v1.3.0: Expanded APJ Country Overlays

Highlights

• SEA overlay expanded to 6 ASEAN countries: Singapore, Malaysia, Thailand, Indonesia, Philippines, Vietnam — 24 identifiers, 16 masking functions (up from SG/MY only with 10 identifiers)
• India overlay improved: Fixed mask_pan_india to hide entity type character, added UAN/EPF, Ration Card, Vehicle Registration (10 identifiers, up from 7), updated DPDP Act enforcement context
• New masking functions: mask_thai_id, mask_nik, mask_npwp, mask_bpjs, mask_philsys, mask_tin_ph, mask_sss_ph, mask_cccd, mask_mst, mask_uan, mask_ration_card, mask_vehicle_reg
• Regulatory coverage: Thailand PDPA (2019), Indonesia PDP Law (2022), Philippines DPA (2012), Vietnam PDPD (2023)

Test Results

• AWS parallel CI: 16/16 scenarios passed
• Azure parallel CI: 15/16 passed (1 unrelated LLM flakiness)
• Unit tests: 71/71 passed

Changed Files

• shared/countries/SEA.yaml — expanded from 2 to 6 countries
• shared/countries/IN.yaml — fixed PAN masking, added 3 identifiers + 3 functions
• shared/function_registry.json — 12 new function entries
• shared/generate_abac.py — expanded category mappings and column hint inference
• shared/docs/country-overlays.md — updated overlay descriptions
• shared/tests/test_country_overlays.py — added ASEAN content tests

v1.2.0: Genie API Partner AI workaround + CI resilience

Highlights

• Genie API PATCH fallback: When Partner Powered AI hasn't propagated to new AWS workspaces, the Genie Space GET endpoint is blocked. The generate flow and test verification now fall back to PATCH (not gated) to read space config.
• LLM generation resilience: Inject USE CATALOG/USE SCHEMA when omitted, strip placeholder ellipsis (..., [...]), validate row filter column refs against table DDLs, expand function category mappings for overlay functions (IN/ANZ/SEA).
• CI infrastructure resilience: Stagger parallel workspace provisioning with jitter to avoid AWS API rate-limit bursts, retry transient errors in masking function deployment.
• aus-bank-demo scenario: Full integration test for Australian bank champion flow (ANZ + financial_services overlay, Genie Space import + promote).

Test Results

• AWS parallel CI: 16/16 scenarios passed
• Azure parallel CI: 16/16 scenarios passed

Changed Files

• shared/generate_abac.py — PATCH fallback, USE CATALOG injection, row filter validation, HCL syntax fixes, overlay function categories
• shared/deploy_masking_functions.py — Transient error retries with backoff
• shared/scripts/provision_test_env.py — Partner AI workspace enablement
• shared/scripts/remap_generated_config.py — Bare quoted catalog remap for genie mode
• shared/scripts/run_integration_tests.py — PATCH verification fallback, promote assertion fallback
• shared/scripts/run_parallel_tests.py — Staggered provisioning with jitter

GenieRails v1.1.0

Highlights

• 15/15 integration test scenarios passing on both AWS and Azure
• Australian Bank Demo with pre-configured Genie Space (sample questions, benchmarks, instructions)
• GenieRails logo added to README

CI & Test Reliability

• Per-run S3 buckets eliminate parallel provisioning race conditions
• Comprehensive governance mode fixes (genie_space_configs no longer leaks)
• sync_tag_policies suffix matching for parallel test environments
• DDL organized by catalog headers helps LLM track multi-catalog output
• Catalog coverage semantic check triggers retry when LLM forgets a catalog
• Ellipsis detection triggers retry when LLM outputs placeholder content
• Unsafe row filter rewrite removes hallucinated column references
• Cross-catalog function reference fix in autofix_fgac_arg_count_mismatch
• max_tokens increased to 32768 to prevent HCL truncation
• Default FMAPI model updated to databricks-claude-sonnet-4-6
• Parallel test timeout increased to 4 hours
• Workspace layer auto-import for orphaned warehouses
• Teardown cleans envs/ config files for fresh re-provisioning

Validation Improvements

• Forward country/industry overlays to post-generation validation
• Merge overlay function categories instead of replacing base categories
• Add customer_pii and patient_pii categories for PII masking functions
• Add payment_card category for card masking functions

Australian Bank Demo

• Pre-populate Genie Space with sample questions, benchmarks, instructions, SQL config
• Generate imports config verbatim from API (not regenerated by LLM)
• Review, tune & validate step added to demo README

Full Changelog: v1.0.0...v1.1.0

v1.0.0: Full ABAC governance with parallel test infrastructure

Highlights

• 15/15 integration test scenarios passing on both AWS and Azure at MAX_PARALLEL=15
• Tag vocabulary normalization eliminates LLM-generated tag value mismatches
• Parallel test runner with per-scenario workspace isolation, per-scenario logs, and --no-fail-fast mode
• wait_for_tag_policies.py with delete+recreate logic handles Databricks tag policy propagation delays

Features

• Role-based groups with fine-grained row and column policies
• AI-generated column masking SQL UDFs (SSN, credit cards, PII)
• Row-level security by region, department, or compliance scope
• Unity Catalog tag policies for sensitive column classification
• Consumer entitlements per group
• Per-space Genie ACLs (CAN_RUN scoped per space)
• Genie Space configuration as code (instructions, benchmarks, SQL measures)
• Dev → prod promotion with catalog remapping
• Country overlays (ANZ, India, Southeast Asia) and industry overlays (Financial Services, Healthcare, Retail)

Documentation

• Prerequisites — OS, Python, Terraform, Databricks account setup
• From UI to Production — import existing Genie Space, add governance, promote
• Version Control & Standalone Terraform — what to commit, version pinning, running Terraform directly
• Architecture — three-layer model, config files, Genie Space lifecycle