DICE API Documentation/com.embabel.dice.projection.memory

Package-level declarations

Types

Runs the mark-and-sweep collector for a context: selects ACTIVE propositions, hands them to one or more CollectorStrategys for marking, then asks the SweepPolicy what to do with each marked one.

CollectorRunResult

data class CollectorRunResult constructor(val runId: String, val dryRun: Boolean, val marks: List<PropositionMark>, val applied: List<PropositionMark>, val skipped: List<PropositionMark>, val hardDeleted: List<String>, val startedAt: Instant, val finishedAt: Instant = Instant.now())

Immutable summary of a single collector run.

CollectorStrategy

fun interface CollectorStrategy

The "mark" half of the mark-and-sweep collector: looks at candidate propositions and reports which ones should be collected, and why.

ConsolidationResult

data class ConsolidationResult(val promoted: List<Proposition>, val reinforced: List<Proposition>, val discarded: List<Proposition>, val merged: List<PropositionMerge>)

Result of memory consolidation.

DecayCollectorStrategy

class DecayCollectorStrategy constructor(retireBelow: Double, retireDecayK: Double = 2.0) : CollectorStrategy

A CollectorStrategy that marks propositions whose decayed confidence has fallen below a staleness threshold.

DefaultCollectorRunner

class DefaultCollectorRunner(repository: PropositionRepository, strategies: List<CollectorStrategy>, policy: SweepPolicy, recordStore: CollectorRecordStore?, listener: DiceEventListener) : CollectorRunner

Default CollectorRunner implementation.

DefaultDreamLoopOrchestrator

data class DefaultDreamLoopOrchestrator(repository: PropositionRepository, passes: List<ConsolidationPass> = emptyList(), allowHardDelete: Boolean = false, changeVolumeThreshold: Int = 10) : DreamLoopOrchestrator

Default DreamLoopOrchestrator: runs a list of ConsolidationPasses as a single cycle.

DefaultMemoryConsolidator

class DefaultMemoryConsolidator(promotionThreshold: Double = 0.6, similarityThreshold: Double = 0.7, reinforcementBoost: Double = 0.1) : MemoryConsolidator

Default implementation of MemoryConsolidator.

DefaultMemoryMaintenanceOrchestrator

data class DefaultMemoryMaintenanceOrchestrator(repository: PropositionRepository, consolidator: MemoryConsolidator, abstractor: PropositionAbstractor? = null, abstractionThreshold: Int = 5, abstractionTargetCount: Int = 3, retireBelow: Double? = null, retireDecayK: Double = 2.0, decayManager: DecayManager? = null, collector: CollectorRunner? = null) : MemoryMaintenanceOrchestrator

Default MemoryMaintenanceOrchestrator: runs consolidation, abstraction, retirement, and optionally the mark-and-sweep collector as a four-stage pipeline.

DefaultMemoryRetriever

class DefaultMemoryRetriever(store: PropositionRepository, recencyWeight: Double = 0.3, confidenceWeight: Double = 0.3, knowledgeTypeClassifier: KnowledgeTypeClassifier = HeuristicKnowledgeTypeClassifier) : MemoryRetriever

Default implementation of MemoryRetriever backed by a PropositionRepository.

DreamLoopOrchestrator

interface DreamLoopOrchestrator : MemoryMaintenanceOrchestrator

Composes ConsolidationPasses into a repeatable consolidation cycle — the "dream loop".

DreamLoopReport

data class DreamLoopReport(val contextId: <Error class: unknown class>, val cycleStarted: Instant, val cycleCompleted: Instant = Instant.now(), val passResults: List<ConsolidationPassResult>, val totalExamined: Int, val totalTransitioned: Int, val totalNewPropositions: Int, val triggered: Boolean)

Summary of a single dream-loop consolidation cycle.

DuplicateCollectorStrategy

class DuplicateCollectorStrategy constructor(similarityThreshold: Double = 0.7, topK: Int = 10, traceStore: CollectorTraceStore? = null) : RunAwareCollectorStrategy

A CollectorStrategy that finds near-duplicate propositions and marks all but the strongest member of each duplicate group.

KnowledgeTypeClassifier

fun interface KnowledgeTypeClassifier

Strategy interface for classifying propositions into knowledge types. Implementations can use different heuristics based on domain needs.

MaintenanceResult

data class MaintenanceResult(val consolidation: ConsolidationResult?, val abstractions: List<Proposition>, val superseded: List<Proposition>, val retired: List<Proposition>, val collectorResult: CollectorRunResult? = null)

Result of a memory maintenance run.

MemoryConsolidator

interface MemoryConsolidator

Consolidates session propositions into long-term memory. End-of-session process that decides what to keep, reinforce, or discard.

MemoryMaintenanceOrchestrator

interface MemoryMaintenanceOrchestrator

Orchestrates memory maintenance — consolidating, abstracting, retiring, and persisting — as a single entry point suitable for end-of-session or scheduled batch runs.

MemoryProjection

data class MemoryProjection(val semantic: List<Proposition> = emptyList(), val episodic: List<Proposition> = emptyList(), val procedural: List<Proposition> = emptyList(), val working: List<Proposition> = emptyList())

Propositions organized by knowledge type.

MemoryProjector

interface MemoryProjector

Projects propositions into memory organized by knowledge type.

MemoryRetriever

interface MemoryRetriever

Retrieves propositions with memory semantics. Combines multiple retrieval strategies: similarity, entity overlap, recency.

PropositionMerge

data class PropositionMerge(val sources: List<Proposition>, val result: Proposition)

Represents a merge of multiple propositions into one.

RunAwareCollectorStrategy

interface RunAwareCollectorStrategy : CollectorStrategy

A CollectorStrategy that also wants the full CollectorRunContext (run id, dry-run flag, clock), not just the bare context id — so it can tag whatever it writes (e.g. trace rows) with the run that produced them.