named proxy attributes and a flag to the risk owner.

specialises
EntityPROV-O
  • REGMAS AIRG5.7 and 5.8: fair outcomes; fairness assessments on protected attributes and proxies
  • KOGKognita roadmaproadmap milestone 0.6 item 11 (evaluation evidence gate)
  • NEWPWM practitioner definition0.8: NEW specialized record or property; no exact standard term identified

evaluation gate

pwm:EvaluationGateslow contextC1

Threshold per use case and metric that an evaluation result must meet before a tier 2 or higher agent is activated or changed: a minimum, or for rates such as hallucination a maximum.

specialises
EntityPROV-O

reasoning trace

pwm:ReasoningTracemedium contextC2

Record that a provider returned a reasoning trace for a model call: its hash and where the content is retained. The model's own account, never a basis for a decision.

specialises
EntityPROV-O

learning export

pwm:LearningExportmedium contextC2

Governed export of interactions, RM edits or outcomes to a training, fine-tuning or evaluation pipeline: purpose, anonymization, validation, approval and requester recorded. Never the default.

specialises
EntityPROV-O

sensitive attribute inference

pwm:SensitiveAttributeInferencemedium contextC3

Special-category inference (religion, health, political views, ethnicity, sexuality or personal vulnerability) labeled by the classifier. Blocked from relationship workflows unless the use case permits. A vulnerability flag goes to RM review; an AI may flag but never determine.

specialises
EntityPROV-O

Properties

signed by

pwm:signedByobject property

Signing key whose signature covers this record. Verification checks the key against the key registry.

domain → range
anysigning key

approved fallback model

pwm:approvedFallbackModelobject property

Model card approved as a fallback for the use case.

domain → range
AI use casemodel card

requires deployment attestation

pwm:requiresDeploymentAttestationobject property

Attestation kind the policy requires for the deployment.

domain → code list
assurance policyattestation kinds

verifier model card

pwm:verifierModelCardobject property

Model card of the model that the verifier runs on. Its model version and provider are the verifier's pinned identity.

domain → range
verifiermodel card

has documentation

pwm:hasDocumentationobject property

Documentation held for an AI use case, such as a model card or design record (MAS AIRG 4.7).

domain → range
AI use caseEntityPROV-O

has component

pwm:hasComponentobject property

Approved software component that an agent deployment is built from (MAS AIRG fn 25).

domain → range
AI agentapproved component

activated on evaluation

pwm:activatedOnEvaluationobject property

Evaluation result that the agent's activation or last change relied on.

domain → range
AI agentevaluation result

memory subject client

pwm:memorySubjectClientobject property

Client that a relationship memory record is about. Institutional memory names no client.

domain → range
memory recordclient

schema version

pwm:schemaVersiondatatype property

Version of the evidence event schema the record was written under. Older events stay readable and verify.

domain → range
evidence recordstringXSD

public key reference

pwm:publicKeyReferencedatatype property

Location or fingerprint of the public key.

domain → range
signing keystringXSD

escrow reference

pwm:escrowReferencedatatype property

Reference to the escrowed recovery material for a key, never the material itself.

domain → range
signing keystringXSD

anchored record hash

pwm:anchoredRecordHashdatatype property

Hash of the chain head that the anchor publishes.

domain → range
evidence anchorstringXSD

anchor location

pwm:anchorLocationdatatype property

Where the checkpoint is kept, in storage the firm does not control.

domain → range
evidence anchorstringXSD

restore verified

pwm:restoreVerifieddatatype property

Whether a restore verified the chain and retention store end to end.

domain → range
recovery bundlebooleanXSD

includes retention state

pwm:includesRetentionStatedatatype property

Whether the bundle holds the historical retention state needed to restore to the cutoff.

domain → range
recovery bundlebooleanXSD

triggers revalidation

pwm:triggersRevalidationdatatype property

Whether the detection requires revalidation of the use case.

domain → range
model change detectionbooleanXSD

dependency target

pwm:dependencyTargetdatatype property

Name of the provider, platform, tool or data source depended on.

domain → range
AI dependencystringXSD

dependency declared

pwm:dependencyDeclareddatatype property

Whether the registers declare the dependency.

domain → range
AI dependencybooleanXSD

dependency observed

pwm:dependencyObserveddatatype property

Whether evidence shows the dependency was actually used.

domain → range
AI dependencybooleanXSD

critical use case share

pwm:criticalUseCaseSharedatatype property

Fraction from 0 to 1 of critical use cases resting on the target.

domain → range
concentration measuredecimalXSD

evidence complete after drill

pwm:evidenceCompleteAfterDrilldatatype property

Whether evidence stayed complete and the chain verified after the simulated fault.

domain → range
failure drillbooleanXSD

control ran

pwm:controlRandatatype property

Control that ran for the incident's interactions.

domain → range
incident reportstringXSD

control failed

pwm:controlFaileddatatype property

Control that failed or was missing for the incident's interactions.

domain → range
incident reportstringXSD

runtime change

pwm:runtimeChangedatatype property

Deployment or configuration change behind the incident.

domain → range
incident reportstringXSD

redacted report

pwm:redactedReportdatatype property

Whether the report exposes fixed labels and aggregate counts only.

domain → range
incident reportbooleanXSD

assurance audience

pwm:assuranceAudiencedatatype property

Audience that external assertions must name.

domain → range
assurance policystringXSD

requires external assertion

pwm:requiresExternalAssertiondatatype property

Whether external-agent assertions are required.

domain → range
assurance policybooleanXSD

requires component approval

pwm:requiresComponentApprovaldatatype property

Whether approved components are required.

domain → range
assurance policybooleanXSD

requires transfer classification

pwm:requiresTransferClassificationdatatype property

Whether a technology transfer classification is required.

domain → range
assurance policybooleanXSD

requires update approval

pwm:requiresUpdateApprovaldatatype property

Whether third-party updates need approval.

domain → range
assurance policybooleanXSD

outcome step-down

pwm:outcomeStepDowndatatype property

Whether adverse outcomes beyond the threshold step autonomy down.

domain → range
assurance policybooleanXSD

manifest digest

pwm:manifestDigestdatatype property

Digest of the manifest document that the signature covers.

domain → range
agent manifeststringXSD

issued at

pwm:issuedAtdatatype property

When a credential or assertion was issued.

domain → range
anydateTimeXSD

expires at

pwm:expiresAtdatatype property

When a credential or assertion stops being valid.

domain → range
anydateTimeXSD

attestation reference

pwm:attestationReferencedatatype property

Reference to the independent evidence behind the attestation.

domain → range
deployment attestationstringXSD

SBOM digest

pwm:sbomDigestdatatype property

SHA-256 of the software bill of materials, 64 lowercase hexadecimal characters.

domain → range
approved componentstringXSD

image digest

pwm:imageDigestdatatype property

Container image digest, sha256: and 64 lowercase hexadecimal characters.

domain → range
approved componentstringXSD

permits self verification

pwm:permitsSelfVerificationdatatype property

Whether the use case explicitly permits a verifier that shares the producer's model, version or provider. Only at tier 1.

domain → range
AI use casebooleanXSD

evaluation deployment hash

pwm:evaluationDeploymentHashdatatype property

SHA-256 of the deployment content that was evaluated, 64 lowercase hexadecimal characters.

domain → range
evaluation resultstringXSD

fairness threshold

pwm:fairnessThresholddatatype property

Disparity above which the report flags a segment to the risk owner.

domain → range
fairness reportdecimalXSD

proxy attribute

pwm:proxyAttributedatatype property

Attribute that stands in for, or correlates with, a protected attribute.

domain → range
fairness reportstringXSD

fairness flagged

pwm:fairnessFlaggeddatatype property

Whether a disparity above the threshold was flagged to the risk owner.

domain → range
fairness reportbooleanXSD

gate metric name

pwm:gateMetricNamedatatype property

Name of the metric that the gate checks.

domain → range
evaluation gatestringXSD

gate minimum score

pwm:gateMinimumScoredatatype property

Lowest acceptable score, from 0 to 1.

domain → range
evaluation gatedecimalXSD

gate maximum score

pwm:gateMaximumScoredatatype property

Highest acceptable score, from 0 to 1, for rates where lower is better.

domain → range
evaluation gatedecimalXSD

trace hash

pwm:traceHashdatatype property

Hash of the reasoning trace in the evidence chain; the content sits in the retention store.

domain → range
reasoning tracestringXSD

requires reasoning trace

pwm:requiresReasoningTracedatatype property

Whether the use case requires a reasoning trace on material output.

domain → range
AI use casebooleanXSD

claim source system

pwm:claimSourceSystemdatatype property

System of record that the claim's source belongs to.

domain → range
claimstringXSD

claim source record

pwm:claimSourceRecorddatatype property

Record in the source system that the claim relies on.

domain → range
claimstringXSD

claim source timestamp

pwm:claimSourceTimestampdatatype property

As-of time of the source record.

domain → range
claimdateTimeXSD

claim retrieval timestamp

pwm:claimRetrievalTimestampdatatype property

When the source record was retrieved.

domain → range
claimdateTimeXSD

calculation method

pwm:calculationMethoddatatype property

Registered calculation service or method that produced a figure in the claim.

domain → range
claimstringXSD

claim prompt version

pwm:claimPromptVersiondatatype property

Prompt version that produced the claim.

domain → range
claimstringXSD

claim confidence

pwm:claimConfidencedatatype property

Confidence from 0 to 1 that the producer assigned to the claim.

domain → range
claimdecimalXSD

claim policy status

pwm:claimPolicyStatusdatatype property

Whether policy allows the claim to reach an RM or client.

domain → range
claimstringXSD

memory correction path

pwm:memoryCorrectionPathdatatype property

How a person can correct or remove the memory record.

domain → range
memory recordstringXSD

anonymization reference

pwm:anonymizationReferencedatatype property

Reference to the anonymization applied before export.

domain → range
learning exportstringXSD

validation reference

pwm:validationReferencedatatype property

Reference to the validation report that cleared the export.

domain → range
learning exportstringXSD

exported record identifier

pwm:exportedRecordIddatatype property

Identifier of a record included in the export.

domain → range
learning exportstringXSD

exclusion reason

pwm:exclusionReasondatatype property

Why the client or record is excluded from learning exports.

domain → range
learning exclusionstringXSD

excludes record identifier

pwm:excludesRecordIddatatype property

Record excluded from every future learning export.

domain → range
learning exclusionstringXSD

changes client treatment

pwm:changesClientTreatmentdatatype property

Whether the inference changes how the client is treated. Requires a person's determination.

Code lists

actor types

pwm:ActorTypeScheme

Used by actorType

CodeConceptDefinition
HUMANhumanpwm:ActorHumanA person.
AGENTagentpwm:ActorAgentAn AI agent.
SYSTEMsystempwm:ActorSystemKognita or another system component.

approval request statuses

pwm:ApprovalStatusScheme

Used by approvalRequestStatus

CodeConceptDefinition
PENDINGpendingpwm:ApprovalPendingAwaiting a decision.
MARKEDmarkedpwm:ApprovalMarkedFirst signature recorded in a two-signature approval.
APPROVEDapprovedpwm:ApprovalStatusApprovedApproved.
REJECTEDrejectedpwm:ApprovalRejectedRejected.
EXPIREDexpiredpwm:ApprovalExpiredExpired without a decision.

egress decisions

pwm:EgressDecisionScheme

Used by egressDecision

CodeConceptDefinition
ALLOWallowpwm:EgressAllowData left unchanged.
REDACTredactpwm:EgressRedactData left after redaction.
DENYdenypwm:EgressDenyData did not leave.

check results

pwm:CheckResultScheme

Used by checkResult

CodeConceptDefinition
passpasspwm:CheckPassThe check passed.
failfailpwm:CheckFailThe check failed.
escalateescalatepwm:CheckEscalateThe check escalates to a person.
requires_humanrequires humanpwm:CheckRequiresHumanA human decision is required.

failover triggers

pwm:FailoverTriggerScheme

Used by failoverTrigger

CodeConceptDefinition
connection_failureconnection failurepwm:TriggerConnectionFailureThe provider connection failed.
http_5xxHTTP 5xxpwm:TriggerHttp5xxThe provider returned a server error.
model_changedmodel changedpwm:TriggerModelChangedThe served model differs from the approved model.

dependency layers

pwm:DependencyLayerScheme

Used by concentrationLayer, dependencyLayer

CodeConceptDefinition
model_providermodel providerpwm:LayerModelProviderModel provider.
cloudcloudpwm:LayerCloudCloud platform.
frameworkframeworkpwm:LayerFrameworkAgent framework.
vector_platformvector platformpwm:LayerVectorPlatformVector store or retrieval platform.
orchestrationorchestrationpwm:LayerOrchestrationOrchestration layer.
tooltoolpwm:LayerToolTool or API.
data_sourcedata sourcepwm:LayerDataSourceData source.

dependency divergences

pwm:DependencyDivergenceScheme

Used by divergenceKind

CodeConceptDefinition
R8-UNDECLAREDundeclaredpwm:DivergenceUndeclaredObserved in evidence but not declared in the registers.
R8-UNUSEDunusedpwm:DivergenceUnusedDeclared in the registers but never observed.

drill kinds

pwm:DrillKindScheme

Used by drillKind

CodeConceptDefinition
model_unavailablemodel unavailablepwm:DrillModelUnavailableThe model is unavailable.
vendor_outagevendor outagepwm:DrillVendorOutageThe vendor is down.
api_outageAPI outagepwm:DrillApiOutageAn API is down.
retrieval_failureretrieval failurepwm:DrillRetrievalFailureRetrieval fails.
identity_service_failureidentity service failurepwm:DrillIdentityServiceFailureThe identity service fails.
corrupted_contextcorrupted contextpwm:DrillCorruptedContextContext is corrupted.
compromised_modelcompromised modelpwm:DrillCompromisedModelThe model is compromised.
tool_failuretool failurepwm:DrillToolFailureA tool fails.

regulatory record classes

pwm:RecordClassScheme

Used by recordClass

CodeConceptDefinition
client communicationsclient communicationspwm:RecordClientCommunicationsClient communications.
adviceadvicepwm:RecordAdviceAdvice.
recommendationsrecommendationspwm:RecordRecommendationsRecommendations.
approvalsapprovalspwm:RecordApprovalsApprovals.
transactionstransactionspwm:RecordTransactionsTransactions.
compliance recordscompliance recordspwm:RecordComplianceRecordsCompliance records.

operational incident kinds

pwm:IncidentKindScheme

Used by incidentKind

CodeConceptDefinition
chain_breakchain breakpwm:IncidentChainBreakThe evidence chain failed verification.
gateway_outagegateway outagepwm:IncidentGatewayOutageThe gateway was unavailable.
degraded_modedegraded modepwm:IncidentDegradedModeThe gateway ran in degraded mode.
provider_failureprovider failurepwm:IncidentProviderFailureA model provider failed.

attestation kinds

pwm:AttestationKindScheme

Used by attestationKind, requiresDeploymentAttestation

CodeConceptDefinition
privacyprivacypwm:AttestationPrivacyPrivacy impact context.
environmentenvironmentpwm:AttestationEnvironmentExecution environment.
monitoringmonitoringpwm:AttestationMonitoringMonitoring coverage.
instructioninstructionpwm:AttestationInstructionHigh-risk instruction authenticity.
transfertransferpwm:AttestationTransferJurisdictional technology transfer.

technology transfer types

pwm:TransferTypeScheme

Used by transferType

CodeConceptDefinition
apiAPI/SaaSpwm:TransferApiAccess through an API or hosted service.
weightsmodel weightspwm:TransferWeightsTransfer of model weights.
licenselicensepwm:TransferLicenseLicense transfer. The Kognita value keeps its original spelling, license.
locallocal deploymentpwm:TransferLocalLocal deployment.
ownershipownership or controlpwm:TransferOwnershipTransfer of ownership or control.

supervisory conformance profiles

pwm:ConformanceProfileScheme

Used by conformanceProfile

CodeConceptDefinition
MASMASpwm:ProfileMASMonetary Authority of Singapore.
HKMA-ARTHKMA ARTpwm:ProfileHKMAARTHong Kong Monetary Authority.
FSBFSBpwm:ProfileFSBFinancial Stability Board sound practices.
OCCOCCpwm:ProfileOCCUS Office of the Comptroller of the Currency examiner considerations.
SECSECpwm:ProfileSECSelected US Securities and Exchange Commission themes.
FCAFCApwm:ProfileFCASelected UK Financial Conduct Authority themes.

special categories

pwm:SensitiveKindScheme

Used by sensitiveKind

CodeConceptDefinition
religionreligionpwm:SensitiveReligionReligion.
healthhealthpwm:SensitiveHealthHealth.
political_viewspolitical viewspwm:SensitivePoliticalPolitical views.
ethnicityethnicitypwm:SensitiveEthnicityEthnicity.
sexualitysexualitypwm:SensitiveSexualitySexuality.
vulnerabilitypersonal vulnerabilitypwm:SensitiveVulnerabilityPersonal vulnerability.

hostile-agent drill scenarios

pwm:DrillScenarioScheme

Used by drillScenario

CodeConceptDefinition
prompt-injection-tool-escalationprompt injection leading to tool escalationpwm:ScenarioPromptInjectionInjected instructions try to escalate tool use.
discovered-credentialsdiscovered third-party credentialspwm:ScenarioDiscoveredCredentialsThe agent finds credentials it was not given.
unauthorized-networkunauthorized external network accesspwm:ScenarioUnauthorizedNetworkThe agent reaches for a network destination outside its allow-list.
machine-speed-probingmachine-speed probingpwm:ScenarioMachineSpeedProbingHigh-rate API probing and multi-technique attack.
compromised-providercompromised providerpwm:ScenarioCompromisedProviderThe model provider is compromised.
agent-privilege-escalationagent-to-agent privilege escalationpwm:ScenarioAgentPrivilegeEscalationOne agent tries to gain another's privileges.
malicious-retrievalmalicious retrieved contentpwm:ScenarioMaliciousRetrievalRetrieved content carries hostile instructions.
containment-recoveryemergency containment and recoverypwm:ScenarioContainmentRecoveryContainment and recovery after a breach.

evaluation levels

pwm:EvaluationLevelScheme

Used by evaluationLevel

CodeConceptDefinition
modelmodelpwm:LevelModelThe model alone.
agentagentpwm:LevelAgentThe agent: planning, tool selection, permission compliance, delegation, retries and escalation.
systemsystempwm:LevelSystemThe end-to-end system: model, retrieval, tools, data, agent, policies, interface and human interaction.

Module 13

Business scope

Costs and charges, product governance, client circumstances, orders and executions, corporate actions and income, tax reporting, lending and performance provenance.

Classes

cost disclosure

pwm:CostDisclosuremedium contextC2

Disclosure to a client of the costs and charges of an investment service or product, ex ante before the service and ex post after the period, with the total and its items.

specialises
EntityPROV-O
  • REGMiFID IIMiFID II Art. 24(4)(c) costs and charges information; Delegated Regulation 2017/565 Art. 50
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

cost item

pwm:CostItemmedium contextC2

One item in a cost disclosure: category, amount and currency, optionally the fee it comes from.

specialises
EntityPROV-O
  • REGMiFID IIMiFID II Art. 24(4)(c) costs and charges information; Delegated Regulation 2017/565 Art. 50
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

inducement disclosure

pwm:InducementDisclosuremedium contextC2

Disclosure of a third-party payment, with whether it is designed to enhance the quality of the service to the client.

specialises
EntityPROV-O
  • REGMiFID IIMiFID II Art. 24(9) inducements; Delegated Regulation 2017/565 Art. 11
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

performance figure

pwm:PerformanceFiguremedium contextC2

Performance of a portfolio over a period, as a figure from a registered calculation service. PWM records its provenance and never computes it.

specialises
EntityPROV-O

Properties

disclosure timing

pwm:disclosureTimingobject property

Whether the disclosure is ex ante or ex post.

domain → code list
cost disclosuredisclosure timings
  • REGMiFID IIMiFID II Art. 24(4)(c) costs and charges information; Delegated Regulation 2017/565 Art. 50
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

disclosed to client

pwm:disclosedToClientobject property

Client who received the disclosure.

domain → range
anyclient
  • REGMiFID IIMiFID II Art. 24(4)(c) costs and charges information; Delegated Regulation 2017/565 Art. 50
  • REGMiFID IIMiFID II Art. 24(9) inducements; Delegated Regulation 2017/565 Art. 11
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

covers mandate

pwm:coversMandateobject property

Mandate whose service the disclosure covers.

domain → range
cost disclosuremandate
  • REGMiFID IIMiFID II Art. 24(4)(c) costs and charges information; Delegated Regulation 2017/565 Art. 50
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

has cost item

pwm:hasCostItemobject property

Item that makes up the disclosed cost.

domain → range
cost disclosurecost item
  • REGMiFID IIMiFID II Art. 24(4)(c) costs and charges information; Delegated Regulation 2017/565 Art. 50
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

cost category

pwm:costCategoryobject property

Category of a cost item.

domain → code list
cost itemcost categories
  • REGMiFID IIMiFID II Art. 24(4)(c) costs and charges information; Delegated Regulation 2017/565 Art. 50
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

cost from fee

pwm:costFromFeeobject property

Fee that the cost item comes from.

domain → range
cost itemfee
  • REGMiFID IIMiFID II Art. 24(4)(c) costs and charges information; Delegated Regulation 2017/565 Art. 50
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

approves product

pwm:approvesProductobject property

Product approved for distribution.

domain → range
product approvalproduct
  • REGMiFID IIMiFID II Art. 16(3) and Art. 24(2) product governance, target market
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

approved for target market

pwm:approvedForTargetMarketobject property

Target market the product is approved for.

  • REGMiFID IIMiFID II Art. 16(3) and Art. 24(2) product governance, target market
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

negative target market category

pwm:negativeTargetMarketCategoryobject property

Investor category the product is not compatible with.

domain → code list
target marketinvestor categories
  • REGMiFID IIMiFID II Art. 16(3) and Art. 24(2) product governance, target market
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

vulnerability determined by

pwm:vulnerabilityDeterminedByobject property

Person who determined the vulnerability. An AI may flag but not determine.

domain → range
vulnerability indicatorpersonFIBO · Release
  • REGFCA FG21/1FCA FG21/1 four drivers of vulnerability: health, life events, resilience, capability
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

order side

pwm:orderSideobject property

Buy or sell.

domain → code list
orderorder sides
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

order type

pwm:orderTypeobject property

Market or limit.

domain → code list
orderorder types
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

order status

pwm:orderStatusobject property

State of the order.

domain → code list
orderorder statuses
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

order account

pwm:orderAccountobject property

Account the order is placed in.

domain → range
orderaccountFIBO · Release
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

order under mandate

pwm:orderUnderMandateobject property

Mandate the order is placed under.

domain → range
ordermandate
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

executes order

pwm:executesOrderobject property

Order that the execution fills.

domain → range
executionorder
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

allocates execution

pwm:allocatesExecutionobject property

Execution that the allocation assigns.

domain → range
allocationexecution
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

allocated to account

pwm:allocatedToAccountobject property

Account that receives the quantity.

domain → range
allocationaccountFIBO · Release
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

disclosed at

pwm:disclosedAtdatatype property

When the disclosure was made.

domain → range
anydateTimeXSD
  • REGMiFID IIMiFID II Art. 24(4)(c) costs and charges information; Delegated Regulation 2017/565 Art. 50
  • REGMiFID IIMiFID II Art. 24(9) inducements; Delegated Regulation 2017/565 Art. 11
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

total cost amount

pwm:totalCostAmountdatatype property

Total cost over the disclosure period.

domain → range
cost disclosuredecimalXSD
  • REGMiFID IIMiFID II Art. 24(4)(c) costs and charges information; Delegated Regulation 2017/565 Art. 50
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

total cost percent

pwm:totalCostPercentdatatype property

Total cost as a percentage of the invested amount.

domain → range
cost disclosuredecimalXSD
  • REGMiFID IIMiFID II Art. 24(4)(c) costs and charges information; Delegated Regulation 2017/565 Art. 50
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

includes inducements

pwm:includesInducementsdatatype property

Whether the disclosed cost includes third-party payments.

domain → range
cost disclosurebooleanXSD
  • REGMiFID IIMiFID II Art. 24(4)(c) costs and charges information; Delegated Regulation 2017/565 Art. 50
  • REGMiFID IIMiFID II Art. 24(9) inducements; Delegated Regulation 2017/565 Art. 11
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

period start

pwm:periodStartdatatype property

Start of the period a disclosure or figure covers.

domain → range
anydateTimeXSD
  • REGMiFID IIMiFID II Art. 24(4)(c) costs and charges information; Delegated Regulation 2017/565 Art. 50
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified
  • NEWPWM practitioner definitionPWM scope: complaint handling, corporate actions, income, lending and performance provenance

period end

pwm:periodEnddatatype property

End of the period a disclosure or figure covers.

domain → range
anydateTimeXSD
  • REGMiFID IIMiFID II Art. 24(4)(c) costs and charges information; Delegated Regulation 2017/565 Art. 50
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified
  • NEWPWM practitioner definitionPWM scope: complaint handling, corporate actions, income, lending and performance provenance

enhances service quality

pwm:enhancesServiceQualitydatatype property

Whether the payment is designed to enhance the quality of the service to the client.

domain → range
inducement disclosurebooleanXSD
  • REGMiFID IIMiFID II Art. 24(9) inducements; Delegated Regulation 2017/565 Art. 11
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

distribution strategy reference

pwm:distributionStrategyReferencedatatype property

Reference to the distribution strategy for the product.

domain → range
product approvalstringXSD
  • REGMiFID IIMiFID II Art. 16(3) and Art. 24(2) product governance, target market
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

next review due at

pwm:nextReviewDueAtdatatype property

When the next review is due.

domain → range
anydateTimeXSD
  • REGMiFID IIMiFID II Art. 16(3) and Art. 24(2) product governance, target market
  • REGMiFID IIMiFID II Art. 25(2) suitability, periodic assessment under Delegated Regulation 2017/565 Art. 54(12)
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

complaint received at

pwm:complaintReceivedAtdatatype property

When the complaint was received.

domain → range
complaintdateTimeXSD

vulnerability flagged at

pwm:vulnerabilityFlaggedAtdatatype property

When the indicator was raised.

domain → range
vulnerability indicatordateTimeXSD
  • REGFCA FG21/1FCA FG21/1 four drivers of vulnerability: health, life events, resilience, capability
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

review frequency in months

pwm:reviewFrequencyMonthsdatatype property

Agreed frequency of periodic suitability reviews, in months.

domain → range
suitability reviewpositiveIntegerXSD
  • REGMiFID IIMiFID II Art. 25(2) suitability, periodic assessment under Delegated Regulation 2017/565 Art. 54(12)
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

performed at

pwm:performedAtdatatype property

When the review was performed.

domain → range
suitability reviewdateTimeXSD
  • REGMiFID IIMiFID II Art. 25(2) suitability, periodic assessment under Delegated Regulation 2017/565 Art. 54(12)
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

order quantity

pwm:orderQuantitydatatype property

Quantity ordered.

domain → range
orderdecimalXSD
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

limit price

pwm:limitPricedatatype property

Limit price of a limit order.

domain → range
orderdecimalXSD
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

order placed at

pwm:orderPlacedAtdatatype property

When the order was placed.

domain → range
orderdateTimeXSD
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

executed quantity

pwm:executedQuantitydatatype property

Quantity filled.

domain → range
executiondecimalXSD
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

execution price

pwm:executionPricedatatype property

Price of the fill.

domain → range
executiondecimalXSD
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

executed at

pwm:executedAtdatatype property

When the fill happened.

domain → range
executiondateTimeXSD
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

execution venue

pwm:executionVenuedatatype property

Venue where the fill happened.

domain → range
executionstringXSD
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

allocated quantity

pwm:allocatedQuantitydatatype property

Quantity assigned to the account.

domain → range
allocationdecimalXSD
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

venue rationale

pwm:venueRationaledatatype property

Why the venue was chosen.

domain → range
best execution recordstringXSD
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

execution policy reference

pwm:executionPolicyReferencedatatype property

Version of the best execution policy applied.

domain → range
best execution recordstringXSD
  • REGMiFID IIMiFID II Art. 27 best execution; Delegated Regulation 2017/565 Art. 64
  • NEWPWM practitioner definition0.10: NEW specialized record or property; no exact standard term identified

reported to jurisdiction

pwm:reportedToJurisdictiondatatype property

Jurisdiction reported to, as an ISO 3166-1 alpha-2 code.

domain → range
tax reporting recordstringXSD

Code lists

disclosure timings

pwm:DisclosureTimingScheme

Used by disclosureTiming

CodeConceptDefinition
ex_anteex antepwm:TimingExAnteBefore the service is provided.
ex_postex postpwm:TimingExPostAfter the period, on the actual costs.

cost categories

pwm:CostCategoryScheme

Used by costCategory

CodeConceptDefinition
one_offone-offpwm:CostOneOffOne-off charges.
ongoingongoingpwm:CostOngoingOngoing charges.
transactiontransactionpwm:CostTransactionTransaction costs.
incidentalincidentalpwm:CostIncidentalIncidental costs, such as performance fees.
ancillaryancillarypwm:CostAncillaryAncillary services.

target market review outcomes

pwm:TargetMarketReviewOutcomeScheme

Used by targetMarketReviewOutcome

CodeConceptDefinition
confirmedconfirmedpwm:ReviewConfirmedThe target market stands.
amendedamendedpwm:ReviewAmendedThe target market was changed.
withdrawnwithdrawnpwm:ReviewWithdrawnThe product was withdrawn from distribution.

complaint resolutions

pwm:ComplaintResolutionScheme

Used by complaintResolution

CodeConceptDefinition
upheldupheldpwm:ResolutionUpheldUpheld.
partially_upheldpartially upheldpwm:ResolutionPartlyUpheldPartially upheld.
not_upheldnot upheldpwm:ResolutionNotUpheldNot upheld.
withdrawnwithdrawn by clientpwm:ResolutionWithdrawnWithdrawn by the client.

vulnerability drivers

pwm:VulnerabilityDriverScheme

Used by vulnerabilityDriver

CodeConceptDefinition
healthhealthpwm:DriverHealthHealth conditions that affect the ability to make decisions.
life_eventslife eventspwm:DriverLifeEventLife events such as bereavement.
resilienceresiliencepwm:DriverResilienceLow ability to withstand financial or emotional shocks.
capabilitycapabilitypwm:DriverCapabilityLow knowledge of financial matters or low digital skills.

suitability review results

pwm:SuitabilityReviewResultScheme

Used by suitabilityReviewResult

CodeConceptDefinition
still_suitablestill suitablepwm:SuitabilityStillThe recommendation remains suitable.
no_longer_suitableno longer suitablepwm:SuitabilityNoThe recommendation is no longer suitable.
update_requiredupdate requiredpwm:SuitabilityUpdateThe assessment needs updating.

order types

pwm:OrderTypeScheme

Used by orderType

CodeConceptDefinition
marketmarketpwm:TypeMarketMarket order.
limitlimitpwm:TypeLimitLimit order.

order statuses

pwm:OrderStatusScheme

Used by orderStatus

CodeConceptDefinition
placedplacedpwm:OrderPlacedPlaced.
partially_filledpartially filledpwm:OrderPartlyFilledPartially filled.
filledfilledpwm:OrderFilledFilled.
canceledcanceledpwm:OrderCanceledCanceled.
rejectedrejectedpwm:OrderRejectedRejected.

best execution factors

pwm:ExecutionFactorScheme

Used by executionFactor

CodeConceptDefinition
pricepricepwm:FactorPricePrice.
costscostspwm:FactorCostsCosts.
speedspeedpwm:FactorSpeedSpeed.
likelihoodlikelihood of execution and settlementpwm:FactorLikelihoodLikelihood of execution and settlement.
sizesizepwm:FactorSizeSize.
naturenaturepwm:FactorNatureNature of the order.

corporate action kinds

pwm:CorporateActionKindScheme

Used by corporateActionKind

CodeConceptDefinition
dividenddividendpwm:ActionDividendDividend.
couponcouponpwm:ActionCouponCoupon payment.
splitsplitpwm:ActionSplitSplit.
mergermergerpwm:ActionMergerMerger.
rights_issuerights issuepwm:ActionRightsIssueRights issue.
redemptionredemptionpwm:ActionRedemptionRedemption.

income kinds

pwm:IncomeKindScheme

Used by incomeKind

CodeConceptDefinition
dividenddividendpwm:IncomeDividendDividend income.
interestinterestpwm:IncomeInterestInterest income.
couponcouponpwm:IncomeCouponCoupon income.

tax reporting regimes

pwm:TaxReportingRegimeScheme

Used by reportingRegime

CodeConceptDefinition
crsCRSpwm:RegimeCRSOECD Common Reporting Standard.
fatcaFATCApwm:RegimeFATCAUS Foreign Account Tax Compliance Act.

lending facility kinds

pwm:FacilityKindScheme

Used by facilityKind

CodeConceptDefinition
lombardLombardpwm:FacilityLombardLombard loan against a securities portfolio.
mortgagemortgagepwm:FacilityMortgageMortgage.
structured_loanstructured loanpwm:FacilityStructuredStructured loan.
otherotherpwm:FacilityOtherOther credit.

performance measures

pwm:PerformanceMeasureScheme

Used by performanceMeasure

CodeConceptDefinition
time_weighted_returntime-weighted returnpwm:MeasureTWRTime-weighted return.
money_weighted_returnmoney-weighted returnpwm:MeasureMWRMoney-weighted return.
absolute_returnabsolute returnpwm:MeasureAbsoluteAbsolute return.

SHACL

Data rules

The shapes file checks data against the rules below. Validate together with the ontology and RDFS inference. A firm can switch a rule off in its own shapes graph and add its own thresholds there.

pyshacl -s pwm-shapes.ttl -e pwm.ttl -i rdfs your-data.ttl

Accountability role

applies to accountability role
  • An accountability role must be held by a named person (Kognita 0.4 item 8; MAS AIRG 4.7 key roles).
  • An AI agent can never be named in an accountability role (Kognita 0.4 item 8).

AI use case

applies to AI use case
  • A use case records its purpose (MAS AIRG 4.7).
  • A use case records its approved jurisdiction as an ISO 3166-1 alpha-2 code (MAS AIRG 4.7 approved scope of use).
  • A use case has one lifecycle status: pilot, live, suspended or retired (MAS AIRG 4.7; Kognita 0.4 item 8).
  • A use case has one risk tier, 1 to 4 (Kognita 0.4 item 8).
  • A use case states whether the control plane mediates it. Embedded vendor AI is registered too, marked unmediated (Kognita 0.4 item 8; MAS AIRG 4.2).
  • A use case carries a review date set by its tier; past it, the use case is refused (Kognita 0.4 item 8).
  • A use case with no business owner is refused (Kognita 0.4 item 8; MAS AIRG 4.7).
  • A use case at tier 2 or above with no named Knowledge Lead is refused (Kognita 0.4 item 8).

Live use case

applies to AI use case
  • A live use case needs a materiality assessment with inherent and residual ratings, approved by the designated control function (MAS AIRG 4.11, 4.13; Kognita 0.4 item 8).

Control function separation

applies to AI use case
  • The person who approves a use case's materiality rating for the control function cannot also be its business or technology owner (Kognita 0.4 item 8; MAS AIRG 4.13 independent oversight).

Risk materiality assessment

applies to risk materiality assessment
  • Rate impact, low to critical (MAS AIRG 4.12(a)).
  • Rate complexity, low to critical (MAS AIRG 4.12(b)).
  • Rate reliance, including the autonomy granted, low to critical (MAS AIRG 4.12(c)).
  • Record materiality before controls (MAS AIRG 4.11).
  • Record materiality after controls (MAS AIRG 4.11).

AI agent

applies to AI agent
  • An agent has one identifier (MAS AIRG footnote 25; Kognita 0.5 item 1).
  • An agent carries one maximum autonomy level, L0 to L6 (Kognita 0.5 item 3).
  • An agent is linked to at least one registered use case (Kognita 0.5 item 1).
  • An agent's deployment version covers prompt, tools, permissions and data sources, not only code (Kognita 0.5 item 1; MAS AIRG 5.25(b)).
  • An agent past its review date is refused, so it needs one (Kognita 0.5 item 1).

Agent tool level

applies to AI agent
  • An agent is permitted a tool that needs a higher autonomy level than the agent's maximum. A call above the agent's level is refused (Kognita 0.5 item 3).

Tool

applies to tool
  • Every tool declares the autonomy level it requires (Kognita 0.5 item 3).
  • A tool declares the verbs it performs, because action permissions are per verb (Kognita 0.5 item 3).

Delegated authority

applies to delegated authority
  • A delegation is granted by one named person (Kognita 0.5 item 2).
  • A delegation is granted to one registered agent (Kognita 0.5 item 2).
  • A delegation is bound to a purpose (Kognita 0.5 item 2).
  • A delegation names the actions it allows (Kognita 0.5 item 2).
  • A delegation must expire even when it also ends on task completion (Kognita 0.5 item 2).

Client communication

applies to client communication
  • A communication that does not trace back to a final decision is refused (Kognita 0.4 item 12).
  • Every communication is classified by communication level (Kognita 0.4 item 12).
  • Every communication carries its content provenance label (Kognita 0.4 item 11).
  • Client communication goes only through a registered, approved channel (Kognita 0.4 item 12).

AI disclosure

applies to client communication
  • When AI generated the content, the client communication must say so (Kognita private-banking pack, client-ai-disclosure; MAS AIRG 5.6).

Claim

applies to claim
  • Unlabeled claims cannot be shown to an RM or sent to a client (Kognita 0.6 item 5).
  • validatorStatus: maximum count 1.
  • A claim typed as verified fact must cite a source and carry the validator status verified. The model's own knowledge is never a source, and an inference cannot become a fact without verification (Kognita 0.6 items 3, 5 and 7).

Fact

applies to fact
  • A fact carries its as-of time (Kognita 0.6 item 2).
  • A fact carries a freshness threshold (Kognita 0.6 item 2).
  • A fact carries its classification (Kognita 0.6 item 2).
  • A fact names its source system (Kognita 0.6 item 2).

Valuation

applies to valuation
  • amount: minimum count 1; maximum count 1; datatype decimal.
  • currency: minimum count 1; maximum count 1.
  • asOf: minimum count 1; datatype dateTime.
  • A figure presented as authoritative must cite a registered calculation service (Kognita 0.6 item 6).

Suitability assessment

applies to suitability assessment
  • Name the recommendation being assessed.
  • Cite the one risk profile in force when the assessment was made (MiFID II Art. 25(2); Delegated Regulation (EU) 2017/565 Art. 54).
  • Check the recommendation against the recorded mandate (Kognita private-banking pack, mandate).
  • suitabilityOutcome: minimum count 1; maximum count 1; allowed values Suitable, SuitableWithConditions, NotSuitable, InsufficientInformation.
  • Record how the recommendation meets the client's objectives and characteristics, not a bare score (MiFID II Art. 25(6)).
  • Record who made the assessment.

Investor risk profile

applies to investor risk profile
  • riskProfileLevel: minimum count 1; maximum count 1.
  • validFrom: minimum count 1; datatype dateTime.
  • wasAttributedTo: minimum count 1.

Investor classification

applies to investor classification
  • investorCategory: minimum count 1; maximum count 1.
  • A classification is made under one jurisdiction's rules.
  • optStatus: maximum count 1.

Governed information asset

applies to governed information asset
  • Minimum label: owner (Filing Cabinet, slide 8).
  • Minimum label: approval status (Filing Cabinet, slide 8).
  • Minimum label: effective date (Filing Cabinet, slide 8).
  • Minimum label: retention date (Filing Cabinet, slide 8).
  • Minimum label: version identifier or content hash (Filing Cabinet, slide 8).

Document section

applies to document section
  • Classify per section, not per document (Filing Cabinet, slide 8).

Policy rule

applies to policy rule
  • Every rule carries a citation to what it implements (Kognita 0.4 item 3).
  • Rules are effective-dated (Kognita 0.3 item 7).
  • A change to a rule is a new version, never an edit in place (Kognita 0.3 item 7).

Cross-border rule

applies to cross-border rule
  • appliesToActivity: minimum count 1.
  • permitted: minimum count 1; maximum count 1; datatype boolean.
  • Client domicile is an ISO 3166-1 alpha-2 code.
  • Booking center is an ISO 3166-1 alpha-2 code.

Policy decision

applies to policy decision
  • Record exactly one decision timestamp with an explicit timezone (Z or an offset).
  • decisionOutcome: minimum count 1; maximum count 1; allowed values OutcomeDeny, OutcomeEscalate, OutcomeHumanApproval, OutcomeAllow, OutcomeObserve.
  • Every decision names the rule and rule version that decided it (Filing Cabinet, slide 14).
  • Every decision references a registered use case (Kognita 0.4 item 8).

Evidence record

applies to evidence record
  • eventType: minimum count 1; maximum count 1.
  • Evidence records are hash-chained (Kognita control frameworks).
  • An evidence record states when the event happened (MAS AIRG 5.23(d)).
  • previousRecordHash: maximum count 1.
  • Every evidence record names the hash of the record before it, except the first record of a chain, which is marked chain start (Kognita control frameworks: hash-chained evidence).

Currency code

applies to anything with currency
  • Currency is an ISO 4217 alphabetic code, such as SGD.

Country code

applies to anything with residence country, anything with domicile country, anything with nationality country, anything with jurisdiction code, anything with tax residence country
  • Residence is an ISO 3166-1 alpha-2 code, such as SG.
  • Domicile is an ISO 3166-1 alpha-2 code, such as SG.
  • Nationality is an ISO 3166-1 alpha-2 code, such as SG.
  • Jurisdiction is an ISO 3166-1 alpha-2 code, such as SG.
  • Tax residence is an ISO 3166-1 alpha-2 code, such as SG.

Legal Entity Identifier

applies to anything with legal entity identifier
  • An LEI is 18 letters or digits followed by two check digits (ISO 17442).

ISIN

applies to anything with ISIN
  • An ISIN is a two-letter prefix, nine letters or digits and a check digit (ISO 6166).

CFI code

applies to anything with CFI code
  • A CFI code is six letters (ISO 10962).

EU sustainability assessment

applies to suitability assessment
  • EU suitability cites one recorded sustainability preference, including explicit no preference (2017/565 Art. 54(2)(a), (5), as amended by 2021/1253 Art. 1(6)).

Sustainability preference

applies to sustainability preference
  • sustainabilityPreferenceKind: minimum count 1.
  • minimumTaxonomyAlignedProportion: maximum count 1; datatype decimal; minimum inclusive 0.0; maximum inclusive 1.0.
  • minimumSFDRSustainableProportion: maximum count 1; datatype decimal; minimum inclusive 0.0; maximum inclusive 1.0.
  • paiConsiderationElements: datatype string.
  • Record the minimum proportion for each selected category and client-determined PAI elements; no preference is exclusive and has no category details (2017/565 Art. 2(7)(a)-(c)).

RunShape

applies to run
  • Record hasBudget (roadmap milestone 0.3).
  • Record runState (roadmap milestone 0.3).
  • A resumed run must have a resumption referencing a resolved approval (0.3 item 2).

RunBudgetShape

applies to run budget
  • Record maxCalls (roadmap milestone 0.3).
  • Record maxTokens (roadmap milestone 0.3).
  • Record maxCostUSD (roadmap milestone 0.3).
  • Record wallClockSeconds (roadmap milestone 0.3).
  • Record consumedCalls (roadmap milestone 0.3).
  • Record consumedTokens (roadmap milestone 0.3).
  • Record consumedCostUSD (roadmap milestone 0.3).
  • Record elapsedSeconds (roadmap milestone 0.3).
  • Record classificationCeiling (roadmap milestone 0.3).

BudgetExceededDenialShape

applies to budget exceeded denial
  • Record exceededBudget (roadmap milestone 0.3).
  • Record budgetDimension (roadmap milestone 0.3).
  • decisionOutcome: required value OutcomeDeny.

CheckpointShape

applies to run checkpoint
  • Record inRun (roadmap milestone 0.3).
  • Record contentHash (roadmap milestone 0.3).

ApprovalRequestShape

applies to approval request
  • Record requestsDecision (roadmap milestone 0.3).
  • Record atCheckpoint (roadmap milestone 0.3).

ApprovalResolutionShape

applies to approval resolution
  • Record resolvesApproval (roadmap milestone 0.3).
  • Record approvalResolution (roadmap milestone 0.3).
  • wasAttributedTo: minimum count 1; maximum count 1; class Person.
  • generatedAtTime: minimum count 1; maximum count 1; datatype dateTime.

RunResumptionShape

applies to run resumption
  • Record inRun (roadmap milestone 0.3).
  • Record resumedFrom (roadmap milestone 0.3).
  • Record resolvedApproval (roadmap milestone 0.3).
  • Resumption must cite a resolved approval for its run checkpoint (0.3 item 2).

ModelCallShape

applies to model call
  • Record usesModel (Kognita 0.4 item 8).
  • Record modelName (roadmap milestone 0.3).
  • Record modelVersion (roadmap milestone 0.3).
  • Record promptTemplateVersion (roadmap milestone 0.3).
  • Record promptHash (roadmap milestone 0.3).
  • Record responseHash (roadmap milestone 0.3).
  • Record providedBy (roadmap milestone 0.3).
  • Record inputTokens (roadmap milestone 0.3).
  • Record outputTokens (roadmap milestone 0.3).

ToolCallShape

applies to tool call
  • Record responseHash (roadmap milestone 0.3).
  • Record invokedTool (roadmap milestone 0.3).

MCPProxiedCallShape

applies to MCP proxied call
  • Record mcpBackend (roadmap milestone 0.3).

ClassifierOutputShape

applies to classifier output
  • Record classifierModelId (roadmap milestone 0.3).
  • Record classifierVersion (roadmap milestone 0.3).
  • Record classifierLabel (roadmap milestone 0.3).
  • Record inputHash (roadmap milestone 0.3).
  • Record confidence (roadmap milestone 0.3).

EnvelopeAttributeShape

applies to envelope attribute
  • Record attributeName (roadmap milestone 0.3).
  • Record attributeValue (roadmap milestone 0.3).
  • Record attributeSource (roadmap milestone 0.3).
  • Classifier-derived attribute needs calibrated confidence and classifier version (0.3 item 4).
  • Identity attributes come from authentication, never classifier text (0.3 item 4 and Tier 0).

ClassifierThresholdShape

applies to anything with has attribute, anything with escalation threshold
  • escalationThreshold: maximum count 1; datatype decimal; minimum inclusive 0.0; maximum inclusive 1.0.
  • Below-threshold or missing-threshold classifier evidence cannot yield ALLOW (0.3 item 4).
  • Classifier fills gaps only; a typed attribute remains authoritative (0.3 item 4).

PolicyCheckShape

applies to policy check
  • Record checkedRule (roadmap milestone 0.3).
  • Record policyContentHash (roadmap milestone 0.3).

RetrievedItemShape

applies to retrieved item
  • Record contentHash (roadmap milestone 0.3).
  • Record embeddingModel (roadmap milestone 0.3).

RetentionEntryShape

applies to retention entry
  • Record contentHash (roadmap milestone 0.3).
  • Record underUseCase (roadmap milestone 0.3).
  • storeLocator: maximum count 1; datatype anyURI.

RetentionPolicyShape

applies to retention policy
  • Record retentionDuration (roadmap milestone 0.3).

ErasureShape

applies to erasure event
  • Record erasedEntry (roadmap milestone 0.3).
  • Record contentHash (roadmap milestone 0.3).
  • Erasure removes stored content reference and keeps the same hash on the entry, event and surviving chain evidence (0.3 item 7).

ReconstructionReportShape

applies to reconstruction report
  • Record partOfInteraction (roadmap milestone 0.3).
  • Record chainVerified (roadmap milestone 0.3).
  • Record pinnedHashesVerified (roadmap milestone 0.3).
  • Cover all ten questions explicitly, marking absent evidence not recorded (0.3 item 8).
  • Report hash mismatches as findings (0.3 item 8).

QuestionCoverageShape

applies to question coverage
  • Record reconstructionQuestion (roadmap milestone 0.3).
  • Record coverageStatus (roadmap milestone 0.3).
  • Answered question must cite recorded evidence (0.3 item 8).

ReconstructionFindingShape

applies to reconstruction finding
  • Record findingKind (roadmap milestone 0.3).
  • Record findingRecord (roadmap milestone 0.3).
  • Hash mismatch finding records different expected and observed hashes (0.3 item 8).

GatewayConfigurationShape

applies to AI use case
  • Record gatewayFailureMode (roadmap milestone 0.3).
  • retentionPolicy: minimum count 1; maximum count 1; class RetentionPolicy.

GovernedCallShape

applies to governed call
  • callAgent: maximum count 1.
  • systemTrigger: maximum count 1.
  • Record inRun (roadmap milestone 0.3).
  • Record governedByDecision (roadmap milestone 0.3).
  • Record clientConfiguration (roadmap milestone 0.3).
  • Call identity must be an agent bound to the authenticated client or a listed system trigger (0.3 Tier 0).
  • A call governed by a held decision requires that decision in the same run and an ApprovalRequest at a checkpoint in the call's run with an ApprovalApproved resolution (KOG roadmap 0.3 item 2; MAS AIRG 5.9).

Pinned decision checks

applies to policy decision
  • Every policy decision carries pinned component checks (0.3 item 7).

Approval checkpoint consistency

applies to approval request
  • An approval request holds a HUMAN_APPROVAL decision at a suspended checkpoint of the same run (0.3 item 2).

IngestedDocumentShape

applies to ingested document
  • Record parserName (Kognita 0.4 item 1).
  • Record parserVersion (Kognita 0.4 item 1).
  • Record contentHash (Kognita 0.4 item 1).

DocumentElementShape

applies to document element
  • Record ingestedFrom (Kognita 0.4 item 1).
  • Record pageNumber (Kognita 0.4 item 1).
  • Record sectionPath (Kognita 0.4 item 1).
  • Record characterStart (Kognita 0.4 item 1).
  • Record characterEnd (Kognita 0.4 item 1).
  • Record classification (Kognita 0.4 item 1).
  • Record contentHash (Kognita 0.4 item 1).

DeclarativePolicyShape

applies to declarative policy
  • Record versionId (Kognita 0.4 item 3).
  • Record containsRule (Kognita 0.4 item 3).
  • Record policySerialization (Kognita 0.4 item 3).
  • Record hasPolicyTest (Kognita 0.4 item 3).

PolicyTestCaseShape

applies to policy test case
  • Record testsRule (Kognita 0.4 item 4).
  • Record givenAttributes (Kognita 0.4 item 4).
  • Record whenAction (Kognita 0.4 item 4).
  • Record expectedOutcome (Kognita 0.4 item 4).

PolicyPackShape

applies to policy pack
  • Record versionId (Kognita 0.4 item 5).
  • Record packPolicy (Kognita 0.4 item 5).
  • Record fixtureLocator (Kognita 0.4 item 5).

RMReviewEditShape

applies to RM review edit
  • Record diffLocator (Kognita 0.4 item 11).
  • Record beforeHash (Kognita 0.4 item 11).
  • Record afterHash (Kognita 0.4 item 11).
  • Record editDiffHash (Kognita 0.4 item 11).

ReviewQueueEntryShape

applies to review queue entry
  • Record inReviewQueue (Kognita 0.4 item 13).
  • Record partOfInteraction (Kognita 0.4 item 13).
  • Record queueReason (Kognita 0.4 item 13).
  • Record queuedAt (Kognita 0.4 item 13).
  • Record reviewBeforeDelivery (Kognita 0.4 item 13).

ReviewCoverageReportShape

applies to review coverage report
  • Record eligibleCount (Kognita 0.4 item 13).
  • Record screenedCount (Kognita 0.4 item 13).
  • Record queuedCount (Kognita 0.4 item 13).
  • Record reviewedCount (Kognita 0.4 item 13).
  • Record underUseCase (Kognita 0.4 item 13).
  • Record windowStart (Kognita 0.4 item 13).
  • Record windowEnd (Kognita 0.4 item 13).

CircuitBreakerTripShape

applies to circuit breaker trip
  • Record trippedBreaker (Kognita 0.4 item 14).
  • Record flagThreshold (Kognita 0.4 item 14).
  • Record observedFlags (Kognita 0.4 item 14).
  • Record prohibitingRule (Kognita 0.4 item 14).
  • Record windowStart (Kognita 0.4 item 14).
  • Record windowEnd (Kognita 0.4 item 14).

AffectedClientQueryShape

applies to affected client query
  • Record windowStart (Kognita 0.4 item 14).
  • Record windowEnd (Kognita 0.4 item 14).
  • Record queryResult (Kognita 0.4 item 14).

InteractionOutcomeShape

applies to interaction outcome
  • Record outcomeKind (Kognita 0.4 item 15).
  • Record partOfInteraction (Kognita 0.4 item 15).
  • Record affectedOutput (Kognita 0.4 item 15).
  • Record implicatedModel (Kognita 0.4 item 15).
  • Record outcomeApprover (Kognita 0.4 item 15).

TwoSignatureShape

applies to anything with requires two signatures
  • A communication requiring two signatures must name two distinct human signers (Kognita 0.4 item 12).

PastReviewDecisionShape

applies to policy decision
  • A decision made after the use-case review time expires must be DENY (Kognita 0.4 item 8).

RiskAppetiteDecisionShape

applies to policy decision
  • A decision outside the recorded risk appetite must be DENY (Kognita 0.4 item 8).

CoverageCountsShape

applies to review coverage report
  • Coverage counts must satisfy reviewed <= queued <= screened <= eligible (Kognita 0.4 item 13).

CircuitBreakerConfigurationShape

applies to circuit breaker
  • Record flagThreshold (Kognita 0.4 item 14).
  • Record windowStart (Kognita 0.4 item 14).
  • Record windowEnd (Kognita 0.4 item 14).

FlightRecorderSessionShape

applies to flight recorder session
  • Record recordedRun (Kognita 0.4 item 7).
  • Record contentHash (Kognita 0.4 item 7).

ReplayShape

applies to replay
  • Record replaysSession (Kognita 0.4 item 7).
  • Record replayReport (Kognita 0.4 item 7).

PolicyChangeShape

applies to policy change
  • Record changesPolicy (Kognita 0.4 item 3).
  • Record deltaText (Kognita 0.4 item 3).
  • Record changeStage (Kognita 0.4 item 3).

NearMissClassificationShape

applies to near miss
  • Record nearMissKind (Kognita 0.4 item 15).
  • Record partOfInteraction (Kognita 0.4 item 15).

ReviewSamplingRateShape

applies to anything with low risk sampling rate
  • Record lowRiskSamplingRate (Kognita 0.4 item 13).

AffectedClientCountShape

applies to affected client result
  • Affected-client count must equal distinct recipients of the returned communications (0.4 item 14).
  • Each affected communication must record withClient (Kognita 0.4 item 14).

ModelCallCardShape

applies to model call
  • Pinned model calls must link to their approved model card and exact version (0.4 item 8).

Live use case risk appetite

applies to AI use case
  • An explicit failed risk-appetite evaluation prevents activation (Kognita 0.4 item 8; MAS AIRG 4.11).

AgentIdentityShape

applies to AI agent
  • pwm:workloadIdentity: minimum count 1, maximum count 1, class pwm:WorkloadIdentity.
  • pwm:agentOwner: minimum count 1, maximum count 1, class fibo-fnd-aap-ppl:Person.
  • pwm:operatesFor: minimum count 1, maximum count 1, class fibo-fnd-aap-ppl:Person.
  • pwm:hasPurpose: minimum count 1, class pwm:Purpose.
  • pwm:deploymentDate: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • pwm:deploymentModel: minimum count 1, maximum count 1, class pwm:AIModel.
  • pwm:deploymentEnvironment: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:agentStatus: minimum count 1, maximum count 1.
  • pwm:approvedDeploymentVersion: minimum count 1, maximum count 1, datatype xsd:string.

WorkloadIdentityShape

applies to workload identity
  • pwm:credentialReference: minimum count 1, maximum count 1, datatype xsd:anyURI.
  • pwm:credentialIssuer: minimum count 1, maximum count 1, datatype xsd:anyURI.
  • pwm:workloadSubject: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:deploymentEnvironment: minimum count 1, maximum count 1, datatype xsd:string.
  • Workload subjects must not be shared (Kognita 0.5 item 1).
  • Credential references must not be shared by distinct workload identities (Kognita 0.5 item 1).

AgentDecisionShape

applies to governed call
  • Agent calls require live approved deployment, current review, registered decision use case and unrevoked own workload identity (Kognita 0.5 item 1).

AuthorityScopeShape

applies to authority scope
  • pwm:aboutClient: minimum count 1, class pwm:Client.
  • pwm:hasPurpose: minimum count 1, class pwm:Purpose.
  • pwm:actionVerb: minimum count 1.
  • pwm:classification: minimum count 1.
  • pwm:scopeTool: minimum count 1, class pwm:Tool.
  • pwm:scopeField: minimum count 1, datatype xsd:string.
  • pwm:jurisdictionCode: minimum count 1, datatype xsd:string.
  • scopeAutonomyLevel: minimum count 1; maximum count 1.

HumanEntitlementShape

applies to human entitlement
  • pwm:authorityScope: minimum count 1, maximum count 1, class pwm:AuthorityScope.
  • pwm:validFrom: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • pwm:validUntil: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • validFrom must be earlier than validUntil (Kognita 0.5 item 2).

DelegationScopeShape

applies to delegated authority
  • pwm:authorityScope: minimum count 1, maximum count 1, class pwm:AuthorityScope.
  • pwm:validFrom: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • pwm:delegationDepth: minimum count 1, maximum count 1, datatype xsd:nonNegativeInteger.
  • pwm:maxDelegationDepth: minimum count 1, maximum count 1, datatype xsd:nonNegativeInteger.
  • validFrom must be earlier than validUntil (Kognita 0.5 item 2).
  • delegationDepth must not exceed maxDelegationDepth (Kognita 0.5 item 2).
  • Grantor must hold an entitlement covering the complete explicit delegation scope and time (Kognita 0.5 item 2).

AuthorityRevocationShape

applies to authority revocation
  • pwm:revokedAuthority: minimum count 1, maximum count 1.
  • pwm:revokedAt: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • pwm:rationale: minimum count 1, maximum count 1, datatype xsd:string.

LiveAuthorityDecisionShape

applies to governed call
  • Both call and decision must cite the same live delegation to this agent and a live, unrevoked human entitlement covering the complete delegation scope (Kognita 0.5 item 2).

SubDelegationShape

applies to anything with parent delegation
  • parentDelegation: maximum count 1; class DelegatedAuthority.
  • Sub-delegation must be acyclic, narrow every explicit scope set and period, and respect depth (Kognita 0.5 item 9).

PromotionGateShape

applies to autonomy promotion gate
  • pwm:minimumSample: minimum count 1, maximum count 1, datatype xsd:positiveInteger.
  • pwm:minimumWindowSeconds: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:qualityThreshold: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:exceptionRateThreshold: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:requiresFullTraceability: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:stepDownThreshold: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:qualityThreshold: minimum inclusive 0, maximum inclusive 1.
  • pwm:exceptionRateThreshold: minimum inclusive 0, maximum inclusive 1.
  • pwm:stepDownThreshold: minimum inclusive 0, must be earlier or lower than pwm:qualityThreshold.
  • pwm:minimumWindowSeconds: minimum exclusive 0.

AutonomyTransitionShape

applies to autonomy transition
  • pwm:transitionAgent: minimum count 1, maximum count 1, class pwm:AIAgent.
  • pwm:promotionGate: minimum count 1, maximum count 1, class pwm:AutonomyPromotionGate.
  • pwm:fromAutonomyLevel: minimum count 1, maximum count 1.
  • pwm:toAutonomyLevel: minimum count 1, maximum count 1.
  • pwm:sampleCount: minimum count 1, maximum count 1, datatype xsd:nonNegativeInteger.
  • pwm:observedQuality: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:observedExceptionRate: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:fullyReconstructable: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:transitionApproved: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:windowStart: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • pwm:windowEnd: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • pwm:windowStart: must be earlier or lower than pwm:windowEnd.
  • pwm:observedQuality: minimum inclusive 0, maximum inclusive 1.
  • pwm:observedExceptionRate: minimum inclusive 0, maximum inclusive 1.

AutonomyEvidenceShape

applies to autonomy transition
  • Promotion requires evidence thresholds, sample/window, owner and independent validation; step-down is one level and alerts owners (Kognita 0.5 item 3).

BlastRadiusLimitShape

applies to blast radius limit
  • pwm:limitDimension: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:limitWindowBasis: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:limitUnit: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:limitMaximum: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:limitMaximum: minimum inclusive 0.
  • pwm:limitDimension: allowed values ( 'records' 'clients' 'value' 'actions' 'transactions' 'clients-contacted' ).
  • pwm:limitWindowBasis: allowed values ( 'action' 'run' 'minute' 'hour' 'UTC-day' ).

BlastRadiusObservationShape

applies to blast radius observation
  • pwm:observedLimit: minimum count 1, maximum count 1, class pwm:BlastRadiusLimit.
  • pwm:projectedConsumption: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:projectedConsumption: minimum inclusive 0.

BlastRadiusDecisionShape

applies to policy decision
  • Projected consumption exceeding any agent blast-radius cap cannot be ALLOW (Kognita 0.5 item 4).
  • ALLOW requires exactly one observation for each calling agent and delegation scope limit and no unrelated limits.

ToolAllowListShape

applies to tool call
  • A tool outside the agent allow-list must be DENY (Kognita 0.5 item 5).

FieldMinimizationShape

applies to tool call
  • Requested fields outside purpose-specific agent permissions must be DENY (Kognita 0.5 item 5).

ExternalContentShape

applies to external content
  • pwm:trustLabel: minimum count 1.
  • pwm:trustLabel: required value pwm:TrustExternal.
  • pwm:contentHash: minimum count 1, maximum count 1, datatype xsd:string.

ContentIsolationShape

applies to tool call
  • Externally tainted runs cannot write; client-data taint cannot reach public-content tools (Kognita 0.5 item 6).

AuthenticatedInstructionShape

applies to authenticated instruction
  • pwm:instructionSource: minimum count 1, maximum count 1, datatype xsd:anyURI.
  • pwm:instructionPrincipal: minimum count 1, maximum count 1, class fibo-fnd-aap-ppl:Person.
  • pwm:assuranceLevel: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:assuranceLevel: allowed values ( 'channel-only' 'authenticated' 'step-up' ).
  • pwm:transactionContext: minimum count 1, maximum count 1, datatype xsd:string.

InstructionAssuranceShape

applies to anything with requires step up
  • Allowed high-risk instructions require signed step-up evidence and a completed approval of a held decision in the same run, strictly before the allowed decision (Kognita 0.5 item 7).

BehaviorBaselineShape

applies to behavior baseline
  • pwm:baselineAgent: minimum count 1, maximum count 1, class pwm:AIAgent.
  • pwm:versionId: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:baselineMetric: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:baselineExpected: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:anomalyThreshold: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:windowStart: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • pwm:windowEnd: minimum count 1, maximum count 1, datatype xsd:dateTime.

BehaviorAnomalyShape

applies to behavior anomaly
  • pwm:anomalyBaseline: minimum count 1, maximum count 1, class pwm:BehaviorBaseline.
  • pwm:anomalyValue: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:containmentApplied: required value true.
  • pwm:anomalyIncident: minimum count 1, maximum count 1, class pwm:AIIncident.
  • pwm:notifiedRole: minimum count 1, class pwm:AccountabilityRole.

CapabilityGrantShape

applies to capability grant
  • pwm:grantee: minimum count 1, maximum count 1, class pwm:AIAgent.
  • pwm:grantedCapability: minimum count 1, maximum count 1, class pwm:Tool.
  • pwm:authorityScope: minimum count 1, maximum count 1, class pwm:AuthorityScope.
  • pwm:validFrom: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • pwm:validUntil: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • pwm:validFrom: must be earlier or lower than pwm:validUntil.

LiveGrantShape

applies to anything with relies on grant
  • Used capability grants must match agent and tool and remain unexpired and unrevoked (Kognita 0.5 item 10).

AgentFleetShape

applies to agent fleet
  • pwm:fleetMember: minimum count 1, class pwm:AIAgent.
  • pwm:fleetQuota: minimum count 1, maximum count 1, datatype xsd:positiveInteger.
  • pwm:fleetIsolation: minimum count 1, maximum count 1, datatype xsd:string.

FleetHaltShape

applies to fleet halt
  • pwm:haltScope: minimum count 1, maximum count 1, class pwm:AgentFleet.
  • prov:generatedAtTime: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • pwm:rationale: minimum count 1, maximum count 1, datatype xsd:string.

SandboxConstraintsShape

applies to sandbox constraints
  • pwm:filesystemScope: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:permitsExport: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:permitsPersistence: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:autoStopSeconds: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:cpuSeconds: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:memoryBytes: minimum count 1, maximum count 1, datatype xsd:nonNegativeInteger.
  • pwm:diskBytes: minimum count 1, maximum count 1, datatype xsd:nonNegativeInteger.
  • pwm:autoStopSeconds: minimum exclusive 0.
  • pwm:cpuSeconds: minimum exclusive 0.

ReversibilityShape

applies to tool
  • pwm:reversibilityClass: minimum count 1, maximum count 1.
  • Reversible and compensatable tools register a compensating action (Kognita 0.5 item 13).

IrreversibleApprovalShape

applies to tool call
  • High-autonomy irreversible action requires human approval resolving a held HUMAN_APPROVAL decision at a suspended checkpoint in the call's run, strictly before the call (Kognita 0.5 item 13).

AdversarialTestShape

applies to adversarial test case
  • pwm:threatModel: minimum count 1, maximum count 1, class pwm:ThreatModel.
  • pwm:attackKind: minimum count 1, maximum count 1, datatype xsd:string.

ThreatModelShape

applies to threat model
  • pwm:versionId: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:threatModelReference: minimum count 1, maximum count 1, datatype xsd:anyURI.

EnforcementAttestationShape

applies to enforcement attestation
  • pwm:attestedUseCase: minimum count 1, maximum count 1, class pwm:AIUseCase.
  • prov:generatedAtTime: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • prov:wasAttributedTo: minimum count 1, maximum count 1, class fibo-fnd-aap-ppl:Person.
  • pwm:gatewayOnlyCredentials: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:egressRestricted: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:evidenceStoreIsolated: minimum count 1, maximum count 1, datatype xsd:boolean.

DeploymentEnforcementShape

applies to AI use case
  • pwm:deploymentMode: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:deploymentMode: allowed values ( 'gateway' 'library' ).
  • Tier 2 or above requires gateway mode and attestation of isolated credentials, routing and evidence (Kognita 0.5 item 15).

BoundedToolShape

applies to tool
  • pwm:maximumDurationSeconds: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:maximumDurationSeconds: minimum exclusive 0.
  • pwm:supportsCancellation: minimum count 1, maximum count 1, datatype xsd:boolean.
  • Uncancellable tools need compensation or steps with intervening decisions (Kognita 0.5 item 16).

ActionStopShape

applies to action stop
  • pwm:stoppedCall: minimum count 1, maximum count 1, class pwm:ToolCall.
  • pwm:stopOutcome: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:stopOutcome: allowed values ( 'canceled' 'completed' 'timed-out' 'unsupported' ).
  • pwm:stopReason: minimum count 1, maximum count 1, datatype xsd:string.
  • prov:generatedAtTime: minimum count 1, maximum count 1, datatype xsd:dateTime.

AuthorityRequestScopeShape

applies to tool call
  • A live delegation must authorize this tool, every declared tool verb, purpose, client, fields, exactly one jurisdictionCode and exactly one classification at or below the scope ceiling by rank (Kognita 0.5 item 2).

ScopeNarrowingLimitsShape

applies to anything with parent delegation
  • Sub-delegation retains all parent numeric caps and may not increase autonomy (Kognita 0.5 item 9).

AutonomyCallShape

applies to tool call
  • Calls cannot exceed effective or maximum autonomy or the delegated scope ceiling; L4 routes to human approval (Kognita 0.5 item 3).

FleetHaltDecisionShape

applies to governed call
  • A halted fleet member must be denied on the next decision (Kognita 0.5 item 11).

ToolDurationShape

applies to anything with call duration seconds
  • pwm:callDurationSeconds: maximum count 1, datatype xsd:decimal, minimum inclusive 0.
  • A call beyond maximum duration needs an evidenced timeout (Kognita 0.5 item 16).

Exclusive workload identity

applies to AI agent
  • Each registered agent has its own workload identity (Kognita 0.5 item 1).

DelegationCeilingShape

applies to delegated authority
  • A delegation's permitted data class must cover the classification of its authority scope, and a child delegation cannot permit a higher class than its parent (Kognita 0.5 item 2).

AgentAutonomyShape

applies to AI agent
  • An agent's effective autonomy level cannot exceed its maximum autonomy level (Kognita 0.5 item 3).

RiskControlApprovalRecordShape

applies to risk control approval
  • Record exactly one approverRole: RiskCommittee, DataGovernanceCommittee or ChiefRiskOfficer.
  • pwm:approvedAt: minimum count 1, maximum count 1, datatype xsd:dateTime.

Live authority status

applies to authority live status
  • statusOf: minimum count 1; maximum count 1; one of classes DelegatedAuthority, KillSwitch.
  • asOf must carry an explicit timezone (Z or an offset) so decision-time comparison cannot use an implicit timezone.
  • observedBy: minimum count 1; class Agent.
  • revoked: maximum count 1; datatype boolean.
  • engaged: maximum count 1; datatype boolean.
  • Record the matching live boolean, subject, xsd:dateTime asOf and observer (KOG 0.4 item 14; 0.3 item 2; MAS AIRG 5.9).

Live delegation at decision time

applies to policy decision
  • An ALLOW relying on delegated authority requires a live status and must not have a latest status showing it revoked as of decision time (KOG 0.4 item 14; 0.3 item 2; MAS AIRG 5.9).

Live kill switch at decision time

applies to policy decision
  • A decision made while a covering kill switch status is engaged must not be ALLOW (KOG 0.4 item 14: circuit breaker; MAS AIRG 5.9).

SigningKeyShape

applies to signing key
  • pwm:keyId: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:publicKeyReference: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:escrowReference: minimum count 0, maximum count 1, datatype xsd:string.

SignedEvidenceShape

applies to evidence record
  • pwm:signature: minimum count 0, maximum count 1, datatype xsd:string.
  • pwm:signedBy: minimum count 0, maximum count 1, class pwm:SigningKey.
  • pwm:evidenceSequence: minimum count 0, maximum count 1, datatype xsd:nonNegativeInteger.
  • A signed evidence record names its signing key and its sequence, so the signature can be checked against the key registry and any revocation cutoff (Kognita 0.7 items T1 and R2).
  • A record signed with a key at or after that key's revocation sequence is not valid (Kognita 0.7 item R2; revoked keys fail closed).

KeyRotationShape

applies to key rotation
  • pwm:rotatesFrom: minimum count 1, maximum count 1, class pwm:SigningKey.
  • pwm:rotatesTo: minimum count 1, maximum count 1, class pwm:SigningKey.
  • pwm:rotatesFrom: must differ from rotatesTo.

KeyRevocationShape

applies to key revocation
  • pwm:revokesKey: minimum count 1, maximum count 1, class pwm:SigningKey.
  • pwm:effectiveFromSequence: minimum count 1, maximum count 1, datatype xsd:nonNegativeInteger.

EvidenceAnchorShape

applies to evidence anchor
  • pwm:anchoredRecordHash: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:anchoredSequence: minimum count 1, maximum count 1, datatype xsd:nonNegativeInteger.
  • pwm:anchorLocation: minimum count 1, maximum count 1, datatype xsd:string.
  • prov:generatedAtTime: minimum count 1, maximum count 1, datatype xsd:dateTime.

RecoveryBundleShape

applies to recovery bundle
  • pwm:bundleHash: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:recoveryCutoffSequence: minimum count 1, maximum count 1, datatype xsd:nonNegativeInteger.
  • pwm:restoreVerified: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:includesRetentionState: minimum count 1, maximum count 1, datatype xsd:boolean.

ExternalReviewSubmissionShape

applies to external review submission
  • pwm:reviewedFromSequence: minimum count 1, maximum count 1, datatype xsd:nonNegativeInteger, not after reviewedToSequence.
  • pwm:reviewedToSequence: minimum count 1, maximum count 1, datatype xsd:nonNegativeInteger.
  • pwm:signedBy: minimum count 1, maximum count 1, class pwm:SigningKey.

ModelProviderRegistrationShape

applies to model provider registration
  • pwm:registeredProvider: minimum count 1, maximum count 1, class pwm:ModelProvider.
  • pwm:dependentUseCase: minimum count 1, class pwm:AIUseCase.
  • pwm:dataResidency: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:dataRetentionTerms: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:encryptionDescription: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:incidentNotificationTerms: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:availabilityCommitment: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:modelChangeNotice: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:exitPlanReference: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:regulatoryCooperation: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:trainingUse: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:auditRights: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:subProcessor: minimum count 1, datatype xsd:string.

RegisteredProviderCallShape

applies to model call
  • A model call names a provider that has no model provider registration (Kognita 0.7 item R3, R3-UNREGISTERED).

ProviderFailoverShape

applies to provider failover
  • pwm:failoverFrom: minimum count 1, maximum count 1, class pwm:ModelCard.
  • pwm:failoverTo: minimum count 1, maximum count 1, class pwm:ModelCard.
  • pwm:failoverUseCase: minimum count 1, maximum count 1, class pwm:AIUseCase.
  • pwm:failoverTrigger: minimum count 1, maximum count 1.
  • A failover target must be a model card approved as a fallback for the use case (Kognita 0.7 item R5, R5-UNAPPROVED-FALLBACK).

ModelChangeDetectionShape

applies to model change detection
  • pwm:approvedModelVersion: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:servedModelVersion: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:triggersRevalidation: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:detectedOnCall: minimum count 1, maximum count 1, class pwm:ModelCall.
  • A served model version that differs from the approved version must trigger revalidation (Kognita 0.7 item R7, R7-MODEL-CHANGED).

AIDependencyShape

applies to AI dependency
  • pwm:dependencyLayer: minimum count 1, maximum count 1.
  • pwm:dependencyOf: minimum count 1, maximum count 1, class pwm:AIUseCase.
  • pwm:dependencyTarget: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:dependencyDeclared: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:dependencyObserved: minimum count 1, maximum count 1, datatype xsd:boolean.

DependencyDivergenceShape

applies to dependency divergence
  • pwm:divergenceKind: minimum count 1, maximum count 1.
  • pwm:divergenceDependency: minimum count 1, maximum count 1, class pwm:AIDependency.
  • An undeclared divergence needs a dependency that is observed and not declared; an unused one needs a dependency that is declared and not observed (Kognita 0.7 item R8).

ConcentrationMeasureShape

applies to concentration measure
  • pwm:concentrationLayer: minimum count 1, maximum count 1.
  • pwm:concentrationTarget: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:criticalUseCaseCount: minimum count 1, maximum count 1, datatype xsd:nonNegativeInteger, not above criticalUseCaseTotal.
  • pwm:criticalUseCaseTotal: minimum count 1, maximum count 1, datatype xsd:nonNegativeInteger.
  • pwm:criticalUseCaseShare: minimum count 1, maximum count 1, datatype xsd:decimal, between 0 and 1.

FailureDrillShape

applies to failure drill
  • pwm:drillKind: minimum count 1, maximum count 1.
  • pwm:drillUseCase: minimum count 1, maximum count 1, class pwm:AIUseCase.
  • pwm:evidenceCompleteAfterDrill: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:gatewayFailureMode: minimum count 1, maximum count 1.

RecordClassRetentionShape

applies to record class retention
  • pwm:recordClass: minimum count 1, maximum count 1.
  • A record class has a retention period in days or is held indefinitely, not both and not neither (Kognita 0.7 item R11).

IncidentReportShape

applies to incident report
  • pwm:reportsIncident: minimum count 1, maximum count 1, class pwm:AIIncident.
  • pwm:redactedReport: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:controlRan: minimum count 1, datatype xsd:string.

AssurancePolicyShape

applies to assurance policy
  • pwm:assuranceUseCase: minimum count 1, maximum count 1, class pwm:AIUseCase.
  • pwm:assuranceAudience: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:maxAdverseOutcomes: minimum count 1, maximum count 1, datatype xsd:positiveInteger.
  • pwm:outcomeWindowDays: minimum count 1, maximum count 1, datatype xsd:positiveInteger, between 1 and 3650.

AgentManifestShape

applies to agent manifest
  • pwm:manifestSchemaVersion: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:manifestOfAgent: minimum count 1, maximum count 1, class pwm:AIAgent.
  • pwm:manifestDigest: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:signedBy: minimum count 1, maximum count 1, class pwm:SigningKey.

ExternalAgentAssertionShape

applies to external agent assertion
  • pwm:assertionIssuer: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:assertionSubject: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:assertionPrincipal: minimum count 1, maximum count 1, class fibo-fnd-aap-ppl:Person.
  • pwm:assertionPurpose: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:assertionAudience: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:assertionScope: minimum count 1, datatype xsd:string.
  • pwm:assertionDepth: minimum count 1, maximum count 1, at most 16.
  • pwm:issuedAt: minimum count 1, maximum count 1, datatype xsd:dateTime, before expiresAt.
  • pwm:expiresAt: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • An assertion has a parent exactly when its delegation depth is above zero (Kognita 0.8 item A2).

DeploymentAttestationShape

applies to deployment attestation
  • pwm:attestationKind: minimum count 1, maximum count 1.
  • pwm:attestationIssuer: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:attestationSubject: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:attestsDeployment: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:attestsPolicy: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:attestationReference: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:issuedAt: minimum count 1, maximum count 1, datatype xsd:dateTime, before expiresAt.
  • pwm:expiresAt: minimum count 1, maximum count 1, datatype xsd:dateTime.

ThirdPartyUpdateApprovalShape

applies to third-party update approval
  • pwm:candidateDigest: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:regressionReference: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:rollbackReference: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:approvedBy: minimum count 1, maximum count 1.

ApprovedComponentShape

applies to approved component
  • pwm:componentName: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:componentVersion: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:provenanceReference: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:sbomDigest: 64 lowercase hexadecimal characters.
  • pwm:imageDigest: sha256: followed by 64 lowercase hexadecimal characters.

ComponentAdvisoryShape

applies to component advisory
  • pwm:advisoryReference: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:affectsComponent: minimum count 1, maximum count 1, class pwm:ApprovedComponent.
  • pwm:quarantined: minimum count 1, maximum count 1, datatype xsd:boolean.

TechnologyTransferClassificationShape

applies to technology transfer classification
  • pwm:transferType: minimum count 1, maximum count 1.
  • pwm:technologyOrigin: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:deploymentJurisdiction: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:transferPolicyReference: minimum count 1, maximum count 1, datatype xsd:string.

ProductionOutcomeShape

applies to production outcome
  • pwm:outcomeId: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:outcomeSourceReference: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:intendedResult: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:actualResult: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:adverse: minimum count 1, maximum count 1, datatype xsd:boolean.

SupervisoryConformanceReportShape

applies to supervisory conformance report
  • pwm:conformanceProfile: minimum count 1, maximum count 1.
  • pwm:reportAsOf: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • pwm:coversControl: minimum count 1, datatype xsd:string.

VerifierShape

applies to verifier
  • pwm:verifierAgent: minimum count 1, maximum count 1, class pwm:AIAgent.
  • pwm:verifierModelCard: minimum count 1, maximum count 1, class pwm:ModelCard.

ClaimVerificationShape

applies to claim verification
  • pwm:verifiedClaim: minimum count 1, maximum count 1, class pwm:Claim.
  • pwm:verifiedBy: minimum count 1, maximum count 1, class pwm:Verifier.
  • pwm:producerModelCard: minimum count 1, maximum count 1, class pwm:ModelCard.
  • pwm:verificationVerdict: minimum count 1, maximum count 1.
  • A verifier must differ from the producer in model, model version and provider unless the use case explicitly permits self verification (Kognita 0.6 item 7, C7).

EvaluationResultShape

applies to evaluation result
  • pwm:evaluatesAgent: minimum count 1, maximum count 1, class pwm:AIAgent.
  • pwm:evaluatedVersion: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:evaluatedModelVersion: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:evaluationLevel: minimum count 1, maximum count 1.
  • pwm:hasMetricValue: minimum count 1, class pwm:EvaluationMetricValue.
  • pwm:evaluationDeploymentHash: 64 lowercase hexadecimal characters.

EvaluationMetricValueShape

applies to evaluation metric value
  • pwm:metricName: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:metricScore: minimum count 1, maximum count 1, between 0 and 1.

FairnessReportShape

applies to fairness report
  • pwm:fairnessThreshold: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:fairnessFlagged: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:proxyAttribute: minimum count 0, datatype xsd:string.

EvaluationGateShape

applies to evaluation gate
  • pwm:gateUseCase: minimum count 1, maximum count 1, class pwm:AIUseCase.
  • pwm:gateMetricName: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:gateMinimumScore: between 0 and 1.
  • pwm:gateMaximumScore: between 0 and 1.
  • A gate states a minimum or a maximum score, not both and not neither (Kognita 0.6 item 11).

ActivationGateShape

applies to AI agent
  • An agent activated on an evaluation result that misses a minimum gate for its use case (Kognita 0.6 item 11).
  • An agent activated on an evaluation result that exceeds a maximum gate for its use case (Kognita 0.6 item 11).

ReasoningTraceShape

applies to reasoning trace
  • pwm:traceOfCall: minimum count 1, maximum count 1, class pwm:ModelCall.
  • pwm:traceHash: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:retainedIn: minimum count 1, maximum count 1.

ClaimEnvelopeShape

applies to claim
  • A verified claim names its source system and record and the source and retrieval times, so an RM can see why they are seeing it (Kognita 0.6 item 1).

MemoryRecordShape

applies to memory record
  • pwm:memoryScope: minimum count 1, maximum count 1.
  • Relationship memory names its client, an expiry and a correction path (Kognita 0.6 item 8).
  • Institutional memory carries no client; client data never crosses into it (Kognita 0.6 item 8).

LearningExportShape

applies to learning export
  • pwm:exportPurpose: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:anonymizationReference: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:validationReference: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:approvedBy: minimum count 1, maximum count 1.
  • pwm:exportRequester: minimum count 1, maximum count 1, class fibo-fnd-aap-ppl:Person.
  • pwm:exportedRecordId: minimum count 1, datatype xsd:string.

LearningExclusionShape

applies to learning exclusion
  • pwm:exclusionReason: minimum count 1, maximum count 1, datatype xsd:string.
  • An exclusion names a client or a record (Kognita 0.6 item 9).

SensitiveAttributeInferenceShape

applies to sensitive attribute inference
  • pwm:sensitiveKind: minimum count 1, maximum count 1.
  • pwm:permittedByUseCase: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:blockedByPolicy: minimum count 1, maximum count 1, datatype xsd:boolean.
  • pwm:changesClientTreatment: minimum count 1, maximum count 1, datatype xsd:boolean.
  • A special-category inference the use case does not permit must be blocked by policy (Kognita 0.6 item 10).
  • An inference that changes how a client is treated needs a person's review and determination; an AI may flag, never determine (Kognita 0.6 item 10).

CostDisclosureShape

applies to cost disclosure
  • pwm:disclosureTiming: minimum count 1, maximum count 1.
  • pwm:disclosedToClient: minimum count 1, maximum count 1, class pwm:Client.
  • pwm:disclosedAt: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • pwm:coversMandate: minimum count 1, maximum count 1, class pwm:Mandate.
  • pwm:totalCostAmount: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:currency: minimum count 1, maximum count 1.
  • pwm:hasCostItem: minimum count 1, class pwm:CostItem.
  • pwm:includesInducements: minimum count 1, maximum count 1, datatype xsd:boolean.
  • An ex post disclosure states the period it covers (MiFID II Art. 24(4); Delegated Regulation 2017/565 Art. 50).

CostItemShape

applies to cost item
  • pwm:costCategory: minimum count 1, maximum count 1.
  • pwm:amount: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:currency: minimum count 1, maximum count 1.

InducementDisclosureShape

applies to inducement disclosure
  • pwm:disclosesPayment: minimum count 1, maximum count 1, class pwm:ThirdPartyPayment.
  • pwm:disclosedToClient: minimum count 1, maximum count 1, class pwm:Client.
  • pwm:disclosedAt: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • pwm:enhancesServiceQuality: minimum count 1, maximum count 1, datatype xsd:boolean.

ProductApprovalShape

applies to product approval
  • pwm:approvesProduct: minimum count 1, maximum count 1, class pwm:Product.
  • pwm:approvedForTargetMarket: minimum count 1, maximum count 1, class pwm:TargetMarket.
  • pwm:distributionStrategyReference: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:approvedBy: minimum count 1, maximum count 1.
  • prov:generatedAtTime: minimum count 1, maximum count 1, datatype xsd:dateTime.

TargetMarketReviewShape

applies to target market review
  • pwm:reviewsTargetMarket: minimum count 1, maximum count 1, class pwm:TargetMarket.
  • pwm:targetMarketReviewOutcome: minimum count 1, maximum count 1.
  • pwm:reviewedAt: minimum count 1, maximum count 1, datatype xsd:dateTime, before the next review date.
  • pwm:nextReviewDueAt: minimum count 1, maximum count 1, datatype xsd:dateTime.

ComplaintOutcomeShape

applies to complaint outcome
  • pwm:complaintOf: minimum count 1, maximum count 1, class pwm:Complaint.
  • pwm:complaintResolution: minimum count 1, maximum count 1.
  • pwm:reviewedAt: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • Redress needs a currency (Kognita record of the outcome).

SuitabilityReviewShape

applies to suitability review
  • pwm:reviewsSuitability: minimum count 1, maximum count 1, class pwm:SuitabilityAssessment.
  • pwm:suitabilityReviewResult: minimum count 1, maximum count 1.
  • pwm:reviewFrequencyMonths: minimum count 1, maximum count 1, datatype xsd:positiveInteger.
  • pwm:performedAt: minimum count 1, maximum count 1, datatype xsd:dateTime, before the next review date.
  • pwm:nextReviewDueAt: minimum count 1, maximum count 1, datatype xsd:dateTime.

OrderShape

applies to order
  • pwm:orderSide: minimum count 1, maximum count 1.
  • pwm:orderType: minimum count 1, maximum count 1.
  • pwm:orderStatus: minimum count 1, maximum count 1.
  • pwm:orderQuantity: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:inInstrument: minimum count 1, maximum count 1, class fibo-fbc-fi-fi:FinancialInstrument.
  • pwm:orderAccount: minimum count 1, maximum count 1, class fibo-fbc-pas-caa:Account.
  • pwm:orderPlacedAt: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • A limit order states its limit price.

ExecutionShape

applies to execution
  • pwm:executesOrder: minimum count 1, maximum count 1, class pwm:Order.
  • pwm:executedQuantity: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:executionPrice: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:executedAt: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • pwm:executionVenue: minimum count 1, maximum count 1, datatype xsd:string.
  • A fill cannot exceed the order quantity.
  • Allocations of one execution cannot add up to more than was executed.
  • A fill cannot happen before the order was placed.

AllocationShape

applies to allocation
  • pwm:allocatesExecution: minimum count 1, maximum count 1, class pwm:Execution.
  • pwm:allocatedToAccount: minimum count 1, maximum count 1, class fibo-fbc-pas-caa:Account.
  • pwm:allocatedQuantity: minimum count 1, maximum count 1, datatype xsd:decimal.

BestExecutionRecordShape

applies to best execution record
  • pwm:bestExecutionOf: minimum count 1, maximum count 1, class pwm:Execution.
  • pwm:executionFactor: minimum count 1.
  • pwm:venueRationale: minimum count 1, maximum count 1, datatype xsd:string.
  • pwm:executionPolicyReference: minimum count 1, maximum count 1, datatype xsd:string.

CorporateActionShape

applies to corporate action
  • pwm:corporateActionKind: minimum count 1, maximum count 1.
  • pwm:affectsInstrument: minimum count 1, maximum count 1, class fibo-fbc-fi-fi:FinancialInstrument.
  • pwm:effectiveAt: minimum count 1, maximum count 1, datatype xsd:dateTime.

IncomeReceiptShape

applies to income receipt
  • pwm:incomeKind: minimum count 1, maximum count 1.
  • pwm:incomeToAccount: minimum count 1, maximum count 1, class fibo-fbc-pas-caa:Account.
  • pwm:amount: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:currency: minimum count 1, maximum count 1.

TaxReportingRecordShape

applies to tax reporting record
  • pwm:reportingRegime: minimum count 1, maximum count 1.
  • pwm:reportingYear: minimum count 1, maximum count 1, datatype xsd:integer.
  • pwm:reportedAccount: minimum count 1, maximum count 1, class fibo-fbc-pas-caa:Account.
  • pwm:reportedToJurisdiction: an ISO 3166-1 alpha-2 code.
  • pwm:amount: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:currency: minimum count 1, maximum count 1.
  • pwm:reportSubmittedAt: minimum count 1, maximum count 1, datatype xsd:dateTime.

LendingFacilityShape

applies to lending facility
  • pwm:facilityKind: minimum count 1, maximum count 1.
  • pwm:facilityClient: minimum count 1, maximum count 1, class pwm:Client.
  • pwm:facilityLimit: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:currency: minimum count 1, maximum count 1.
  • pwm:maxLoanToValue: between 0 and 1.

PerformanceFigureShape

applies to performance figure
  • pwm:performanceOf: minimum count 1, maximum count 1, class pwm:Portfolio.
  • pwm:performanceMeasure: minimum count 1, maximum count 1.
  • pwm:performanceValue: minimum count 1, maximum count 1, datatype xsd:decimal.
  • pwm:periodStart: minimum count 1, maximum count 1, datatype xsd:dateTime, before periodEnd.
  • pwm:periodEnd: minimum count 1, maximum count 1, datatype xsd:dateTime.
  • A performance figure cites a registered calculation service; PWM never computes it (Kognita 0.6 item 6).

Code lists

applies to every coded property
  • Each of the 101 coded properties takes its values only from the code list it is bound to with pwm:valuesFrom. The shapes, named pwmsh:CodeList_ followed by the property name, are generated from the bindings. List replacement requires firm shapes and review of handwritten constraints and code-dependent rules; see https://github.com/mze3e/kognita/blob/main/docs/ontology/pwm/README.md.

For institutions

Extending PWM

Keep a firm's detail in its own ontology that imports PWM by its version IRI, which resolves to the Turtle file. Specialise PWM classes, add the firm's code lists and bind coded properties to them with pwm:valuesFrom. Keep confidential content out of public repositories.

@prefix owl:  <http://www.w3.org/2002/07/owl#> .
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
@prefix skos: <http://www.w3.org/2004/02/skos/core#> .
@prefix sh:   <http://www.w3.org/ns/shacl#> .
@prefix pwm:  <https://ahmedmuzammil.com/ontology/pwm#> .
@prefix pwmsh: <https://ahmedmuzammil.com/ontology/pwm/shapes#> .
@prefix firm: <https://example.org/firm/ontology#> .

<https://example.org/firm/ontology> a owl:Ontology ;
    owl:imports <https://ahmedmuzammil.com/ontology/pwm/0.10> .

# The firm's own risk scale, in place of the illustrative one
firm:RiskScale a skos:ConceptScheme ; skos:prefLabel "Firm risk scale"@en .
firm:R1 a skos:Concept ; skos:inScheme firm:RiskScale ; skos:notation "R1" ;
    skos:prefLabel "capital preservation"@en .

# Additional bindings do not remove the defaults; see the separate shapes below
pwm:riskProfileLevel pwm:valuesFrom firm:RiskScale .
pwm:productRiskLevel pwm:valuesFrom firm:RiskScale .
pwm:targetRiskProfile pwm:valuesFrom firm:RiskScale .

# A firm product type specializing a PWM class
firm:StructuredNote a owl:Class ;
    rdfs:subClassOf pwm:Product ;
    rdfs:label "structured note"@en .

Sources

Reference register

MarkReferenceJurisdictionStatus
ONTFIBOFinancial Industry Business Ontology (FIBO)EDM Council and Object Management Group, 2026-10-27Every FIBO IRI and maturity used here was rechecked on 3 November 2026 against raw .rdf files from master at GitHub commit 33a92cd34285d532c158da97bb42add089f135d3 (27 October 2026). Each term's maturity level, Release or Provisional, is taken from the file that defines it.Globalverified
ONTOMG CommonsCommons Ontology LibraryObject Management Group, 2025-08-01Collections/hasMember was verified against https://www.omg.org/spec/Commons/20250801/Collections.rdf on 3 November 2026. PartiesAndSituations and Organizations were read at version 20250801. Jurisdiction, RegulatoryAgency, Authorization, isPlayedBy and Document are referenced by FIBO production ontologies; their module files were not opened.Globalverified
ONTPROV-OPROV-O: The PROV OntologyW3C, 2013-04-30Section 3.1 Activity and Agent definitions verified against the W3C Recommendation on 3 November 2026 for Run and SystemTrigger. Other PROV-O terms retain their existing instrument-level grounding.Globalverified
ONTSKOSSKOS Simple Knowledge Organization System ReferenceW3C, 2009-08-18Globalinstrument-level
ONTDCMI TermsDCMI Metadata TermsDublin Core Metadata InitiativeGlobalinstrument-level
ONTDPVData Privacy Vocabulary (DPV) 2.3, with the AI extensionW3C Data Privacy Vocabularies and Controls Community GroupConcept IRIs checked against the DPV 2.3 Turtle files (dpv and ai) in the w3c-cg/dpv repository.Globalverified
ISOISO 17442ISO 17442-1 Financial services: Legal entity identifier (LEI)ISOGlobalinstrument-level
ISOISO 6166ISO 6166 Securities: International securities identification number (ISIN)ISOGlobalinstrument-level
ISOISO 10962ISO 10962 Securities: Classification of financial instruments (CFI)ISOGlobalinstrument-level
ISOISO 4217ISO 4217 Codes for the representation of currenciesISOGlobalinstrument-level
ISOISO 3166ISO 3166-1 Codes for the representation of names of countriesISOGlobalinstrument-level
ISOISO 20022ISO 20022 Universal financial industry message schemeISOGlobalinstrument-level
ISOISO 22222ISO 22222:2005 Personal financial planning: Requirements for personal financial plannersISO, 2005Listed as current; confirmed at review in 2023.Globalverified
ISOISO/IEC 22989ISO/IEC 22989:2022 Artificial intelligence concepts and terminologyISO/IECCited by the MAS AI Risk Management Guidelines (footnote 27) for the AI life cycle.Globalinstrument-level
ISOISO/IEC 42001ISO/IEC 42001:2023 Artificial intelligence management systemISO/IECGlobalinstrument-level
ISOISO 15489ISO 15489-1:2016 Records management: Concepts and principlesISOGlobalinstrument-level
REGFATF RecommendationsInternational Standards on Combating Money Laundering and the Financing of Terrorism and Proliferation (the FATF Recommendations)Financial Action Task Force, 2012, as updatedRecommendation 12 (politically exposed persons, including source of wealth and source of funds) was checked; other Recommendation numbers are cited from the published text.Globalinstrument-level
REGOECD CRSStandard for Automatic Exchange of Financial Account Information in Tax Matters: Common Reporting StandardOECDControlling Persons is defined in Section VIII(D)(6).Globalverified
REGIMDA agentic AI frameworkModel AI Governance Framework for Agentic AIInfocomm Media Development Authority (Singapore)Referenced by the MAS AI Risk Management Guidelines, footnote 11.SGinstrument-level
REGMAS AIRGGuidelines on Artificial Intelligence Risk ManagementMonetary Authority of Singapore, 2026-10-07Read in full from the MAS PDF. Effective 7 October 2027: Sections 3 and 4 from that date; Sections 5 and 6 by 7 October 2028 (para 1.8). MAS intends to consult in 2027 on further guidance for agentic AI.SGverified
REGMAS FEATPrinciples to Promote Fairness, Ethics, Accountability and Transparency (FEAT) in the Use of AI and Data Analytics in Singapore's Financial SectorMonetary Authority of Singapore, 2018SGverified
REGMAS FAA-N16Notice FAA-N16 Recommendations on Investment ProductsMonetary Authority of SingaporeSGverified
REGMAS SFA 04-N12Notice SFA 04-N12 Sale of Investment ProductsMonetary Authority of SingaporeSGverified
REGMAS Notice 626Notice 626 Prevention of Money Laundering and Countering the Financing of Terrorism: BanksMonetary Authority of SingaporeSGverified
REGSecurities and Futures Act (SG)Securities and Futures Act 2001, section 4A (accredited, expert and institutional investors), with the opt-in regime for accredited investorsSingapore Parliament; Monetary Authority of SingaporeSGverified
REGBanking Act (SG)Banking Act 1970, section 47 (banking secrecy) and the Third ScheduleSingapore ParliamentSGinstrument-level
REGMAS Fair Dealing GuidelinesGuidelines on Fair Dealing: Board and Senior Management Responsibilities for Delivering Fair Dealing Outcomes to CustomersMonetary Authority of SingaporeCited by the MAS AI Risk Management Guidelines, footnote 15.SGinstrument-level
REGMiFID IIDirective 2014/65/EU on markets in financial instruments (MiFID II)European Parliament and Council, 2014-05-15EUinstrument-level
REGMiFID II Delegated RegulationCommission Delegated Regulation (EU) 2017/565 supplementing MiFID II (organizational requirements and operating conditions)European CommissionEUinstrument-level
REGPRIIPs RegulationRegulation (EU) No 1286/2014 on key information documents for packaged retail and insurance-based investment products (PRIIPs)European Parliament and CouncilEUinstrument-level
REGEU AI ActRegulation (EU) 2024/1689 laying down harmonized rules on artificial intelligence (AI Act)European Parliament and CouncilEUinstrument-level
REGFCA COBSFCA Handbook, Conduct of Business Sourcebook (COBS 3 client categorization; COBS 9A suitability; COBS 10A appropriateness)Financial Conduct AuthorityGBinstrument-level
REGFCA FG21/1FG21/1 Guidance for firms on the fair treatment of vulnerable customersFinancial Conduct Authority, 2021GBinstrument-level
REGSFC Code of ConductCode of Conduct for Persons Licensed by or Registered with the Securities and Futures Commission (para 5.2 suitability; para 15 professional investors)Securities and Futures CommissionParagraph 5.2 (suitability) checked. Paragraph 15 cited from the published Code.HKverified
REGHK Professional Investor RulesSecurities and Futures (Professional Investor) Rules (Cap. 571D)Hong Kong SAR GovernmentHKinstrument-level
REGFinSAFederal Act on Financial Services (FinSA), SR 950.1Swiss ConfederationArticle 4 (client segmentation) and Article 5 (opting out and opting in) checked. Articles 10 to 14 (suitability and appropriateness) cited as a range.CHverified
REGSEC Regulation DSEC Regulation D, Rule 501(a) definition of accredited investor (17 CFR 230.501)US Securities and Exchange CommissionUSinstrument-level
REGSEC Regulation Best InterestSEC Regulation Best Interest (17 CFR 240.15l-1)US Securities and Exchange CommissionUSinstrument-level
REGMiFID sustainability preferencesCommission Delegated Regulation (EU) 2021/1253 amending Delegated Regulation (EU) 2017/565European Commission, 2021-04-21Verified against EUR-Lex CELEX 32021R1253, EN/ALL text and relationship table on 3 November 2026. Article 1(1) adds Article 2(7)(a)-(c) to 2017/565. Article 1(6)(a)-(f) amends Article 54(2)(a), (5), (9), (10), (12), (13). These are articles of the amended 2017/565, not Articles 2(7) and 54 of 2021/1253. No preference is an explicit PWM recording code, not a fourth statutory category.EUverified
KOGKognita roadmapKognita Roadmap: From Governance Engine to Agent Harness (docs/ROADMAP.md)Kognita (open source, MIT), 2026-09, at release 0.7.0Read at software release 0.7.0, commit 9303769, not roadmap milestone 0.7. Locators such as 0.4 item 8 refer to roadmap sections as recorded at that software release. Milestone 0.5 items 1 to 16 and its Definition of Done were also re-read in the local docs/ROADMAP.md on 3 November 2026 for ontology 0.5; the original pinned software-release reference is retained.Globalverified
KOGKognita regulations libraryAI Control Frameworks: Wealth Management and Agentic AI Coverage (docs/control-frameworks.md), the Kognita regulations libraryKognita (open source, MIT), 2026-10-12Consolidates the 40-domain bank-grade framework, the 70-control wealth management framework (MAS, PDPC, SFC, FCA, Bank of England, APRA, ASIC, CSBS, BIS/FSI, ESMA), a paragraph map of the MAS AI Risk Management Guidelines, and the trust-architecture principles.Globalverified
KOGKognita vocabulariesKognita controlled vocabularies (src/kognita/vocabulary.py, usecases.py, claims.py, memory.py, origination.py)Kognita (open source, MIT), release 0.7.0Globalverified
KOGKognita private-banking packKognita private-banking policy pack 1.0.0 (src/kognita/packs/data/private-banking)Kognita (open source, MIT)Globalverified
AMEVOLVEEVOLVE: Building the AI-Native Enterprise (organizational maturity model for AI)Ahmed MuzammilGlobalverified
AMGoverning agentic AI (essay)Governing agentic AI inside a regulated bankAhmed Muzammil, 2026-09-21Globalverified
AMFiling Cabinet talkYour AI Is Only as Good as Your Filing Cabinet (AIIM ASEAN Forum, Singapore)Ahmed Muzammil, 2026-10-22Checked https://www.ahmedmuzammil.com/sitemap.xml and the speaking page on 3 November 2026. No talk-specific page exists in the sitemap or speaking-page links; the general speaking page is retained.Globalverified
NEWPWM practitioner definitionPractitioner definition in this ontologyAhmed MuzammilUsed where no term was found in FIBO, OMG Commons or DPV at the versions checked. Each use is a candidate for proposal to the relevant standard.Globalverified

0.1 to 1.0

Roadmap

PWM versions generally follow the Kognita milestone with the same number; versions 0.8 and 0.9 are not tied to Kognita milestones. 1.0 is due before Sections 5 and 6 of the MAS AI risk guidelines apply on 7 October 2028. It must meet four tests. Every Kognita control and MAS paragraph is mapped or explicitly marked as boundary or out of scope with a reason. A client interaction can be reconstructed with published queries. Kognita and an institution extension both use PWM. The IRIs are frozen.

VersionStatusTheme
0.1DoneFirst public draft
0.2DoneVerified FIBO grounding and sustainability preferences
0.3DoneRuns, budgets, approvals, gateway calls, pinned evidence and reconstruction
0.4DoneIngestion, policy artifacts, review, circuit breaker and outcomes
0.5DoneAgent identity, scoped authority, autonomy, blast radius and fleets
0.6Built in 0.8Claims: provenance envelope, verification, memory, evaluation, definitions
0.7Built in 0.8Trust and resilience: signed evidence, provider register, dependencies, drills
0.8DoneSupervisory alignment and runtime assurance, covering the 0.6 and 0.7 scope
0.9Built in 0.10Business scope: verified sources, costs and charges, product governance, operations
0.10DraftSupervisory readiness and binding to Kognita
1.0Before 7 Oct 2028Stable, after release candidates and external review

Full roadmap with gates, versioning policy and risks

Change log

  • 0.10 (2026-11-19): Built the roadmap releases 0.9 (business scope) and 0.10 (supervisory readiness and binding to Kognita) as one release. New module 13: costs and charges (CostDisclosure, CostItem, InducementDisclosure), product governance (ProductApproval, TargetMarketReview, negative target market), client circumstances (ComplaintOutcome, SuitabilityReview, structured vulnerability drivers), operations (Order, Execution, Allocation, BestExecutionRecord, CorporateAction, IncomeReceipt, TaxReportingRecord for CRS and FATCA, LendingFacility with LombardFacility as one kind) and PerformanceFigure, which records provenance and never computes. Module 12 additions: EvaluationGate with an activation check, the eight hostile-agent drill scenarios as codes, inventory attributes hasDocumentation and hasComponent. Artifacts: an AI inventory profile (MAS AIRG 4.7 and footnote 25), the ten reconstruction questions as SPARQL queries plus an inventory report and a superseded-source query, a mapping table from all 114 Kognita vocabulary values and 25 policy-pack policies to PWM IRIs, a JSON-LD context and a round trip that reads a signed Kognita evidence export as PWM RDF and checks the chain and the shapes, five jurisdiction profiles (SG, HK, CH, EU, UK) as data, a worked code-list profile, and an AIRG paragraph map that maps every paragraph row in the MAS table of the control frameworks to PWM terms or to boundary. Datatype migration: nextReviewDate and retentionUntil are xsd:dateTime and freshnessThreshold is replaced by freshnessThresholdSeconds, so every datatype is in the OWL 2 datatype map; a review time is now an instant, and a date-only value from earlier versions reads as the end of that day. The 0.8 deprecated aliases are removed. Not done: verification of the 24 instrument-level citations (each is marked in SPEC), FIBO alignment for the new operations terms, the OWL DL profile and HermiT runs (need ROBOT), and a JSON-LD context for register exports (evidence exports only).
  • 0.8 (2026-11-18): Brought PWM level with Kognita software 0.8, built as one release that covers milestones 0.6 to 0.8: milestone 0.7 trust and resilience (signing keys, rotation and revocation, anchors, recovery bundles, external reviews, the model-provider register, failover, model-change detection, the dependency map, concentration measures, failure drills, retention by record class, incident reports), milestone 0.8 runtime assurance (assurance policy, agent manifest, external-agent assertions, deployment attestations, update approvals, approved components and advisories, technology transfer classification, production outcomes, supervisory conformance reports) and the milestone 0.6 records the 0.5 review found missing (verifier independence, evaluation evidence, fairness reports, reasoning traces, claim envelope fields, memory governance, learning exports and exclusions, sensitive attribute inferences). New module 12. Every Kognita vocabulary enum now has a PWM code list whose notations equal the enum values, checked by tests/test_kognita_vocabulary.py: 15 missing evidence event types and the ActorType, ApprovalStatus, EgressDecision and CheckResult lists added, and every one of the 320 concepts has a notation. Renamed to US English with deprecated aliases kept for one version: AuthorisedSignatory, BehaviourAnomaly, BehaviourBaseline, BookingCentre, bookingCentreCode, CommContextualisedInsight, InternationalOrganisationPEP, UnusualAgentBehaviour and VerbAnalyse; PossibleProviderCapability renamed PossibleBankCapability to match Kognita. Ranges added to ten properties that had none. Reasoning axioms added: disjoint mandate types and disjoint agent, model, use case, system and provider, AI agent disjoint from person, and 14 functional properties. Every class and property now has example data or a documented exception, enforced by tests/test_example_coverage.py. Kognita itself moved to US spelling for its identifiers, values and modules, so the verb, drill, reason-code and attribute notations changed with it (analyze, license, unauthorized-network, C1-LABELED, booking_center). Source pinned to Kognita 0.9.0.
  • 0.5 (2026-11-04): Merged the 0.4 live authority observations and evidence-chain build check. Corrected delegation rank direction, typed review expiry, resolved held approvals before allowed actions, explicit field scope and PROV prefixes; refined constraint messages. Aligned to roadmap milestone 0.5: workload identity, explicit scoped and expiring human authority, autonomy evidence and step-down, blast-radius caps, content trust, instruction assurance, anomaly records, narrowing delegation chains, capability grants, fleets, sandbox output, reversibility, adversarial tests, enforcement attestations and stopping in-flight calls. Existing governance terms are reused. Configuration/grants are SlowContext; execution evidence and sandbox decision output are MediumContext. AuthorityLiveStatus uses FastContext for delegation revocation and kill-switch state; firm limits and promotion gates require dated approval by a risk committee, data governance committee or chief risk officer. Authority checks bind completed approval to the governing policy, use case and run, compare call classification by rank, cover every tool verb and require live entitlement scope coverage. Review fixes reject duplicate credential references on distinct workload identities, apply maximum autonomy as fallback for irreversible approval, and clarify that grant checks apply only when a call relies on a grant. Review of 4 Nov 2026: removed the duplicate liveStatusSubject, liveStatusValue and observedAt fields, keeping the canonical statusOf, revoked or engaged, asOf and observedBy; a delegation's permitted data class is now a checked ceiling (DelegationCeilingShape); an agent's effective autonomy cannot exceed its maximum (AgentAutonomyShape).
  • 0.4 (2026-11-03): Context review: durable execution evidence, snapshots and checkpoints use MediumContext; DelegatedAuthority and KillSwitch use SlowContext; FastContext is reserved for live financial state. Aligned to roadmap milestone 0.4: structured ingestion provenance, versioned declarative policy and scenarios, policy packs, recorder replay, RM edit hashes, review queues and coverage, scoped breaker incidents, affected-client results and outcome classifications. Existing lifecycle and register terms are reused; removed graph item 2 adds nothing.
  • 0.3 (2026-11-03): Aligned to roadmap milestone 0.3: run budgets, durable approval continuation, gateway model and MCP calls, classifier evidence, pinned checks and retention, reconstruction coverage, failure mode and authenticated client identity bindings. Review fixes: ProxyCall uses the pre-drafting PolicyDecisionModelAllow for its portfolio read, preserving the evidence chain. retentionDuration uses xsd:nonNegativeInteger in whole days; legacy date and duration ranges remain unchanged with an explicit datatype-map limitation and migration open item. RetentionPolicy uses SlowContext/C1; AuthenticatedClientConfiguration and SystemTrigger use SlowContext/C2. Governed calls require a held decision in the same run with an ApprovalRequest at a checkpoint in the call's run resolved as ApprovalApproved; pending, denied, missing-request and cross-run cases, including decisions naming two runs with approval only in the other run, violate GovernedCallShape (KOG roadmap 0.3 item 2; MAS AIRG 5.9).
  • 0.2 (2026-11-03): Reverified FIBO grounding and maturity; added EU sustainability preferences, code list, two written rules and fixtures; versioned shapes IRI; clarified software release 0.7.0 provenance and repaired the control framework link.
  • 0.1 (2026-11-02): First public draft: 10 modules, reference register, SHACL shapes, example data and negative tests.

The controls this ontology supports are built in Kognita. Comments and corrections are welcome on GitHub .