mcp_fact_graph 0.2.3

Temporal knowledge graph with evidence-based fact management, candidate confirmation, and automatic summarization for the MCP ecosystem.

0.2.3 - 2026-06-30 - Storage-port injection seam for persistence

Added

• The *StoragePort interfaces are now re-exported from the public barrel (mcp_fact_graph.dart). They were the injection seam of the primary FactGraphRuntime(...) constructor all along — exporting them lets a host supply persistent (disk-backed) storage implementations instead of the default in-memory stack, without path-importing src/.
• The in-memory storage stack (InMemory*Storage + InMemoryStorageContainer) is re-exported as the subclassing base for out-of-package persistence adapters: a persistent port can extends InMemory*Storage, override only the mutating methods to flush to disk, and reuse all query/index logic unchanged.

Notes

• Backward-compatible, additive only — no behavioral or signature changes to existing types. Consumers on ^0.2.2 resolve 0.2.3 with no constraint changes.
• The five *Query value types whose names also exist in mcp_bundle (ClaimQuery, EntityQuery, FactQuery, PatternQuery, RunQuery) are intentionally hidden from the barrel to avoid ambiguous-export errors in consumers that depend on both packages (e.g. mcp_knowledge). They remain reachable by path-importing src/ports/storage_port.dart.

0.2.2 - 2026-05-23 - mcp_bundle 0.4.0 cascade

Changed (cascade)

• mcp_bundle caret bumped from ^0.3.1 to ^0.4.0. mcp_fact_graph does not touch UiSection.pages directly, so this release is a caret-only cascade. Consumers should bump to ^0.2.2.

0.2.1 - 2026-05-04 - Microsecond consistency overlap

Fixed

• ConsistencyChecker._periodsOverlap now flags overlap iff two facts share the same occurredAt instant to the microsecond. Previously the helper reduced both timestamps to calendar-day precision, which caused every same-day, same (entity, factType) record to be reported as a triple-mismatch conflict regardless of the actual time gap. Point-in-time lifecycle records (e.g., agent.fork.assigned mirrors emitted by host runtimes) can now be ingested with enableConsistencyCheck: true without spurious conflicts.

Dependencies

• mcp_bundle: ^0.3.1 — additive; no behavioral change for existing call sites.

---

0.2.0 - 2026-04-28 - Standard Ports & Domain Expansion

Added

• Standard port adapters under src/adapters/ implementing mcp_bundle Contract Layer (capability-named ports).
• UnifiedStoragePort consolidating per-domain storage.
• L1 FactGraph domain entities — artifact, automation, classification, fact_cluster, fact_policy, relation, run, skill_run.
• LLM-call-reduction domain entities — extraction rules / validators, classifier memory, disambiguation decisions, LLM call log, idempotency records.
• Claim signal entity bridging L2 ContextOps and L3 SkillOps.
• Response validation entity.

Changed

• Domain entities reorganized into the 4-layer (L0 Evidence, L1 FactGraph, L2 ContextOps, L3 SkillOps) architecture.
• New dependency: mcp_bundle ^0.3.0.

Removed

• Legacy per-domain EvidencePort and event entity — replaced by Contract Layer ports and UnifiedStoragePort.

---

0.1.1 Dependency Update

Changed

• Updated mcp_bundle dependency to ^0.2.1 for Contract Layer compatibility

---

0.1.0 Initial Release

Added

Core Features

• Temporal Knowledge Graph

  • Fact model with domain, category, value, and temporal scope
  • Evidence model for source data with confidence scores
  • Candidate model for pending facts awaiting confirmation
  • Claim model for skill execution assertions

• Fact Graph Service

  • FactGraphService as the main entry point
  • Evidence ingestion and processing
  • Candidate creation and confirmation workflow
  • Fact querying with domain, entity, and temporal filters

• Context Bundles

  • ContextBundle for complete entity context retrieval
  • Facts, summaries, and recent claims aggregation
  • Optimized for LLM consumption

• Port-Based Architecture

  • FactStoragePort for fact persistence
  • EvidenceStoragePort for evidence storage
  • CandidateStoragePort for candidate management
  • EntityStoragePort for entity data
  • EventStoragePort for event sourcing
  • ViewStoragePort for materialized views

• In-Memory Implementations

  • Complete in-memory storage implementations for testing
  • Thread-safe operations with proper isolation

Storage Ports

• FactStoragePort - CRUD operations for facts
• EvidenceStoragePort - Evidence management
• CandidateStoragePort - Candidate lifecycle
• EntityStoragePort - Entity metadata
• EventStoragePort - Event sourcing support
• ViewStoragePort - Materialized views

Data Models

• Fact - Confirmed knowledge with temporal validity
• Evidence - Source data with metadata
• Candidate - Pending facts with confidence
• Claim - Skill execution assertions
• Entity - Entity metadata and summaries
• ContextBundle - Aggregated entity context

---

Support and Contributing

• Report Issues
• Join Discussions
• Read Documentation
• Support Development