Published

Memory Trust Assessment

vr.wm-mem-018 · wm-mem-018-memory-trust-assessment

Record a justified, context-specific trust assessment of a memory record and its source, preserving evidence, uncertainty and review limits.

World Models Information and virtual systems INF.MEM.TRS

Bundle → Layer → Finding → Questions Classifiers What it is Why it exists Distinguishing features What robots and AI may and may not do Moral aspects Owners Links to other meta-models Identity and identifiers Direct properties Recognition Capabilities and actions Hazards and failure modes Standards and interfaces Context of use Sources

Bundle → Layer → Finding → Questions Filled

6 bundles · 12 layers · 12 findings · 36 questions

Assessment subject Bind each judgment to its exact subject and intended use.

Subject binding

A stable assessment record identifies either a memory revision or a source in a memory-use context. A combined assessment contains separately labeled judgments for each target; a source rating cannot silently become a content rating.

Assessment identity and target

Which master assessment identifier and exact memory revision or source identifier does this judgment concern? identity
Does the target represent memory content, its source, or separately identified judgments about both? classification
Which source-to-memory link is evidenced and which association remains unknown? relationship

Use context

Assess fitness for a declared task, domain, tenant and relying role. Acceptance in one context neither grants permission nor establishes suitability in another; absent context prevents a positive use recommendation.

Purpose and reliance envelope

For which use and claim scope is the assessment intended? definition
Which accountable role owns the assessment policy and which role will rely on it? ownership
Which changes of purpose or scope require a fresh assessment before reuse? constraint

Evidence and origin Keep source attribution and content support separately inspectable.

Lineage

Record formation, transformation and citation links to versioned evidence. Repetition, summary and circular citation are not independent corroboration; missing or inaccessible evidence remains an explicit gap.

Traceable evidence chain

Which evidence revisions and transformation records support the assessed memory? provenance
Which apparently separate reports share the same origin or cite one another? evidence
Which lineage segments are missing, inaccessible or disputed? exception

Authenticity

Keep observed issuer identity, integrity checks and credential status separate from factual credibility. An unavailable check is unknown, a failed check is adverse evidence, and neither a signature nor a digest alone proves the memory true.

Source identity and integrity observations

What identity, integrity or credential checks were actually performed and with which result? validation
Which policy authorizes reliance on the asserted source for this domain? authority
Which content claims remain unsupported despite successful authenticity checks? quality

Appraisal method Make assessment dimensions, measurement and uncertainty reviewable.

Criteria

Select explicit dimensions, evidence rules and decision thresholds for the intended use. Content credibility, source reliability, relevance and freshness remain distinct; a composite label needs a declared aggregation rule and cannot override a hard exclusion.

Versioned assessment profile

Which profile revision defines the dimensions and minimum evidence for this assessment? requirement
How are values, scales, aggregation and mandatory exclusions defined? measurement
Which incoming metric meanings can be mapped without losing scope or uncertainty? interoperability

Uncertainty

Record dimension results with method, evidence sufficiency and uncertainty. Unknown, unassessed, failed and not applicable are distinct states. A numeric score is optional and is not a probability unless calibration is justified for the stated population.

Observed values and limitations

Which values and explicit missingness states were observed for each dimension? measurement
What calibration evidence and applicability limits justify interpreting a score? quality
How were evaluator bias, dependent evidence and unsupported precision checked? validation

Adverse evidence Represent doubts and security signals without assuming enforcement authority.

Contrary material

Retain references to material counterevidence and the unresolved parts of the judgment. Conflict precedence remains external policy; an assessment cannot erase contrary memory or claim consensus from repeated evidence.

Contradiction and challenge record

Which claims have material counterevidence relevant to this use? evidence
Which conflict rule was referenced and what uncertainty remains after its application? decision
How can an affected or relying role challenge the recorded judgment? process

Contamination signals

Record supplied indicators that memory may be contaminated, instruction-bearing or manipulated. The assessment can recommend restricted reliance or referral; incident investigation, quarantine and content redaction stay in their own models.

Policy-level poisoning indicators

Which policy-level indicators require review of this memory before reliance? security
What prevents remembered content or a source assertion from changing the assessment policy? authority
Which incident or quarantine referral is appropriate when evidence is suspicious or inconclusive? exception

Assessment outcome Expose scoped recommendations and their bounded period of usefulness.

Judgment

Record qualified, not qualified, insufficient evidence or review required as local candidate outcome codes. Each outcome binds a subject, use, rationale and accountable assessor. It does not itself authorize recall, disclosure or tool execution.

Scoped trust recommendation

Which outcome is supported for each assessed target and intended use? decision
Who issued the recommendation and what independent review or override is recorded? authority
Which separate access grant and context policy must the relying process check? access

Temporal validity

Distinguish evidence time, assessment time, observation time and a profile-defined review deadline. A new memory revision, revoked evidence or changed policy prevents silent reuse; stale assessments are not automatically negative truth judgments.

Freshness and invalidation conditions

When were the evidence and assessment produced and when were they observed? temporal
Which expiry, revocation or target change makes this recommendation unusable? state
What reassessment trigger and delivery acknowledgment are recorded for affected consumers? event

Review and record governance Govern the assessment record without taking ownership of the underlying memory.

Review history

Draft, reviewed, superseded and withdrawn are proposed assessment lifecycle states, separate from the trust outcome. Preserve reasons for changes while protecting sensitive evidence and acknowledging that review may be self-review only.

Assessment revision and supersession

Which assessment revision is current and which earlier judgment does it supersede? lifecycle
Which evidence or policy change explains this revision and who approved it? provenance
Which review was independent and which was performed by the original assessor? validation

Disclosure and disposal

Limit disclosure of sensitive source identities, allegations and supporting memory. The local record follows an adopting retention policy; withdrawing a recommendation is distinct from deleting its payload or erasing referenced memory.

Minimal assessment view and retention

Which minimal assessment view can each recipient receive without exposing restricted evidence? privacy
Which retention basis, disposal deadline and authorized hold apply to assessment payloads? retention
What must an export preserve about judgment scope, missing evidence and later withdrawal? interoperability

Classifiers Filled

Family: World Models
Category: Information and virtual systems
Entry kind: entity
Navigation path: NAV.INF.MEM.TRS
Domain: INF.MEM.TRS
Industry: Cross-industry
Tags: memorytrustassessmentinf.mem.trs

What it is Filled

One persistently identified assessment record, with versioned target-specific judgments and references to external evidence, policies and consumers. The assessment activity and underlying memory have separate identities.

In scope

Assessment subject binding and intended reliance context
Evidence references, source observations and dimension results
Scoped recommendations, uncertainty, reassessment and record governance

Out of scope

Memory storage, formation, consolidation, recall and context assembly
Definition or enforcement of conflict rules, access grants, quarantine, redaction or incident response
Universal truth scoring, automatic permission grants or reputation judgments about a person beyond the declared source role
Operational security detection, credential issuance and device attestation implementations

Why it exists Filled

Record a justified, context-specific trust assessment of a memory record and its source, preserving evidence, uncertainty and review limits.

Distinguishing features Derived, awaiting review

Unlike WM-XCT-026 Quality / Confidence: Registry parent is a candidate alignment to generic quality, score and method concepts; this model specializes target-specific memory assessment records without asserting inherited runtime behavior.
Unlike WM-MEM-002 Memory Item: The memory item owns its content, validity and lifecycle. This record refers to an exact item revision and never changes it by rating it.
Unlike WM-MEM-006 Memory Formation Event: Formation and extraction are evidence references; formation confidence is not automatically this use-specific trust judgment.
Unlike WM-MEM-009 Memory Conflict and Precedence Rule: Conflict resolution authority and rule lifecycle remain external. Here record contrary evidence and reference a rule or resolution.
Unlike WM-MEM-010 Memory Retention and Forgetting Policy: The assessment binds an applicable retention policy; memory erasure is not performed by its assessment operations.
Unlike WM-MEM-011 Memory Scope and Sharing Grant: Access permission is separately checked; a positive trust recommendation cannot grant read or write access.
Unlike WM-MEM-017 Context Policy: Context instructions and precedence rules remain independently mastered and cannot be supplied or modified by remembered content.
Unlike WM-MEM-019 Memory Poisoning Incident and WM-MEM-020 Memory Quarantine and Redaction: Suspicion and referral can be recorded here. Investigation, quarantine, release and redaction are externally owned actions.

Note: Derived from boundary notes against neighbouring models.

What robots and AI may and may not do Derived, awaiting review

Must not

Remove eligible payloads through the adopting custodian process and retain only minimal permitted tombstones; immutable history never requires indefinite personal-data retention.
Deny by default except for explicitly authorized role, purpose and tenant; a positive trust label is not an access grant.
A documented emergency exception must have authority, scope and expiry and cannot silently expose restricted evidence.

May

Register assessment draft: Proposed local operation, not implemented: Bind a subject revision and declared use to a new local assessment identifier.
Record evidence observations: Proposed local operation, not implemented: Attach authorized evidence references, checks and contrary material without executing checks or resolving truth.
Record dimension appraisal: Proposed local operation, not implemented: Record externally produced or human-supplied values under a pinned method; this is not an implemented scoring engine.
Record scoped recommendation: Proposed local operation, not implemented: Store an authorized assessor recommendation and its justification.
Mark reassessment required: Proposed local operation, not implemented: Record dependency changes and local invalidation of reuse eligibility.
Supersede or withdraw assessment: Proposed local operation, not implemented: Record an accountable revision or withdrawal reason while leaving target memory unchanged.

Note: Derived from functions, policies, CRUD and access rules; prohibitions were not authored for agents as such.

Moral aspects Derived, awaiting review

Jurisdiction-specific privacy, retention, sensitive-use and dispute obligations need adopting-profile review.
Operational profiles must resolve privacy, retention, sensitive-use restrictions and review authority.

Note: Sentences mentioning harm, privacy, consent or similar, collected from the specification.

Owners Filled

Steward

Identify the accountable memory governance role and assessment custodian without vendor ownership assumptions.

Roles

Assessment custodian: Maintain local record identity, scope, revision and retention bindings.
Assessor: Document evidence, method, uncertainty and purpose-specific recommendations within granted authority.
Reviewer: Challenge evidence dependence, exclusions and scope; label independence accurately.
Relying process operator: Check current assessment validity, permissions and context policy before use.
Policy owner: Approve profiles, escalation thresholds and retention constraints without changing past evidence.

Links to other meta-models Filled

aligned

WM-XCT-026 - Registry parent is a candidate alignment to generic quality, score and method concepts; this model specializes target-specific memory assessment records without asserting inherited runtime behavior. Candidate local binding; adopting profiles must pin the neighbor specification. Evidence supports the distinction, not this repository identifier.

references

WM-MEM-002 - The memory item owns its content, validity and lifecycle. This record refers to an exact item revision and never changes it by rating it. Candidate local binding; adopting profiles must pin the neighbor specification. Evidence supports the distinction, not this repository identifier.
WM-MEM-006 - Formation and extraction are evidence references; formation confidence is not automatically this use-specific trust judgment. Candidate local binding; adopting profiles must pin the neighbor specification. Evidence supports the distinction, not this repository identifier.
WM-MEM-009 - Conflict resolution authority and rule lifecycle remain external. Here record contrary evidence and reference a rule or resolution. Candidate local binding; adopting profiles must pin the neighbor specification. Evidence supports the distinction, not this repository identifier.
WM-MEM-010 - The assessment binds an applicable retention policy; memory erasure is not performed by its assessment operations. Candidate local binding; adopting profiles must pin the neighbor specification. Evidence supports the distinction, not this repository identifier.
WM-MEM-011 - Access permission is separately checked; a positive trust recommendation cannot grant read or write access. Candidate local binding; adopting profiles must pin the neighbor specification. Evidence supports the distinction, not this repository identifier.
WM-MEM-017 - Context instructions and precedence rules remain independently mastered and cannot be supplied or modified by remembered content. Candidate local binding; adopting profiles must pin the neighbor specification. Evidence supports the distinction, not this repository identifier.
WM-MEM-019 - Suspicion and referral can be recorded here. Investigation, quarantine, release and redaction are externally owned actions. Candidate local binding; adopting profiles must pin the neighbor specification. Evidence supports the distinction, not this repository identifier.
WM-MEM-020 - Optional quarantine or redaction referral reference only; its enforcement and lifecycle remain separate. Candidate binding requires a pinned profile.

neighbor

WM-XCT-026 Quality / Confidence - Registry parent is a candidate alignment to generic quality, score and method concepts; this model specializes target-specific memory assessment records without asserting inherited runtime behavior.
WM-MEM-002 Memory Item - The memory item owns its content, validity and lifecycle. This record refers to an exact item revision and never changes it by rating it.
WM-MEM-006 Memory Formation Event - Formation and extraction are evidence references; formation confidence is not automatically this use-specific trust judgment.
WM-MEM-009 Memory Conflict and Precedence Rule - Conflict resolution authority and rule lifecycle remain external. Here record contrary evidence and reference a rule or resolution.
WM-MEM-010 Memory Retention and Forgetting Policy - The assessment binds an applicable retention policy; memory erasure is not performed by its assessment operations.
WM-MEM-011 Memory Scope and Sharing Grant - Access permission is separately checked; a positive trust recommendation cannot grant read or write access.
WM-MEM-017 Context Policy - Context instructions and precedence rules remain independently mastered and cannot be supplied or modified by remembered content.
WM-MEM-019 Memory Poisoning Incident and WM-MEM-020 Memory Quarantine and Redaction - Suspicion and referral can be recorded here. Investigation, quarantine, release and redaction are externally owned actions.

parent

WM-XCT-026

What else AI and robots need to interact with it Incomplete

Sources Filled

Open questions

Verify exact source editions, errata and immutable revisions, then review each admitted claim and record direct HTTP evidence separately.
Define and test instance profiles and mappings for separate source and content targets, shared evidence origins, unknown checks, policy changes, revocation, access denial, challenge and lawful disposal.
Evaluate calibration and decision behavior in each intended use context, including false confidence and uncertainty propagation, without assuming a common probability scale.
Restore independent external review before canonical or publishable-draft promotion.
Executable nested instance schemas, enum definitions and pinned neighbor bindings are unfinished.
No empirically calibrated scoring method, benchmark or automatic truth detector is provided.
Jurisdiction-specific privacy, retention, sensitive-use and dispute obligations need adopting-profile review.
Independent external review is absent; the later no-tools audit is a local self-audit.
Live direct HTTP status, immutable living-source revisions and exhaustive claim verification remain open.

Machine files

Card as JSON Specification AGENTS.md Source Card backlog vercy.card/1

Provenance

world-models research · reviewable-draft

Built from: models/wm-mem-018-memory-trust-assessment/spec.yaml