Changelog

All notable changes to bigquery-data-science are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.8.0] - 2026-10-20

Added

• AI.AGG Structured Output & Error Tolerance Support:
  • Added coverage for GA output_schema parameter emitting typed STRUCT records directly from aggregation via SQL DDL schema strings.
  • Added coverage for GA max_error_ratio parameter for fault-tolerant aggregation.
  • Added Section 11 (output_schema) and Section 12 (max_error_ratio) to bigquery-ai/functions/ai_agg/ai_agg.ipynb and matching Examples 9 & 10 to bigquery-ai/functions/ai_agg/ai_agg.sql.
• gemini-4-argon Foundation Model Support:
  • Added gemini-4-argon to skills/pinned-models.json, bigquery-ai/RESOURCES.md, and bigquery-ai/reference/general-purpose-functions.md. Verified live via BigQuery.
• GoogleSQL Reusable Column Aliases GA:
  • Documented GoogleSQL GA support for referencing SELECT aliases inside expressions and WHERE clauses.

Changed

• AI.AGG Status Transition:
  • Graduated AI.AGG from Preview to GA in bigquery-ai/README.md, bigquery-ai/reference/managed-functions.md, and skills/bigquery-ai/reference/managed-functions.md.
• Skill Progressive Lenses Re-synchronized:
  • Lens 3 (skills/bigquery-ai/syntax/aiagg.md): Regenerated surgical syntax card with output_schema and max_error_ratio signatures.
  • Lens 4 (skills/bigquery-ai/narrative/ai_agg.md): Regenerated textbook narrative from the executed ai_agg.ipynb.
  • Refreshed manifests across all 3 agent skills (bigquery-data-science, bigquery-ml, bigquery-ai).

[0.7.0] - 2026-10-19

Added

• Standalone Repository & 1:1 Domain Normalization (bigquery-ml/ and bigquery-ai/):
  • Extracted and normalized all 61 BigQuery ML notebooks (bigquery-ml/) and 43 BigQuery AI notebooks (bigquery-ai/) with companion .sql scripts, exhaustive reference/*.md documentation, and unified bigquery-data-science Jupyter kernel metadata.
• Hybrid 3-Skill Hierarchy & 4-Lens Progressive Disclosure (skills/):
  • skills/bigquery-data-science/: Umbrella task router, 10 head-to-head empirical comparisons, cross-namespace composition guide, reviewer rubric, and reference/namespace-matrix.md prerequisite matrix.
  • skills/bigquery-ml/ and skills/bigquery-ai/: 4-Lens architecture — Lens 1 (SKILL.md routing & cross-cutting gotchas), Lens 2 (reference/*.md use-case decision guides), Lens 3 (syntax/*.md surgical H2-sliced per-model/per-function syntax cards + syntax/INDEX.md), and Lens 4 (narrative/*.md 104 boilerplate-stripped textbook walkthroughs with GitHub provenance banners).
• Zero-Dependency Python Package, CLI & Stdio MCP Server (src/bigquery_data_science/):
  • Skill & kernel installer (bigquery-data-science install, list, path, kernel) and federated statmike.agent_skills entry point.
  • Instant offline catalog engine (bigquery-data-science lookup, search, triage).
  • Zero-dependency JSON-RPC stdio MCP server (bigquery-data-science mcp) exposing triage_bigquery_task, lookup_bigquery_syntax, search_bigquery_gotchas, and get_bigquery_walkthrough.
  • Maintainer verification suite (sweep, validate, check-links, check-reference, check-syntax-sync, check-pinned-models, check-narratives, sync-syntax, convert-notebook, manifest).
• Unified Packaging & Layered uv Extras (pyproject.toml & uv.lock):
  • Zero-dependency < 1.5 MB PyPI wheel bundling skills/, with optional extras [notebooks], [pipelines], [workflows], and [all].