v2.0 Release Note
Document version: v2.0
Position: the additional-capability whitepaper after v1.0. v2.0 continues the Foggy semantic-layer engine story without deprecating v1.0 or replacing its foundational syntax references.
Implementation policy: this document focuses on Java-engine capabilities that are implemented and signed off. Planned, prompt-only, and unsigned experimental behavior is not described as stable functionality.
Core changes
v1.0 established the foundation: TM/QM semantic modeling, JSON Query DSL, MCP tools, permission governance, query evidence, and Compose analysis.
v2.0 advances governed query into governed complex analysis:
- Java MCP can recalibrate a route and redispatch instead of continuing an expired plan.
- Query processing can emit non-blocking trace correlation for catalog, planner, tool call, validator, and result stages.
- DSL_CTE provides staged analysis with governed aggregation, derivation, windows, post-filtering, ordering, and limited cross-result bridges.
- Relation result-stage expressions are a signed formula set for ratios, differences, delta ratios, labels, and ordered buckets.
- Memory Grid provides bounded handles, lifecycle, resource limits, and fail-closed guards for secondary analysis.
- Pivot enters the signed Java-engine scope with tree axes, axis windows, drilldown contracts, and database date/weekday corrections.
- Experience Recipe becomes a minimal registry lookup with namespace, tenant, permission, owner, and active-state filtering.
Capability summary
| Area | v2.0 addition | Boundary |
|---|---|---|
| Java MCP orchestration | calibrated-route redispatch and non-blocking trace correlation | Not an autonomous analyst; validator, policy, and fail-closed rules remain required |
| DSL_CTE staged analysis | aggregate, derive, window, postSlice, orderBy, and governed bridge templates | No arbitrary SQL CTE or physical-table join |
| Governed expressions | ratio, difference, delta ratio, absolute delta ratio, threshold label, ordered bucket, alias DAG, signed ranking | No arbitrary functions, CASE, physical fields, or free expression language |
| Memory Grid | bounded result handle, resolver, lifecycle, limits, cross-model alignment | No unbounded detail, arbitrary DuckDB API, full outer/multi-key join, or durable backend promise |
| Pivot | tree axis, axis window/domain selection, cascade drilldown, weekday dialect parity | No unsigned shape, arbitrary interactive expansion, or complete BI pivot product |
| Experience Recipe | exact registry lookup, governance filtering, publish evidence, signature-verifier SPI | Not a recipe marketplace, vector search, admin console, or trust root |
| Multi-database evidence | SQLite, MySQL8, PostgreSQL pivot/DSL_CTE evidence and SQL Server weekday parity | Not a complete SQL Server Maven profile release gate |
State vocabulary
| State | Meaning | Documentation rule |
|---|---|---|
| Signed off | Covered by Java runtime, validator, fixture, CI, or acceptance evidence | May be described as a v2.0 stable capability |
| Narrow contract available | Works only within a signed template, model, shape, or dialect evidence | Always state the use case and rejection boundary |
| Follow-up direction | Planned or evaluated, but not a stable Runtime promise | Keep it in exclusions or future direction |
Relationship to v1.0
v2.0 keeps v1.0 principles:
- The LLM does not access a bare database schema directly.
- Requests enter through semantic models, structured DSL, and tools.
- The engine owns permissions, field validation, dialect translation, and execution evidence.
- Prompt instructions are not a security boundary.
The new contracts increase analytical power without relaxing governance. The more complex the analysis, the more important signed contracts, test evidence, and fail-closed behavior become.
Not included in v2.0
- Free physical SQL execution.
- Arbitrary Semantic SQL, CTE, or expression language.
- Generic cross-model joins, generic funnels, generic attribution, or arbitrary period comparisons.
- A complete BI product, report designer, SQL notebook, or customer-facing AI analytics workbench.
- A complete SQL Server release gate, production durable Memory Grid backend, full Recipe marketplace, or governed vector search.
- Full parity for all new capabilities between Python and Java.