GMEOW Metacognition Module

What This Slice Covers

This slice owns 11 terms and contributes 0 mapping or projection rows. Use it when its terms match the native fact you want to preserve; use the linkage tables to see how those facts leave GMEOW for consumer vocabularies.

Dependencies

Consumers

Local Map

metacognition map

Examples

Dunning Kruger

# SPDX-FileCopyrightText: 2026 Blackcat Informatics® Inc. <paudley@blackcatinformatics.ca>
# SPDX-License-Identifier: CC-BY-4.0
#
# Worked example — Dunning–Kruger as a MODELLED metacognitive miscalibration, the
# second-order companion to the cognition slice's dunning-kruger.ttl. Where that
# example records two coexisting first-order gmeow:KnowledgeProficiency claims at
# divergent depths (Principle 9), this one turns the reflexive lens on Sam's OWN
# self-assessment: a gmeow:MetacognitiveState whose gmeow:metaTarget is Sam's
# self-attributed mastery, carrying gmeow:calibration gmeow:overconfident. The
# numeric Brier-style gap rides gmeow:calibrationError — a SOLVER-LAYER annotation
# (Principle 12), invisible to the reasoner. The calibration is itself a
# vantage-indexed claim (Principle 9): the assessor judges Sam overconfident; GMEOW
# records the judgement, it does not adjudicate a global verdict.
@prefix gmeow: <https://blackcatinformatics.ca/gmeow/> .
@prefix ex:    <https://blackcatinformatics.ca/gmeow/examples/metacognition/> .
@prefix wd:    <http://www.wikidata.org/entity/> .

ex:sam      a gmeow:Person ; gmeow:name "Sam"@en .
ex:assessor a gmeow:Person ; gmeow:name "External Assessor"@en .

# --- First-order: Sam, by Sam's own lights, has MASTERED quantum computing
#     (reusing the cognition slice's gmeow:KnowledgeProficiency by reference).
ex:samQuantumSelf a gmeow:KnowledgeProficiency ;
    gmeow:knowledgeProficiencyAgent   ex:sam ;
    gmeow:knowledgeProficiencySubject wd:Q192995 ;        # quantum computing
    gmeow:knowledgeProficiencyLevel   gmeow:knowledgeMastered ;
    gmeow:knowledgeProficiencyScale   gmeow:scaleKnowledgeDepth ;
    gmeow:accordingTo                 ex:sam .            # self-attributed vantage

# --- Second-order: Sam's metacognitive state reflecting on that self-assessment.
#     metaTarget is the reflexivity edge onto Sam's own first-order proficiency
#     (the same-agent ownership is a SHACL concern, not an OWL axiom); calibration
#     reads the direction of the confidence/accuracy gap.
ex:samSelfRegard a gmeow:MetacognitiveState ;
    gmeow:metaTarget  ex:samQuantumSelf ;
    gmeow:calibration gmeow:overconfident .

# --- The calibration is a vantage-indexed claim (Principle 9): the assessor,
#     observing Sam's self-regard against a knows-about assessment, judges it
#     overconfident and records the Brier-style magnitude as a solver-layer
#     annotation on the claim (computed-not-asserted, Principle 12).
ex:samCalibrationClaim a gmeow:StandpointClaim ;
    gmeow:vantage           ex:assessor ;
    gmeow:observedFeature   ex:samSelfRegard ;
    gmeow:observationMethod gmeow:methodExpertJudgement ;
    gmeow:claimModality     gmeow:probable ;
    gmeow:calibrationError  0.62 .

Known Unknown

# SPDX-FileCopyrightText: 2026 Blackcat Informatics® Inc. <paudley@blackcatinformatics.ca>
# SPDX-License-Identifier: CC-BY-4.0
#
# Worked example — a KNOWN-UNKNOWN that mints an inquiry, with NO new bridge term.
# Lillith recognises the boundary of her own knowledge (gmeow:awareOfNotKnowing),
# the second-order complement of the cognition knowledge spectrum. The recognised
# gap motivates a question by REUSING inquiry's open-domain gmeow:evokes (the
# not-known subject evokes the question) and the inquiry spine (gmeow:seeksToKnow)
# — there is no gmeow:motivates term (Principle 6), and the bridge to the inquiry
# slice is documented routing, not an entailment. This is the trustworthy-agent
# move: surfacing what it does not know as an open inquiry rather than confabulating.
@prefix gmeow: <https://blackcatinformatics.ca/gmeow/> .
@prefix ex:    <https://blackcatinformatics.ca/gmeow/examples/metacognition/> .
@prefix rdfs:  <http://www.w3.org/2000/01/rdf-schema#> .

ex:lillith   a gmeow:Agent .
ex:protocolX a gmeow:Entity ; rdfs:label "protocol X"@en .

# --- Known-unknown: Lillith is aware that she does NOT know protocol X — a
#     recognised gap (second-order), not the open-world silence of an unasserted
#     triple. The boundary of her knowledge, recorded for a trustworthy memory.
ex:lillith gmeow:awareOfNotKnowing ex:protocolX .

# --- The gap evokes a question (reusing inquiry's open-domain gmeow:evokes — the
#     not-known subject sits legally in its subject position, so no new motivates
#     term is minted), which Lillith then holds open and seeks to know.
ex:protocolX gmeow:evokes ex:howDoesProtocolXWork .

ex:howDoesProtocolXWork a gmeow:Question ;
    gmeow:questionType gmeow:typeHow ;
    rdfs:label "how does protocol X work?"@en .

ex:lillith gmeow:seeksToKnow ex:howDoesProtocolXWork .   # the inquiry spine, reused

Reflection Revision

# SPDX-FileCopyrightText: 2026 Blackcat Informatics® Inc. <paudley@blackcatinformatics.ca>
# SPDX-License-Identifier: CC-BY-4.0
#
# Worked example — REFLECTION defeating one's own inference, belief revision as
# SUPPRESSION (Principle 10). The metacognition layer over the inference slice's
# belief-revision pattern: Lillith inductively concluded "the server is down" from
# "the ping timed out". On REFLECTION (a gmeow:Event typed gmeow:eventTypeReflection,
# tied to the reflexive layer by a gmeow:MetacognitiveState whose gmeow:metaTarget
# is her OWN gmeow:InferenceCommitment) she surfaces an UNDERCUTTING defeater — a
# firewall rule blocked the ping, removing the warrant. Revision does NOT delete:
# the conclusion-claim is set gmeow:displayable false, the gmeow:InferenceTenure is
# closed, and the gmeow:InferenceCommitment is retained as audit. The defeater is
# installed via the inference slice's gmeow:hasDefeater (reused by reference); the
# reflection → revision link is a documented bridge, never an axiom.
@prefix gmeow: <https://blackcatinformatics.ca/gmeow/> .
@prefix ex:    <https://blackcatinformatics.ca/gmeow/examples/metacognition/> .
@prefix rdfs:  <http://www.w3.org/2000/01/rdf-schema#> .
@prefix xsd:   <http://www.w3.org/2001/XMLSchema#> .

ex:lillith a gmeow:Agent ; gmeow:name "Lillith"@en .

ex:propPingTimedOut rdfs:label "the ping timed out"@en .
ex:propServerDown   rdfs:label "the server is down"@en .
ex:propFirewall     rdfs:label "a firewall rule blocks ping"@en .

# --- The original premise and conclusion (the prior inference Lillith held).
ex:premPingTimedOut a gmeow:StandpointClaim ;
    gmeow:vantage ex:lillith ;
    gmeow:observedFeature ex:propPingTimedOut ;
    gmeow:observationMethod gmeow:methodDirectObservation ;
    gmeow:claimModality gmeow:unequivocal .

# The conclusion — now SUPPRESSED on reflection, but retained.
ex:conclServerDown a gmeow:StandpointClaim ;
    rdfs:label "Concluded (then retracted): the server is down"@en ;
    gmeow:vantage ex:lillith ;
    gmeow:observedFeature ex:propServerDown ;
    gmeow:observationMethod gmeow:methodExpertJudgement ;
    gmeow:claimModality gmeow:probable ;
    gmeow:inferenceMode gmeow:modeInduction ;
    gmeow:inferredFrom ex:premPingTimedOut ;
    gmeow:displayable false .                     # Principle 10: suppressed by the fired defeater

# --- The reflection act: Lillith reviews her OWN reasoning (the metacognition
#     layer). The event carries the gmeow:eventTypeReflection value; a
#     gmeow:MetacognitiveState ties the review to the reflexive layer, its
#     gmeow:metaTarget pointing at her own inference commitment.
ex:lillithReview a gmeow:Event ;
    gmeow:eventType gmeow:eventTypeReflection .

ex:lillithReflectsOnInf1 a gmeow:MetacognitiveState ;
    gmeow:metaTarget ex:revisedCommitment .

# --- The defeater the reflection surfaces (undercutting: removes the warrant,
#     does not assert the negation).
ex:defeaterFirewall a gmeow:StandpointClaim ;
    rdfs:label "Defeater: a firewall rule blocked the ping"@en ;
    gmeow:vantage ex:lillith ;
    gmeow:observedFeature ex:propFirewall ;
    gmeow:observationMethod gmeow:methodDirectObservation ;
    gmeow:claimModality gmeow:unequivocal ;
    gmeow:defeaterKind gmeow:defeaterUndercutting .

# --- The reified argument, retained as audit; the reflection installs the
#     defeater on Lillith's own commitment via the inference slice's hasDefeater.
ex:revisedCommitment a gmeow:InferenceCommitment ;
    rdfs:label "Inductive argument: ping-timeout therefore server-down (defeated on reflection)"@en ;
    gmeow:inferenceModeOf gmeow:modeInduction ;
    gmeow:premise ex:premPingTimedOut ;
    gmeow:conclusion ex:conclServerDown ;
    gmeow:hasDefeater ex:defeaterFirewall .

# --- The tenure — opened then closed (never deleted, Principle 10).
ex:serverDownTenure a gmeow:InferenceTenure ;
    rdfs:label "The held interval of the 'server is down' conclusion"@en ;
    gmeow:tenureOf ex:revisedCommitment ;
    gmeow:duringInterval ex:tenureInterval .

ex:tenureInterval a gmeow:TimeInterval ;
    gmeow:startedAtTime "2026-06-16T09:00:00Z"^^xsd:dateTime ;
    gmeow:endedAtTime   "2026-06-16T09:12:00Z"^^xsd:dateTime .  # closed when the defeater fired

Terms

Classes

Term Label Definition
gmeow:CalibrationStatus Calibration Status The qualitative fit between an agent's avowed confidence and its actual accuracy — a closed value vocabulary: gmeow:wellCalibrated (confidence tracks accuracy)...
gmeow:MetacognitiveState Metacognitive State An agent's intrinsic SECOND-ORDER mental moment — a state whose subject is one of the agent's OWN first-order mental moments or claims (a belief about a belief...

Properties

Term Label Definition
gmeow:awareOfNotKnowing aware of not knowing A known-unknown: an agent is aware that it does NOT know a subject — the second-order recognition of a gap in its own knowledge, the explicit boundary of what...
gmeow:calibration calibration Assigns a qualitative gmeow:CalibrationStatus (wellCalibrated / overconfident / underconfident) to a gmeow:MetacognitiveState — the direction of the gap betwee...
gmeow:calibrationError calibration error A SOLVER-LAYER numeric measure of miscalibration (Principle 12) — a Brier-style score in the closed interval [0,1] attached to the statement asserting a metaco...
gmeow:epistemicSelfTrust epistemic self-trust The reliability an agent accords to one of its OWN epistemic faculties or sources — feeling-of-knowing: how far the agent trusts its own memory, perception, or...
gmeow:metaTarget meta target The reflexivity edge: relates a gmeow:MetacognitiveState to the first-order mental moment or claim it reflects upon. The RANGE is left intentionally open — the...

Individuals

Term Label Definition
gmeow:eventTypeReflection reflection An event of an agent reviewing its OWN reasoning — re-examining a prior inference, belief, or self-assessment. A value individual of gmeow:EventType (the event...
gmeow:overconfident overconfident Confidence exceeds accuracy: the agent is more sure than it should be — the Dunning–Kruger pole, characteristically overconfidence at low competence. The confa...
gmeow:underconfident underconfident Accuracy exceeds confidence: the agent is less sure than it should be — the impostor pole, where knowledge is held but distrusted. The opposite failure to over...
gmeow:wellCalibrated well calibrated Confidence tracks accuracy: the agent's avowed confidence matches how often it is right — the target calibration of a trustworthy agent. The midpoint of the ov...

Guide

Metacognition

Slice: https://blackcatinformatics.ca/gmeow/slices/metacognition · tier: core

Second-order epistemics — the reflexive turn. The whole epistemic stack is first-order: an agent believes a proposition (epistemics), knows about a subject (cognition), claims a feature from a vantage (standpoint). Metacognition is the turn whose subject is one of the agent's own first-order states or claims, plus calibration — the fit between avowed confidence and accuracy. For a trustworthy AI memory this is decisive: "how sure am I that I'm right, and do I know where my knowledge stops" is exactly what separates a calibrated agent from a confabulating one (Principle 15). The slice mints almost no new machinery — a metacognitive judgement is a gmeow:StandpointClaim whose gmeow:observedFeature is one of the agent's own claims / mental moments and whose gmeow:vantage is that same agent.

Reflexivity by reuse, not new machinery

gmeow:MetacognitiveState (gufo:Kind ⊑ gmeow:MentalMoment) is the endurant second-order mode, and gmeow:metaTarget is the reflexivity edge to the first-order state it reflects upon. It is the co-equal sibling of the first-order mental modes, gathered under gmeow:MentalMoment (kernel) so an agent's whole mental life is one queryable family:

Mode Slice Stands toward
gmeow:CognitiveState cognition a subject the agent knows (objectual, first-order)
gmeow:IntentionalMode teleology a state of affairs the agent aims at (conative, first-order)
doxastic states epistemics a proposition the agent holds (first-order)
gmeow:MetacognitiveState metacognition one of the agent's own first-order states / claims (reflexive, second-order)

These are documented siblings, not subsumed: there is deliberately no rdfs:subClassOf making gmeow:MetacognitiveState a kind of gmeow:CognitiveState. "Thinking about thinking is a kind of thinking" would erase the reflexive-vs-first-order distinction — exactly as gmeow:Question is not a gmeow:Proposition. The siblinghood lives in prose and the tests/test_metacognition.py no-subsumption guard, never in an axiom.

gmeow:MetacognitiveState · gmeow:metaTarget

gmeow:MetacognitiveState types an agent's second-order mental moment, inhering in exactly one agent. gmeow:metaTarget (domain gmeow:MetacognitiveState, open range, non-functional) points at the first-order gmeow:MentalMoment, gmeow:StandpointClaim, or gmeow:InferenceCommitment it reflects upon. The reflexivity is conceptual (the target is the agent's own state), never OWL reflexivity: gmeow:metaTarget is deliberately not owl:ReflexiveProperty (which leaves EL and is semantically false — the metastate and its target are distinct individuals). Same-agent ownership is a SHACL concern, not a reasoned axiom (Principle 12).

Calibration

gmeow:calibration (domain gmeow:MetacognitiveState, range gmeow:CalibrationStatus) assigns the qualitative direction of the confidence/accuracy gap. gmeow:CalibrationStatus is a closed value vocabulary (gufo:AbstractIndividualType ⊑ gufo:QualityValue), its members individuals, never subclasses:

Value Reading
gmeow:wellCalibrated confidence tracks accuracy — the target of a trustworthy agent
gmeow:overconfident confidence exceeds accuracy — the Dunning–Kruger pole, the confabulation risk
gmeow:underconfident accuracy exceeds confidence — the impostor pole

The numeric magnitude of the gap is gmeow:calibrationError — a Brier-style score in [0,1], solver-layer (Principle 12). It is an owl:AnnotationProperty by design, exactly like gmeow:confidence: invisible to the reasoner, computed externally, and never a materialised axiom. Modelling it as an owl:DatatypeProperty with a domain would make it a reasoned ABox edge and risk the reasoner touching it — the guard test_calibration_error_is_a_solver_layer_annotation pins it as annotation-only.

Dunning–Kruger is overconfidence at low competence — the cognition slice's coexisting-claims fixture, now modelled. The dunning-kruger.ttl example turns the reflexive lens on Sam's own self-assessment: a gmeow:MetacognitiveState whose gmeow:metaTarget is Sam's self-attributed gmeow:KnowledgeProficiency, carrying gmeow:calibration gmeow:overconfident, with the assessor's gmeow:calibrationError riding the calibration claim. The calibration is itself a vantage-indexed claim (Principle 9); GMEOW records the judgement, it does not adjudicate a verdict.

gmeow:CalibrationStatus · gmeow:calibration · gmeow:calibrationError

The closed status vocabulary, the property that assigns it (non-functional — a self-assessment and an external assessment coexist), and the solver-layer Brier-style magnitude annotation. The qualitative direction and the numeric gap are kept distinct: gmeow:calibration reasons (an EL range axiom over a closed vocab); gmeow:calibrationError does not (an annotation the reasoner never sees).

Known-unknowns

gmeow:awareOfNotKnowing (domain gmeow:Agent, open range, non-functional) records the boundary of an agent's knowledge — a recognised gap (second-order), not the open-world silence of an unasserted triple. It is the reflexive complement of the cognition knowledge spectrum: where gmeow:isAwareOf … gmeow:hasMastered record what an agent knows, this records a recognised edge of that knowledge.

A known-unknown motivates a question by reference — with no new bridge term. The recognised gap mints a gmeow:Question by reusing inquiry's open-domain gmeow:evokes (the not-known subject sits legally in evokes's subject position) and the inquiry spine (gmeow:seeksToKnow); there is no gmeow:motivates partner (Principle 6), and the bridge is documented routing, not an entailment. The known-unknown.ttl example shows the trustworthy-agent move: surfacing what it does not know as an open inquiry rather than confabulating an answer.

gmeow:awareOfNotKnowing

The known-unknown edge. Flat (no rdfs:subPropertyOf), open range, domain gmeow:Agent — the cheap surface a trustworthy memory reads to surface its blind spots. The inquiry bridge reuses gmeow:evokes; this slice asserts no owl:subPropertyOf into the inquiry slice.

Epistemic self-trust

gmeow:epistemicSelfTrust (domain gmeow:Agent, open range, non-functional) is feeling-of-knowing — the reliability an agent accords to one of its own epistemic faculties or sources (its memory, perception, reasoning). It is the flat 80% surface; its rich, graded form is a gmeow:StandpointClaim whose gmeow:vantage is the agent and whose gmeow:observedFeature is the agent's own faculty, carrying gmeow:trustLevel by reference to the trust slice — self-trust is trust-about-oneself, the gmeow:trustor and gmeow:trustee coinciding. The trust machinery is reused, not re-minted (Principle 6); this slice asserts no triples into the trust slice.

gmeow:epistemicSelfTrust

The self-trust edge. Use it for reliance on one's own faculty; trust in an external agent or source is the trust slice's gmeow:TrustAssertion directly. Promote to the standpoint-claim idiom carrying gmeow:trustLevel when the degree, scale, or vantage of self-trust is itself the fact.

Reflection & belief revision

gmeow:eventTypeReflection (a gmeow:EventType, the gmeow:eventTypeDeception pattern) types an event of an agent reviewing its own reasoning. A reflection may install a gmeow:hasDefeater on one of the agent's own gmeow:InferenceCommitment individuals (the inference slice), triggering belief revision as suppression (Principle 10): the defeated conclusion-claim is set gmeow:displayable false and the prior inference is retained as audit, never deleted. The reflection-revision.ttl example walks it: Lillith reflects on her own abductive inference, surfaces an undercutting defeater (a firewall blocked the ping), installs it via gmeow:hasDefeater, and suppresses the defeated conclusion — the reflection → revision link is a documented bridge, never an axiom.

The boundary with the mentation slice is documented (Principle 4): gmeow:eventTypeReflection types the reviewable occurrent kind; a reflection that unfolds as a reasoning process is a gmeow:MentalProcess — one canonical source per fact, never both for the same individual.

gmeow:eventTypeReflection

The reflection-act value individual. Type a reflection gmeow:Event with it; record revision by installing gmeow:hasDefeater on the agent's own inference and suppressing the defeated conclusion with gmeow:displayable false.

Alignments (mappings/equivalences.ttl)

The slice's alignments are deliberately prose-only: its grounding literatures — Nelson & Narens' metamemory model, Flavell's metacognition, and the Brier-score calibration tradition — have no stable external IRI vocabulary to map to, so the mapping set carries the scholarly grounding in its comment and seeds no gmeow:TermEquivalence rows (the same intentional non-mapping the inquiry slice applies to its question-type vocabulary). The reused first-order quantities and the cross-slice bridges carry their alignments in their home slices.

Dependencies

Slice Why
kernel gmeow:MentalMoment (the mode genus) and gmeow:Agent (the known-unknown / self-trust domain)
epistemics the first-order doxastic states and credence that metacognition assesses — referenced, the reflexive gmeow:metaTarget points at them
standpoint gmeow:StandpointClaim — a metacognitive judgement is a standpoint claim over the agent's own state (gmeow:vantage = the agent)
observations the Observation spine the metacognitive claim rides (gmeow:vantage, gmeow:observedFeature)
cognition gmeow:CognitiveState (the first-order sibling) and gmeow:KnowledgeProficiency (the Dunning–Kruger gmeow:metaTarget)

Bridges to inquiry (gmeow:evokes), inference (gmeow:hasDefeater, gmeow:InferenceCommitment), and trust (gmeow:trustLevel) are reused by reference and documented, not declared as dependencies — the slice asserts no axioms into them (Principle 9).

Verified by construction

tests/test_metacognition.py pins the load-bearing shape of the slice: