live wire
IBM makes watsonx Orchestrate AgentOps, custom LLM judging and Bedrock-agent discovery generally availableIBMSchemaGate 0.1.45 fixes broken Oracle ADB wallet connections and an OCI stack pinned 28 releases behindSchemaGatePDI’s Amazon Quick procurement agent grounds spend answers in vendor, category and contract contextAWS Business Intelligence BlogBigQuery’s ML.METRICS example returns 0.84 accuracy but 0.30 macro-F1 on the same 100-row classification queryGoogle Cloud BigQuery docsSchemaGate 0.1.44 auto-selects sentence embeddings, lifting bundled-schema retrieval from 90/98 to 93/98SchemaGateSchemaGate 0.1.43 adds read-only SQL execution with per-principal table checks—and documents unauthenticated client assertionsSchemaGateDatabox adds reusable AI Analyst Skills with personal/company scope, auto-matching and marketplace installsDataboxFabric previews an AI builder for data-agent instructions, source guidance and example queriesMicrosoft FabricDatabricks trains data-agent retriever to stop early or spend bounded extra search steps, reporting 5.8-second latencyDatabricksThoughtSpot adds SpotterCode coding agent to its Visual Embed PlaygroundThoughtSpotLongMemEval-S audit: 67–73% of restore-fixable 80k-budget errors came from evicted evidence under three policiesarXivSchemaGate 0.1.42 adds dimension-aware retrieval and fixes complex multi-table SQL promptsSchemaGateSnowflake agent toolsets can silently drop inherited tools when callers lack accessSnowflake DocumentationLooker’s VS Code extension reaches GA with MCP-assisted LookML generation, editing and validationGoogle Cloud Looker release docsIBM makes watsonx Orchestrate AgentOps, custom LLM judging and Bedrock-agent discovery generally availableIBMSchemaGate 0.1.45 fixes broken Oracle ADB wallet connections and an OCI stack pinned 28 releases behindSchemaGatePDI’s Amazon Quick procurement agent grounds spend answers in vendor, category and contract contextAWS Business Intelligence BlogBigQuery’s ML.METRICS example returns 0.84 accuracy but 0.30 macro-F1 on the same 100-row classification queryGoogle Cloud BigQuery docsSchemaGate 0.1.44 auto-selects sentence embeddings, lifting bundled-schema retrieval from 90/98 to 93/98SchemaGateSchemaGate 0.1.43 adds read-only SQL execution with per-principal table checks—and documents unauthenticated client assertionsSchemaGateDatabox adds reusable AI Analyst Skills with personal/company scope, auto-matching and marketplace installsDataboxFabric previews an AI builder for data-agent instructions, source guidance and example queriesMicrosoft FabricDatabricks trains data-agent retriever to stop early or spend bounded extra search steps, reporting 5.8-second latencyDatabricksThoughtSpot adds SpotterCode coding agent to its Visual Embed PlaygroundThoughtSpotLongMemEval-S audit: 67–73% of restore-fixable 80k-budget errors came from evicted evidence under three policiesarXivSchemaGate 0.1.42 adds dimension-aware retrieval and fixes complex multi-table SQL promptsSchemaGateSnowflake agent toolsets can silently drop inherited tools when callers lack accessSnowflake DocumentationLooker’s VS Code extension reaches GA with MCP-assisted LookML generation, editing and validationGoogle Cloud Looker release docs
nl2sql.ai
guidePRACTITIONER GUIDE

A correct energy SQL query can still answer the wrong time period

For time-series analytics, query review cannot reveal measurements that never arrived. Completeness, units and gap treatment have to travel with the answer.

Valid query, incomplete period.
AI-generated illustration
By The News Desk· Sep 7, 2026the quick take — two AI hosts go live when you do

An energy text-to-SQL system can generate valid SQL, execute it successfully and still answer a different question from the one the user asked.

Logiciel Solutions describes the failure mode with a quarterly site-average request. If one site has eleven days of missing readings, a query that averages the rows present is arithmetically correct—but the result covers less than the requested quarter. Because the missing intervals are absent from the table, reviewing the generated SQL does not reveal the substitution.

That distinction matters for any natural-language interface over meter, sensor or historian data: execution success is not period completeness.

Put completeness in the result contract

Logiciel’s recommended design starts before generation. The semantic layer should encode each asset’s expected interval frequency, allowing the system to compute the proportion of expected readings that arrived alongside every aggregate. The answer should display that completeness measure rather than leave it as optional metadata.

This changes the acceptance test. Do not ask only whether generated SQL matches an approved query. Ask whether an answer for a requested period also states:

  • the expected and observed interval coverage;
  • how missing intervals were treated;
  • whether any values were interpolated; and
  • which period the returned aggregate actually represents.

A system that cannot supply those fields should not present a partial-period average as an unqualified quarterly result.

Units belong in the same boundary

Time-series sources may report at different granularities and in different units. Logiciel argues that conversion should happen centrally in the semantic layer, with mixed-unit aggregation prevented rather than repaired by each generated query. Aggregation rules also need explicit definitions: averaging pre-aggregated site values can weight a small site the same as a large one, even when that is not the intended business calculation.

The safe generation surface is therefore smaller than the raw warehouse schema. It consists of curated metrics with expected intervals, normalized units, gap policy and aggregation semantics attached.

Separate measurement from inference

Interpolation introduces another provenance boundary. If filled values are indistinguishable from observed readings, an answer can turn an estimate into apparent fact. Logiciel recommends marking interpolated values and stating gap treatment with the result.

This is practitioner guidance from a consultancy, not a published cross-system benchmark; the page does not name the energy client behind its example or provide an independently reproducible evaluation. But the implementation test is concrete: remove a known block of intervals, ask for the full-period aggregate, and verify that the interface reports incomplete coverage instead of returning a bare number.

For energy data, visible SQL is useful but insufficient. The answer contract must expose what the query cannot: the measurements that should have existed and did not.

Filed by The News Desk. Corrections: desk@nl2sql.ai · Our standards →

comments · 0

    Comments are moderated before they appear. Your email is used once to confirm it is you — never shown, never sold. Corrections and questions get an answer from the desk when we have one.

    A correct energy SQL query can still answer the wrong time period — nl2sql.ai