@@ -7,8 +7,91 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
77
88## [ Unreleased]
99
10+ ## [ 0.3.0] - 2026-05-15
11+
12+ ### Release highlights
13+
14+ Substantial feature release covering three major workstreams that landed
15+ between 0.2.3 and 0.3.0:
16+
17+ - ** Compiled structured extractors — full Phase C pipeline**
18+ ([ #75 ] ( https://github.com/GoogleCloudPlatform/BigQuery-Agent-Analytics-SDK/issues/75 ) )
19+ — deterministic source generation, JSON-to-plan parsing, LLM-driven plan
20+ resolution, retry-on-gate-failure orchestration, runtime fallback wiring,
21+ runtime extractor-registry adapter, orchestrator call-site swap,
22+ compile-and-measure utility, revalidation harness, BigQuery-table bundle
23+ mirror, `` bqaa-revalidate-extractors `` CLI with `` --events-bq-query-file `` ,
24+ and an operational rollout guide. Replaces per-event LLM extraction with
25+ deterministic code on the hot path while preserving the LLM fallback for
26+ unrecognized event shapes.
27+ - ** Ontology runtime reader**
28+ ([ #58 ] ( https://github.com/GoogleCloudPlatform/BigQuery-Agent-Analytics-SDK/issues/58 )
29+ reader follow-on to PR #92 's concept-index emission) —
30+ `` OntologyRuntime `` façade, `` EntityResolver `` Protocol with two reference
31+ implementations (`` ExactEntityResolver `` , `` LabelSynonymResolver `` ), and
32+ `` ConceptIndexLookup `` with fingerprint-strict verification across three
33+ trust points (eager `` verify() `` at construction, explicit re-checks,
34+ per-query `` WHERE compile_fingerprint `` defense in depth). Stable failure
35+ codes (`` FingerprintMismatchError `` , `` MetaTableMissingError `` ,
36+ `` MetaTableEmptyError `` , `` MetaTableMultipleRowsError `` ).
37+ - ** Binding + extraction validation toolkit**
38+ ([ #76 ] ( https://github.com/GoogleCloudPlatform/BigQuery-Agent-Analytics-SDK/issues/76 ) ,
39+ [ #105 ] ( https://github.com/GoogleCloudPlatform/BigQuery-Agent-Analytics-SDK/issues/105 ) )
40+ — `` validate_extracted_graph(...) `` ,
41+ `` validate_extracted_graph_from_ontology(...) `` ,
42+ `` validate_binding_against_bigquery(...) `` Python APIs;
43+ `` bq-agent-sdk binding-validate `` CLI for pre-flight validation;
44+ `` ontology-build --validate-binding `` and `` --location `` flags.
45+
46+ Examples shipped alongside the SDK release: the MAKO four-guarantee notebook
47+ demonstrating the compiled-extractor pipeline end-to-end
48+ ([ #107 ] ( https://github.com/GoogleCloudPlatform/BigQuery-Agent-Analytics-SDK/issues/107 ) ),
49+ and the A2A joint-lineage demo with auditor projections, the receiver
50+ `` A2A_INTERACTION `` typed view, and an audit-analyst agent that closes the
51+ BQAA loop
52+ ([ #129 ] ( https://github.com/GoogleCloudPlatform/BigQuery-Agent-Analytics-SDK/issues/129 ) ).
53+
1054### Added
1155
56+ - ** `` A2A_INTERACTION `` typed view (`` adk_a2a_interactions `` )** in
57+ `` src/bigquery_agent_analytics/views.py `` (PR
58+ [ #136 ] ( https://github.com/GoogleCloudPlatform/BigQuery-Agent-Analytics-SDK/pull/136 ) ).
59+ Surfaces the BQ AA Plugin's caller-side A2A delegation rows
60+ with JSON-extracted lineage columns —
61+ `` a2a_task_id `` , `` a2a_context_id `` , `` a2a_request `` ,
62+ `` a2a_response `` , plus a `` receiver_session_id_from_response ``
63+ COALESCE — so downstream consumers can join caller and receiver
64+ traces without writing the JSON-extraction SQL by hand. Used by
65+ the A2A joint-lineage demo's auditor projection.
66+ - ** `` CodeEvaluator.context_cache_hit_rate(...) `` + CLI support**
67+ (PR [ #114 ] ( https://github.com/GoogleCloudPlatform/BigQuery-Agent-Analytics-SDK/pull/114 ) ).
68+ New pre-built evaluator in
69+ `` src/bigquery_agent_analytics/evaluators.py `` that measures
70+ Gemini context-cache prefix-hit rate
71+ (`` cached_tokens / input_tokens `` ) per session, with
72+ cold-start / warm rate thresholds and an explicit
73+ `` fail_on_missing_telemetry `` switch. Wired through the
74+ `` bqaa evaluate --evaluator context_cache_hit_rate `` CLI path
75+ (`` src/bigquery_agent_analytics/cli.py `` ).
76+ - ** `` gm compile --emit-concept-index `` / `` --concept-index-table `` **
77+ CLI flags in `` src/bigquery_ontology/cli.py `` (PR
78+ [ #92 ] ( https://github.com/GoogleCloudPlatform/BigQuery-Agent-Analytics-SDK/pull/92 ) ,
79+ issue
80+ [ #58 ] ( https://github.com/GoogleCloudPlatform/BigQuery-Agent-Analytics-SDK/issues/58 )
81+ Phase 1). Emits a fingerprint-stamped concept-index table
82+ (`` label `` / `` synonym `` / `` notation `` rows + `` __meta `` )
83+ that the ontology runtime reader (above) verifies against.
84+ `` --concept-index-table `` is required when `` --emit-concept-index ``
85+ is set — no silent global default.
86+ - ** `` ontology-build --skip-property-graph `` ** flag in
87+ `` src/bigquery_agent_analytics/cli.py `` (PR
88+ [ #108 ] ( https://github.com/GoogleCloudPlatform/BigQuery-Agent-Analytics-SDK/pull/108 ) ,
89+ issue
90+ [ #104 ] ( https://github.com/GoogleCloudPlatform/BigQuery-Agent-Analytics-SDK/issues/104 ) ).
91+ Materializes node and edge tables without issuing the
92+ `` CREATE OR REPLACE PROPERTY GRAPH `` statement, letting users
93+ own their property-graph DDL while still letting the SDK
94+ populate the backing tables.
1295- ** Ontology runtime reader** in
1396 `` bigquery_agent_analytics.ontology_runtime `` and
1497 [ ` docs/ontology_runtime_reader.md ` ] ( docs/ontology_runtime_reader.md ) .
0 commit comments