v1.0.6 — pydantic CVE fix

Bump pydantic to >=2.7 to resolve CVE-2024-3772 (regex DoS).

v1.0.5 — Silence Smithery -32601 warning

Register empty list_resources and list_prompts handlers in the MCP server so Smithery's probe no longer hits 'method not found' (-32601). No functional change to tools.

v1.0.4 — Geo-Compound Blocking & Cardinality Filter

Geo-compound blocking and cardinality filter in column-context config

The pipeline's column-context config builder now produces smarter blocking and matchkey configurations.

What changed

• Geo-compound blocking: When geo columns exist in pipeline context, blocking keys compound them with name/string columns. Picks lowest-cardinality geo (state > city) for broadest discrimination.
• Soundex isolation: Soundex passes only apply to name fields — never compound geo+name keys where soundex on "CA" produces meaningless "C000".
• Cardinality filter: Low-cardinality string columns (e.g., hospital_type with 5 values) are excluded from fuzzy matchkeys to prevent score inflation.
• Null-safe cardinality: Uses drop_nulls().n_unique() instead of n_unique() which counts null as distinct.
• Helper extraction: _make_blocking() reduces config duplication across 4 blocking paths.

Install

pip install goldenpipe==1.0.4

Full Changelog: v1.0.3...v1.0.4

v1.0.3

Bug Fixes

• Fix schema mismatch in dedupe stage: Cast all DataFrame columns to Utf8 before calling GoldenMatch to prevent mixed-type crashes
• Fix OOM on oversized blocks: Set skip_oversized=True on pipeline-generated blocking config to prevent massive blocks (e.g. 7,600 empty-name records) from causing memory exhaustion
• Improve error diagnostics: Log full tracebacks on stage failures and debug-level config details

Testing

• Full pipeline on UK GIAS schools register (52,288 rows, 135 columns) — all 4 stages SUCCESS in 40s
• 120 tests passing

v1.0.2

Changes since v1.0.1

• feat: column context pipeline — GoldenCheck, GoldenFlow, and GoldenMatch now share column metadata through the pipeline context. Column types, null rates, and cardinality are classified once and flow downstream for smarter auto-config.
• refactor: ColumnType/CardinalityBand enums — type-safe column classification with post_init validation
• fix: expose unique/dupes/match_stats in pipeline artifacts — DedupeStage now stores full dedup results
• fix: utf8-lossy encoding for CSV reads — prevents crashes on Latin-1 data

Tested on 208,505 NC voter records: 191,962 distinct people in 34 seconds with zero config.

v1.0.1

Changes

• Add MCP Registry metadata (server.json, mcp-name verification)
• Add CI test workflow (ruff + pytest)
• Add community files (CODE_OF_CONDUCT.md, SECURITY.md)
• Fix version mismatch in init.py
• Add dynamic CI badge to README
• Add Documentation and Changelog URLs to pyproject.toml
• Clean up tracked internal files and expand .gitignore

v1.0.0

First stable release of GoldenPipe — the Golden Suite orchestrator.

Features

• End-to-end pipeline: chains GoldenCheck (scan) → GoldenFlow (transform) → GoldenMatch (dedupe) in one command
• Adaptive logic: skips transformation if no issues found, routes to PPRL if sensitive fields detected
• Pluggable stages: works with Golden Suite tools or custom stages via entry points
• 4 MCP tools: list_stages, validate_pipeline, run_pipeline, explain_pipeline
• Multiple interfaces: CLI, REST API, TUI, MCP server, A2A protocol
• Zero-config default: goldenpipe run data.csv just works
• Rich reporting: reasoning for every decision, timing metrics, artifact outputs

Install

pip install goldenpipe              # Core
pip install goldenpipe[golden-suite] # + all Golden Suite tools
pip install goldenpipe[all]          # Everything