Changelog

Changelog

Significant changes to Douto, organized chronologically. Follows Keep a Changelog conventions.

Format: each entry lists the date, associated commit(s), and categorized changes (Added, Changed, Fixed, Removed).

---

v0.1.0 — Initial Setup (2026-02)

The Douto repository was created and populated with the pipeline scripts (migrated from an Obsidian vault) and the knowledge base structure.

2026-02-28 — North Star Definition

Commit: b7930d3 Reference: SEN-368

Added

• AGENTS.md updated with Jude.md north star — Douto is formally a component of the unified legal research platform (Juca + Leci + Douto + Valter = Jude.md)
• Issue epic SEN-368 defined as the convergence target

---

2026-02-28 — Knowledge Base Population

Commit: c4e2c5b

Added

• Populated MOCs with real corpus data — 56 books classified across 4 legal domains
• MOC_CIVIL.md — 35 books, ~9,365 chunks (largest domain)
• MOC_PROCESSUAL.md — 8 books, ~22,182 chunks
• MOC_EMPRESARIAL.md — 7 books
• MOC_CONSUMIDOR.md — placeholder structure (not yet populated)

---

2026-02-28 — Pipeline Migration

Commit: 8f9c702 Reference: SEN-358

Added

• pipeline/process_books.py — PDF to markdown extraction via LlamaParse (supports tiers: agentic, cost_effective, fast)
• pipeline/rechunk_v3.py — Intelligent legal chunking with 5 processing passes, 16 section patterns, running header detection, footnote grouping
• pipeline/enrich_chunks.py — Chunk enrichment via MiniMax M2.5 (5 concurrent threads, structured legal metadata)
• pipeline/embed_doutrina.py — Embedding generation using Legal-BERTimbau (768-dim, normalized)
• pipeline/search_doutrina_v2.py — Hybrid search (semantic cosine + BM25) with interactive CLI
• pipeline/requirements.txt — Python dependencies (sentence-transformers, torch, numpy, anthropic, llama-parse)

Note

These 5 scripts were developed inside an Obsidian vault before being migrated to this repository. They predate the repo itself. The migration commit brought them under version control for the first time.

---

2026-02-28 — Initial Setup

Commit: ce16dbc

Added

• AGENTS.md — Agent identity, responsibilities, boundaries, and git protocol
• knowledge/INDEX_DOUTO.md — Skill graph index mapping 8 legal domains
• knowledge/mocs/ — MOC directory structure
• knowledge/nodes/.gitkeep — Placeholder for future atomic notes
• tools/.gitkeep — Placeholder for future auxiliary tools
• .gitignore — Excludes node_modules, embeddings, .env, pycache

---

Documentation Session (2026-02-28)

In a single documentation session, the following strategic documents were created:

Added

• CLAUDE.md — Coding guidelines for AI code agents, aligned with the sens.legal ecosystem conventions (priority order, Python conventions, pipeline rules, knowledge base rules, embedding conventions, git patterns)
• PROJECT_MAP.md — Full project diagnostic: directory tree, stack details, architecture, data flow diagrams, gap analysis, recommendations
• ROADMAP.md — Product roadmap with 42 features across 6 milestones, 7 pending decisions, risk matrix, and mitigation plan
• PREMORTEM.md — Risk analysis: 6 false premises, 14 technical risks, 5 product risks, 4 execution risks, 7 edge cases, and the IP risk disclosure
• docs/ — Starlight documentation site with 22+ pages covering getting started, architecture, features, configuration, development, roadmap, and reference

---

Pre-History

The pipeline was developed over an undetermined period inside an Obsidian vault before this repository existed. Key facts about the pre-history:

• The 5 Python scripts in pipeline/ predate the repository
• The corpus (~50 books, ~31,500 chunks) was processed before migration
• Hardcoded paths in the scripts reflect the original development environments (Linux native and WSL)
• The enrichment prompt (enrich_prompt.md) was lost during migration and is not in the repository
• Linear issues (SEN-XXX) were used for tracking before the repository had GitHub Issues