Skip to content
sens Valter

Changelog

Changelog

Section titled “Changelog”

Significant changes organized by date, newest first. For session-level development logs, see PROGRESS.md in the repository root.

Categories: Added (new feature), Changed (modification to existing feature), Fixed (bug fix), Security (security-related change), Removed (removed feature or dead code).

February 2026

Section titled “February 2026”

2026-02-28 — Reorg, Roadmap v2, Project Map

Section titled “2026-02-28 — Reorg, Roadmap v2, Project Map”

Changed

  • Reorg Phases 0-2: security fixes, dependency cleanup, dead code removal
  • Roadmap v2: premortem analysis, reasoning chain vision, milestone restructuring with risk-informed planning
  • Issue organization: 5 GitHub milestones created, 9 new issues filed, 2 closed, 11 updated with labels and milestones

2026-02-27 — App Directory Preparation

Section titled “2026-02-27 — App Directory Preparation”

Added

  • Security hardening for App Directory requirements (input validation, error sanitization)
  • UX improvements for MCP tool descriptions and parameter schemas
  • Metadata alignment with App Directory submission format
  • Dry-run validation of App Directory submission flow

2026-02-25 — Quality Gates

Section titled “2026-02-25 — Quality Gates”

Added

  • Enforced ruff check and ruff format as quality gates
  • mypy strict typing on critical modules (api/deps.py, api/routes/ingest.py, mcp/tools.py)
  • make quality target combining lint + type check + test

2026-02-24 — Rate Limit Fail-Safe, Tracing, Decoupling

Section titled “2026-02-24 — Rate Limit Fail-Safe, Tracing, Decoupling”

Added

  • Rate limit fail-safe mechanism (initial implementation)
  • OpenTelemetry tracing with console exporter and trace_id propagation
  • Architectural decoupling: 10 sprints separating concerns across layers

Changed

  • Store implementations decoupled from direct core imports (dependency injection)

2026-02-22 — ChatGPT Integration, Search Enrichment

Section titled “2026-02-22 — ChatGPT Integration, Search Enrichment”

Added

  • ChatGPT integration live via MCP remote (HTTP/SSE transport)
  • Search result enrichment with graph metadata (citation count, connectivity score)

2026-02-21 — KG Quality, MCP Remote, Auth, Governance

Section titled “2026-02-21 — KG Quality, MCP Remote, Auth, Governance”

Added

  • MCP remote server with HTTP/SSE transport (mcp/remote_server.py)
  • Authentication hardening (API key validation, HMAC verification)
  • Operational governance documentation (CLAUDE.md governance section)

Changed

  • Knowledge graph quality improvements (node deduplication, relation validation)

2026-02-17 — Batch Classification, Railway Deploy

Section titled “2026-02-17 — Batch Classification, Railway Deploy”

Added

  • Batch document classification pipeline
  • Railway production deployment configuration

2026-02-16 — Pagination, STJ Metadata, Ingestion, Embeddings

Section titled “2026-02-16 — Pagination, STJ Metadata, Ingestion, Embeddings”

Added

  • Cursor-based pagination for search results
  • STJ metadata store (stores/stj_metadata.py) for tribunal-specific data
  • Ingestion workflow with state machine (core/workflow_state_machine.py)
  • Shortest-path graph queries between decisions
  • Embedding service with configurable model (embeddings/)

2026-02-15 — Features Search, MCP Desktop, Reranking, Graph Endpoints

Section titled “2026-02-15 — Features Search, MCP Desktop, Reranking, Graph Endpoints”

Added

  • Features search endpoint (search by extracted legal features)
  • MCP server for Claude Desktop (stdio transport)
  • Cross-encoder reranking for search results
  • 4 additional graph analytics endpoints (temporal analysis, minister comparison, criterion evolution, citation network)

2026-02-13 — Graph Analytics Endpoints

Section titled “2026-02-13 — Graph Analytics Endpoints”

Added

  • Graph analytics endpoints via POST /v1/graph/:
    • Divergence detection between ministers and turmas
    • Optimal legal argument path finding
    • Minister profile analysis (voting patterns, specializations)
    • Temporal evolution of legal criteria

2026-02-10 — Data Ingestion at Scale

Section titled “2026-02-10 — Data Ingestion at Scale”

Added

  • Large-scale data ingestion: 23,441 documents, 2,119 extracted features, 810,225 metadata records
  • Ingestion pipeline with PDF extraction, text processing, and metadata parsing

2026-02-09 — Project Foundation

Section titled “2026-02-09 — Project Foundation”

Added

  • Hybrid search engine (BM25 lexical + semantic vector search with RRF fusion)
  • Anti-hallucination verification against STJ public data (core/verifier.py)
  • IRAC legal analysis framework (models/irac.py)
  • MCP server skeleton with tool definitions
  • Initial corpus: 3,673 documents indexed with embeddings
  • PostgreSQL document store with JSONB metadata
  • Qdrant vector store for semantic search
  • Redis cache layer with rate limiting
  • Neo4j knowledge graph with FRBR-based ontology
  • FastAPI REST API with structured error handling
  • structlog JSON logging with request-scoped trace_id