Foggy Capability Baseline and Evidence Matrix

This page is the current capability source for the implementation manual. It is not frozen v3.0 whitepaper content. The evidence was collected on 2026-08-22 against a disposable SQLite fixture and the Java Runtime. Production deployments must repeat the checks and add authentication, authorization, audit, network, and secret controls.

Evidence boundary

Item	Value in this run
Launcher	foggy-runtime-launcher v0.1.17
Engine	java
Runtime API	foggy-runtime-api/v1
Schema	2026-06-06
Security mode	none-dev-test-only
Namespace	salesdrop
Fixture	disposable SQLite + sales_drop_daily
Runtime source	foggy-data-mcp-bridge, published framework 9.2.0

none-dev-test-only is a development/test boundary, not a production security conclusion. Every deployment should read /api/v1/capabilities first and select the next steps from the reported capabilities.

Capability matrix

Status meanings: Verified means the disposable fixture executed successfully; Declared/not exercised means the Runtime advertises the capability but this run did not consume it; Conditional depends on configuration, identity, or an optional service; Disabled is explicitly unavailable in this instance.

Capability	Current status	Runtime/API or MCP entry point	Evidence and documentation decision
Readiness and capability discovery	Verified	GET /readyz, GET /api/v1/capabilities	Returned Java, API v1, and schema 2026-06-06; every implementation flow starts here.
Runtime security mode	Conditional	securityMode, management-auth configuration	Current mode is none-dev-test-only; the manual must not present the local Launcher as production-ready.
Datasource list/test	Verified	GET /api/v1/datasources, POST /api/v1/datasources/{name}/test	The default datasource passed its test; diagnostics must not expose passwords or complete connection strings.
Namespace datasource binding	Verified	PUT /api/v1/namespaces/{namespace}/datasource	Model validation failed in a fresh unbound namespace; binding default enabled the successful flow. Binding is a model-validation prerequisite.
Datasource diagnostics	Verified	GET /api/v1/datasources/diagnostics	Reported namespaceBindings.salesdrop=default; managedDatasourceCount=0 does not mean that a configured datasource is unavailable.
Table list and inspection	Verified	POST /api/v1/tables/list, POST /api/v1/tables/inspect	sales_drop_daily was inspected; this is controlled operations/modeling access, not a substitute for semantic queries.
Runtime SQL	Verified but restricted	POST /api/v1/sql/query	A read-only top-five query passed; the SQL surface must not become a normal MCP user-query endpoint.
Bundle registration	Verified	GET/POST /api/v1/bundles	sales-drop-models was registered before refresh; registration alone does not publish a model.
Model validation	Verified	POST /api/v1/models/validate	Passed after datasource binding; validation produces candidate diagnostics and does not replace the catalog directly.
Model refresh/list/describe	Verified	/api/v1/models/refresh, GET /api/v1/models, POST /api/v1/models/{model}/describe	Refresh exposed SalesDropDailyQueryModel; describe returned 21 fields and physical-table mapping.
Query validate/execute	Verified	`/api/v1/query/{model}/validate	execute`
Query explain	Verified	/api/v1/query/{model}/explain, dataset.explain_query	basis=RECOMPILED; this is fresh compilation evidence in the current context, not a historical execution trace.
Compose/FSScript	Declared/not exercised	`/api/v1/compose/validate	preview
Authoring workspace/resources/diff/validate	Declared/not exercised	/api/v1/authoring/...	The capability is supported; publish, recovery, and concurrent-revision behavior need a separate management-plane acceptance run.
Release package export	Declared/not exercised	/api/v1/authoring/workspaces/{id}/release-package	Advertised as supported; this run does not claim that a release drill is complete.
Production apply/rollback	Disabled	authoring.production.apply/rollback capabilities	Explicitly disabled in this instance; the v3.0 draft must preserve this conditional/disabled wording.
Release package import	Disabled	authoring.releasePackage.import capability	Explicitly disabled; the deployment manual must not promise direct production import.
MCP model discovery	Verified	dataset.list_models	Preferred discovery tool; it returns routing information, followed by dataset.describe_model_internal.
MCP model details	Verified	dataset.describe_model_internal	Takes model; new integrations must not use dataset.get_metadata for first-pass discovery.
Legacy metadata tool	Compatibility/deprecated	dataset.get_metadata	tools/list marks it Deprecated; new integrations must migrate to list + describe.
Single-model query	Declared with API evidence	dataset.query_model	Single-model filters, grouping, time windows, and pivot route here; cross-model work must not be forced into it.
Cross-model composition	Declared/not exercised	dataset.compose_script	Use only for Join/Union/derived/multi-plan flows; scripts must return { plans: plan }.
Chart export	Conditional	dataset.export_with_xchart, dataset.export_with_echarts	XChart is JVM-native with no browser/Node dependency; ECharts requires an optional rendering service and an explicit choice.
Natural-language query	Conditional	dataset_nl.query	Requires a ChatModel/AI Provider; structured tools should not be described as unavailable when the provider is absent.

Items for the v3.0 review

• Keep DRAFT / NOT FROZEN until the product and architecture owners confirm this matrix and the compatibility policy.
• Make datasource binding, model validate/refresh, catalog generation, and tool migration explicit implementation prerequisites.
• Define RECOMPILED separately from historical execution traces, and state that SQL/physical-name exposure is governed by namespace policy.
• Put the disabled/conditional state of production apply, rollback, and import in the compatibility matrix instead of promising a complete release loop early.

Reproduction entry point

See Runtime API and MCP Examples for curl, MCP JSON-RPC, and CLI examples. Delivery evidence should retain the JSON responses for capabilities, datasource diagnostics, model validate/refresh, and query validate/execute.