MEMORY-LD v0.1.0

Status: Draft Extends: CORE-LD v0.1.0 Constitutional dependencies: URS v0.1.0 → CORE-LD v0.1.0 Spec family: Functional Profile (see CORE-LD §6.1 — Composition Rules) Dimension registry prefix: MEMORY-LD Canonical @context: https://specs.initialcore.net/ns/memory-ld.jsonld

1. Preamble

MEMORY-LD is the functional profile for persistent knowledge organization. It defines how knowledge stores, memories, episodes, retrieval operations, and forgetting are expressed as CORE-LD entities with standardised memory semantics.

What MEMORY-LD is:

• A set of entity types and properties specialised for representing persistent knowledge
• A memory model: what is stored (episodes, facts, concepts), how it is organised (stores, indices, schemas), how it is retrieved (queries, recall), and how it evolves (consolidation, forgetting)
• A standard envelope for any system that maintains structured long-term knowledge

What MEMORY-LD is not:

• Not a database specification — MEMORY-LD defines the representation, not the engine
• Not a vector store format — embeddings are parameters, not entity types
• Not a replacement for CORE-LD — MEMORY-LD inherits all CORE-LD invariants and adds its own

1.1 Composition rule

MEMORY-LD = CORE-LD + MemoryStore + Episode + KnowledgeAtom + Schema + RetrievalSpec + ForgettingPolicy

Any conforming MEMORY-LD implementation MUST also conform to CORE-LD at the same conformance level. An encoded MEMORY-LD binding MUST preserve both CORE-LD and MEMORY-LD invariants.

1.2 Memory as a functional complement

MEMORY-LD is the complement to INPUT-LD in the functional profile family:

INPUT-LD answers "what does the system need to process?" MEMORY-LD answers "what does the system need to remember?"

2. Entity Types

MEMORY-LD extends the CORE-LD kind set with the following entity types:

2.1 MemoryStore

A MemoryStore represents a managed collection of persistent knowledge — the top-level container for memories.

{
  "@context": "https://specs.initialcore.net/ns/memory-ld.jsonld",
  "@id": "_:store1",
  "@type": "MemoryStore",
  "name": "ICoreConstitutionalMemory",
  "storeType": "episodic",
  "schema": "_:schema1",
  "indexes": [ "_:idx1", "_:idx2" ],
  "policy": "_:policy1"
}

Registered store types:

2.2 Episode

An Episode represents a single recorded experience or event — a unit of episodic memory.

{
  "@id": "_:ep1",
  "@type": "Episode",
  "timestamp": "2026-07-30T14:00:00Z",
  "duration": 2.5,
  "summary": "User requested memory profile creation",
  "content": "_:context1",
  "associations": [ "_:entity1", "_:entity2" ],
  "salience": 0.85
}

2.3 KnowledgeAtom

A KnowledgeAtom represents a single atomic fact or proposition — a unit of semantic memory.

{
  "@id": "_:ka1",
  "@type": "KnowledgeAtom",
  "subject": "_:entity1",
  "predicate": "hasProvenance",
  "object": "did:icore:admin",
  "confidence": 0.95,
  "source": "_:ep1"
}

KnowledgeAtoms are composable: multiple atoms about the same subject with the same predicate constitute alternative or corroborating knowledge (distinguished by source and confidence).

2.4 Schema

A Schema defines the valid structure of memories within a MemoryStore — an ontology for memory organization.

{
  "@id": "_:schema1",
  "@type": "Schema",
  "name": "ConstitutionalEntitySchema",
  "entityTypes": [ "Entity", "Identifier", "Provenance" ],
  "requiredProperties": [ "name", "identifier" ],
  "forbiddenProperties": [],
  "validators": [
    {
      "field": "identifier.scope",
      "operator": "in",
      "value": ["local", "constitutional"]
    }
  ]
}

2.5 Index

An Index provides an access structure for efficient retrieval.

{
  "@id": "_:idx1",
  "@type": "Index",
  "name": "byActor",
  "indexType": "hash",
  "field": "actor",
  "unique": false
}

Registered index types:

2.6 RetrievalSpec

A RetrievalSpec describes a memory retrieval operation — what to find, how to rank, what to return.

{
  "@id": "_:ret1",
  "@type": "RetrievalSpec",
  "query": {
    "field": "predicate",
    "operator": "equals",
    "value": "hasProvenance"
  },
  "ranking": "relevance",
  "maxResults": 10,
  "includeAssociations": true
}

QueryExpression:

{
  "field": "...",
  "operator": "equals | contains | matches | exists",
  "value": "..."
}

2.7 ForgettingPolicy

A ForgettingPolicy defines when and how memories are consolidated, archived, or pruned.

{
  "@id": "_:policy1",
  "@type": "ForgettingPolicy",
  "retentionDuration": "P90D",
  "salienceThreshold": 0.3,
  "consolidationStrategy": "summarize",
  "maxEpisodes": 10000
}

3. Memory Lifecycle

A memory progresses through a lifecycle from acquisition through potential forgetting.

Acquired → Indexed → Consolidated → Archived → Forgotten

A memory may also be Recalled (temporarily promoted from Archived to active) without changing its lifecycle state.

4. Invariants

MEMORY-LD inherits all 23 CORE-LD invariants (I1–I23). The following MEMORY-LD-specific invariants are ADDITIONAL:

4.1 Store invariants (I-MEM-1 through I-MEM-4)

4.2 Episode invariants (I-MEM-5 through I-MEM-8)

4.3 KnowledgeAtom invariants (I-MEM-9 through I-MEM-12)

4.4 Retrieval invariants (I-MEM-13 through I-MEM-15)

4.5 Policy invariants (I-MEM-16 through I-MEM-18)

5. MEMORY-LD @context Extension

The MEMORY-LD @context extends the CORE-LD @context with the following term mappings:

{
  "@context": {
    "MemoryStore": "https://specs.initialcore.net/ns/memory#MemoryStore",
    "Episode": "https://specs.initialcore.net/ns/memory#Episode",
    "KnowledgeAtom": "https://specs.initialcore.net/ns/memory#KnowledgeAtom",
    "Schema": "https://specs.initialcore.net/ns/memory#Schema",
    "Index": "https://specs.initialcore.net/ns/memory#Index",
    "RetrievalSpec": "https://specs.initialcore.net/ns/memory#RetrievalSpec",
    "ForgettingPolicy": "https://specs.initialcore.net/ns/memory#ForgettingPolicy",

    "storeType": "https://specs.initialcore.net/ns/memory#storeType",
    "capacity": "https://specs.initialcore.net/ns/memory#capacity",
    "schema": { "@id": "https://specs.initialcore.net/ns/memory#schema", "@type": "@id" },
    "indexes": { "@id": "https://specs.initialcore.net/ns/memory#indexes", "@container": "@set" },
    "policy": { "@id": "https://specs.initialcore.net/ns/memory#policy", "@type": "@id" },
    "episodic": "https://specs.initialcore.net/ns/memory#episodic",
    "semantic": "https://specs.initialcore.net/ns/memory#semantic",
    "procedural": "https://specs.initialcore.net/ns/memory#procedural",
    "associative": "https://specs.initialcore.net/ns/memory#associative",
    "working": "https://specs.initialcore.net/ns/memory#working",

    "timestamp": { "@id": "https://specs.initialcore.net/ns/memory#timestamp", "@type": "xsd:dateTime" },
    "duration": "https://specs.initialcore.net/ns/memory#duration",
    "summary": "https://schema.org/description",
    "content": { "@id": "https://specs.initialcore.net/ns/memory#content", "@type": "@id" },
    "associations": { "@id": "https://specs.initialcore.net/ns/memory#associations", "@container": "@set" },
    "salience": "https://specs.initialcore.net/ns/memory#salience",

    "subject": { "@id": "https://specs.initialcore.net/ns/memory#subject", "@type": "@id" },
    "predicate": "https://specs.initialcore.net/ns/memory#predicate",
    "object": "https://specs.initialcore.net/ns/memory#object",
    "confidence": "https://specs.initialcore.net/ns/memory#confidence",
    "source": { "@id": "https://specs.initialcore.net/ns/memory#source", "@type": "@id" },

    "entityTypes": { "@id": "https://specs.initialcore.net/ns/memory#entityTypes", "@container": "@set" },
    "requiredProperties": { "@id": "https://specs.initialcore.net/ns/memory#requiredProperties", "@container": "@set" },
    "forbiddenProperties": { "@id": "https://specs.initialcore.net/ns/memory#forbiddenProperties", "@container": "@set" },
    "validators": { "@id": "https://specs.initialcore.net/ns/memory#validators", "@container": "@set" },

    "indexType": "https://specs.initialcore.net/ns/memory#indexType",
    "field": "https://specs.initialcore.net/ns/memory#field",
    "unique": "https://specs.initialcore.net/ns/memory#unique",
    "hash": "https://specs.initialcore.net/ns/memory#hash",
    "ordered": "https://specs.initialcore.net/ns/memory#ordered",
    "vector": "https://specs.initialcore.net/ns/memory#vector",
    "graph": "https://specs.initialcore.net/ns/memory#graph",

    "query": "https://specs.initialcore.net/ns/memory#query",
    "ranking": "https://specs.initialcore.net/ns/memory#ranking",
    "maxResults": "https://specs.initialcore.net/ns/memory#maxResults",
    "includeAssociations": "https://specs.initialcore.net/ns/memory#includeAssociations",
    "relevance": "https://specs.initialcore.net/ns/memory#relevance",
    "recency": "https://specs.initialcore.net/ns/memory#recency",

    "retentionDuration": "https://specs.initialcore.net/ns/memory#retentionDuration",
    "salienceThreshold": "https://specs.initialcore.net/ns/memory#salienceThreshold",
    "consolidationStrategy": "https://specs.initialcore.net/ns/memory#consolidationStrategy",
    "summarize": "https://specs.initialcore.net/ns/memory#summarize",
    "prune": "https://specs.initialcore.net/ns/memory#prune",
    "archive": "https://specs.initialcore.net/ns/memory#archive",
    "merge": "https://specs.initialcore.net/ns/memory#merge"
  }
}

6. Composition Examples

6.1 INPUT-LD → MEMORY-LD request

When MEMORY-LD entities are the target of an INPUT-LD Request:

{
  "@context": [
    "https://specs.initialcore.net/ns/core-ld.jsonld",
    "https://specs.initialcore.net/ns/input-ld.jsonld",
    "https://specs.initialcore.net/ns/memory-ld.jsonld"
  ],
  "@id": "_:storeQuery",
  "@type": "Request",
  "intent": "query",
  "target": "_:store1",
  "parameters": [
    {
      "@id": "_:ret1",
      "@type": "RetrievalSpec",
      "query": { "field": "storeType", "operator": "equals", "value": "episodic" },
      "ranking": "recency",
      "maxResults": 50
    }
  ],
  "context": { "actor": "did:icore:user:alice" }
}

6.2 Episodic record with provenance

{
  "@context": [
    "https://specs.initialcore.net/ns/core-ld.jsonld",
    "https://specs.initialcore.net/ns/memory-ld.jsonld"
  ],
  "@id": "_:ep2",
  "@type": "Episode",
  "timestamp": "2026-07-30T15:30:00Z",
  "summary": "Constitutional update applied",
  "content": "_:entity3",
  "associations": [ "_:entity1", "_:entity2" ],
  "salience": 0.92,
  "provenance": {
    "Representer": "did:icore:system",
    "Timestamp": "2026-07-30T15:30:05Z",
    "Authority": { "authorityType": "inherent" },
    "RepresentationContext": "urn:icore:memory:episodic:v1"
  }
}

6.3 KnowledgeAtom store with schema and forgetting

{
  "@context": [
    "https://specs.initialcore.net/ns/core-ld.jsonld",
    "https://specs.initialcore.net/ns/memory-ld.jsonld"
  ],
  "@id": "_:semStore",
  "@type": "MemoryStore",
  "storeType": "semantic",
  "schema": "_:schema1",
  "policy": "_:policy1",
  "indexes": [ "_:idx1" ],
  "provenance": {
    "Representer": "did:icore:admin",
    "Timestamp": "2026-07-30T10:00:00Z",
    "Authority": { "authorityType": "inherent" },
    "RepresentationContext": "urn:icore:memory:stores:v1"
  }
}

{
  "@id": "_:schema1", "@type": "Schema",
  "entityTypes": ["KnowledgeAtom"],
  "requiredProperties": ["subject", "predicate", "object"]
}

{
  "@id": "_:policy1", "@type": "ForgettingPolicy",
  "retentionDuration": "P180D",
  "salienceThreshold": 0.2,
  "consolidationStrategy": "archive"
}

{
  "@id": "_:idx1", "@type": "Index",
  "name": "byPredicate",
  "indexType": "hash",
  "field": "predicate"
}

6.4 Memory backing for INPUT-LD

The MEMORY-LD profile pairs with INPUT-LD to create a complete request/remember loop:

Actor → INPUT-LD Request → System processes
                                  ↓
                           Episode recorded → MEMORY-LD Store
                                  ↓
                           Future queries recall via RetrievalSpec

7. Conformance

7.1 Implementation requirements

A conforming MEMORY-LD implementation MUST: 1. Satisfy all CORE-LD conformance requirements (§9 of CORE-LD) 2. Parse the MEMORY-LD @context extension (§5) in addition to the CORE-LD @context 3. Validate all 18 MEMORY-LD invariants (I-MEM-1 through I-MEM-18) 4. Reject documents that violate any CORE-LD or MEMORY-LD invariant 5. Accept any valid CORE-LD document as also valid at the CORE-LD level (MEMORY-LD is additive)

7.2 Conformance levels

8. Relationship to Other Profiles

9. Future Considerations

• Embedding integration: Standard embedding vector representation as Index metadata
• Memory federation: Cross-store query routing and knowledge merging
• Episodic compression: Timeline merging and conflict resolution for overlapping episodes
• Forgetting as a constitutional right: Audit trails for policy-driven memory removal
• Dream/inference: MEMORY-LD as substrate for reasoning across stored knowledge without external input

MEMORY-LD v0.1.0 — July 30, 2026. Draft. Extends CORE-LD v0.1.0. Second functional profile in the DAG--LD specification family.*