# Memory Consolidation

- Name: Memory Consolidation
- Catalogue ID: WM-MEM-008
- Registry ID: vr.wm-mem-008
- Type: Published Vercy event
- Version: 0.1.0-reviewable-draft
- Specification: https://ver.cy/models/wm-mem-008-memory-consolidation/spec.yaml
- Storage type: format-independent; select a binding in the page constructor
- Interface: https://ver.cy/models/wm-mem-008-memory-consolidation/#template-builder
- Processes: https://ver.cy/processes/
- Model agent protocol: https://ver.cy/model-agent-protocol.md
- Research evidence: https://github.com/ver-cy/world-models/tree/feat/mega-model-registry/research/runs/wm-mem-008

Read this file first, then `spec.yaml`. Traverse Bundle to Layer to Finding to
Questions and Artifacts. The specification is published, while its research
assurance remains `reviewable-draft`. Do not claim canonical completeness
while `researchAdjudication.publicationHolds` is non-empty. Preserve source
references, master-system identity and access rules.

## Five-facet gate

Before creating or changing an instance, inspect the specification for explicit
coverage of: identity and class; direct properties; recognition and observation;
capabilities, behaviour and possible actions; context and evidence. Treat each
facet as required, optional, not applicable with a reason, or delegated through
a pinned model reference. Do not invent physical properties for abstract
objects. For physical measurements preserve units, tolerances, method,
conditions, observation time, uncertainty and provenance.

Keep capability, behaviour, state, transition, affordance, operation,
precondition, effect, reversibility, permission, constraint, hazard and failure
mode distinct. Resolve delegated child models and master systems instead of
copying their domains into this model.

## Agent lifecycle route

1. **Fill:** read the nearest owner and Dimension policies, resolve master
   identities, ask only for missing facts or authority, then record facts,
   hypotheses and unknowns distinctly with evidence and observation time.
2. **Extend:** propose a versioned extension when the published bundles and
   layers cannot express a required concept. Preserve stable IDs and namespace
   ownership; do not silently widen the model boundary.
3. **Edit:** validate authority, purpose, current version and concurrency before
   mutation. Preserve history and provenance. Treat access expansion, external
   writes and material side effects as confirmation-gated actions.
4. **Validate:** run schema, semantic, reference, permission, evidence and
   five-facet checks before and after a write. Report unresolved references,
   lossy projections, stale observations and publication holds.
5. **Retire:** deprecate and supersede through the declared lifecycle. Preserve
   immutable history and resolvable references. Require confirmation before
   deletion, destructive migration or irreversible external action.

Autonomous work is limited to policy-authorized reading, local population of
confirmed facts, validation, indexing and non-destructive maintenance. Propose
schema changes and new imports. Require confirmation for publication, broader
access, external writes, deletion and irreversible effects. Follow the model
agent protocol and the nearest local `AGENTS.md` whenever their rules are more
restrictive.

## Local package assurance

This is a local, noncanonical reviewable draft. Codex is the sole active provider. Claude and Grok were skipped with zero attempts under the owner-authorized waiver. The separate no-tools self-audit is not independent external review. Catalogue URLs are distribution targets; this local package does not establish remote availability.

For this model, history preservation means authorized minimal evidence under the pinned retention policy. It does not require preserving erased payload or restricted lineage indefinitely. External owners retain store mutation, grants, index updates and erasure execution. All subject functions are proposed local recording operations, not an implemented consolidation engine.

Publication holds:

- Independent external review is absent under the owner-authorized single-provider waiver. Claude and Grok were skipped with zero attempts; the separate local Codex no-tools self-audit is not an independent second-provider review.
- Live source, version, errata and reuse-term verification remains incomplete. Selected text from eight official sources was read through the web service. Direct HTTP checks were not attempted under the owner-reported sandbox restriction: zero measured HTTP 200 responses, with statuses unmeasured. The DQV dated publication pin remains unresolved; NIST Privacy Framework 1.0 is a historical pinned edition, not a latest-version claim.
- Adopting profiles must pin neighbor versions, processing purpose, grants, retention and erasure authority, release review, atomicity and retry contracts. No universal duplicate threshold, precedence rule, summary acceptance score or jurisdiction-specific legal basis is established.
- Executable nested instance schemas, PROV-O and DQV mappings, store and index adapters, empirical quality evaluation and adversarial acceptance fixtures remain unimplemented. Research schema validation does not certify operational behavior, factual accuracy or compliance.
- Independent external review was explicitly waived by the repository owner; this codex-only result remains a reviewable draft.