@duckcodeailabs/dql-agent 1.8.3 → 1.8.4

package/dist/answer-loop.js

@@ -29,7 +29,7 @@ import { evaluateDbtFirstGeneratedSql } from './metadata/dbt-first-safety.js';
 import { planAnalyticalPath, humanizeAnalyticalEntityId, analyticalPolicyUserFacingReason, } from './metadata/analytical-policy.js';
 import { planCertifiedAdaptation } from './metadata/block-adapt.js';
 import { compactSqlSnippet, extractSimpleSelectShape, selectExpressionOutputName, } from './metadata/sql-shape.js';
-import { composeSemanticQueryForQuestion, composeSemanticQueryFromMembers, } from './semantic-bridge/compose.js';
+import { composeSemanticQueryForQuestion, composeSemanticQueryFromCompiledMembers, composeSemanticQueryFromMembers, } from './semantic-bridge/compose.js';
 import { runAgenticToolLoop } from './agentic/tool-loop.js';
 import { buildSemanticStageTools } from './agentic/toolset.js';
 import { deriveAgenticTrust } from './agentic/answer-contract.js';
@@ -125,6 +125,93 @@ export function resolveAgentFilterValueBindings(value, schemaContext) {
         && normalizeValueIndexText(candidate.canonicalValue) === normalizeValueIndexText(best.canonicalValue))
         .map(({ distance: _distance, ...binding }) => binding);
 }
+/**
+ * AGT-005/AGT-012 — turn bounded value-index evidence into typed question
+ * state before route compatibility is evaluated. The value index, never an
+ * LLM or a metadata-token match, owns the canonical warehouse member. A value
+ * that matches multiple dimensions is intentionally left unbound so the
+ * certified lane cannot silently run with the wrong parameter.
+ */
+function questionPlanWithResolvedMemberBindings(plan, schemaContext) {
+    const existing = plan.requestedShape.memberBindings ?? [];
+    const existingValues = new Set(existing.flatMap((binding) => binding.values.map(normalizeValueIndexText)));
+    const candidates = Array.from(new Set([
+        ...(plan.valueMentions ?? []).map((mention) => mention.text.trim()),
+        ...(plan.requestedShape.filters ?? []).map((value) => value.trim()),
+    ].filter(Boolean)));
+    const resolved = candidates.flatMap((value) => {
+        if (existingValues.has(normalizeValueIndexText(value)))
+            return [];
+        const bindings = resolveAgentFilterValueBindings(value, schemaContext);
+        const dimensions = Array.from(new Set(bindings.map((binding) => binding.column.trim()).filter(Boolean)));
+        const canonicalValues = Array.from(new Set(bindings.map((binding) => binding.canonicalValue.trim()).filter(Boolean)));
+        if (dimensions.length !== 1 || canonicalValues.length !== 1)
+            return [];
+        const binding = bindings[0];
+        return [{
+                dimension: dimensions[0],
+                values: [canonicalValues[0]],
+                source: 'question',
+                confidence: binding.match === 'fuzzy' ? 'unique_partial' : 'exact',
+            }];
+    });
+    if (resolved.length === 0)
+        return plan;
+    const byIdentity = new Map();
+    for (const binding of [...existing, ...resolved]) {
+        const key = `${normalizeValueIndexText(binding.dimension)}\0${binding.values.map(normalizeValueIndexText).join('\0')}`;
+        if (!byIdentity.has(key))
+            byIdentity.set(key, binding);
+    }
+    return {
+        ...plan,
+        requestedShape: {
+            ...plan.requestedShape,
+            memberBindings: Array.from(byIdentity.values()),
+        },
+    };
+}
+/**
+ * Map typed member bindings to a block's declared parameter contract. The
+ * mapping is identifier-based (parameter name, semantic field, or declared
+ * filter binding) and must select exactly one parameter. Structural SQL is
+ * never accepted as a value.
+ */
+function certifiedInvocationInputs(block, plan) {
+    const definitions = block.parameters ?? [];
+    if (definitions.length === 0)
+        return {};
+    const parameters = {};
+    const parameterSources = {};
+    const declaredFilterBindings = new Map((block.filterBindings ?? []).map((entry) => [entry.filter, entry.binding]));
+    const topN = plan.requestedShape.topN?.n;
+    if (topN) {
+        const limitCandidates = definitions.filter((definition) => definition.name === 'top_n' || definition.binding?.kind === 'limit');
+        if (limitCandidates.length === 1) {
+            parameters[limitCandidates[0].name] = topN;
+            parameterSources[limitCandidates[0].name] = 'question';
+        }
+    }
+    for (const member of plan.requestedShape.memberBindings ?? []) {
+        const candidates = definitions.filter((definition) => {
+            if (Object.prototype.hasOwnProperty.call(parameters, definition.name))
+                return false;
+            const target = definition.binding?.kind === 'semantic_filter'
+                ? definition.binding.field
+                : declaredFilterBindings.get(definition.name) ?? definition.name;
+            return outputConceptMatches(target, member.dimension);
+        });
+        if (candidates.length !== 1)
+            continue;
+        const definition = candidates[0];
+        const values = Array.from(new Set(member.values.map((value) => value.trim()).filter(Boolean)));
+        if (values.length === 0 || (!definition.type.endsWith('[]') && values.length !== 1))
+            continue;
+        parameters[definition.name] = definition.type.endsWith('[]') ? values : values[0];
+        parameterSources[definition.name] = member.source === 'prior_result' ? 'prior_result' : 'question';
+    }
+    return Object.keys(parameters).length > 0 ? { parameters, parameterSources } : {};
+}
 function damerauLevenshteinDistance(left, right) {
     const rows = left.length + 1;
     const columns = right.length + 1;
@@ -181,6 +268,12 @@ function renderContextValidationRefusalForUser(code, machineError, memberBinding
         case 'unsafe_sql':
             return 'The drafted query used a statement type that is not allowed in this governed preview, so I did not run it.';
         case 'insufficient_context':
+            if (/could not be parsed|parse error|syntax/i.test(machineError)) {
+                return 'I drafted a query, but its SQL syntax did not match the connected warehouse, so I did not run it. The failed draft is available in Inspect; retrying will generate a warehouse-specific query.';
+            }
+            return machineError.trim().length > 0 && !/inspect_metadata_context/i.test(machineError)
+                ? `I could not prepare a governed query yet: ${machineError.replace(/\s*Use inspect_metadata_context[^.]*\.\s*$/i, '').trim()}`
+                : 'I could not prepare a governed query from the retrieved metadata. Name the specific metric or table and how to break it down, and I can generate a review-required draft.';
         default:
             return machineError.trim().length > 0 && !/inspect_metadata_context/i.test(machineError)
                 ? `I could not prepare a governed query yet: ${machineError.replace(/\s*Use inspect_metadata_context[^.]*\.\s*$/i, '').trim()}`
@@ -498,9 +591,10 @@ async function runAnswerLoop(input) {
     const considered = mergeHits(artifactHits, semanticHits, manifestHits, kg.search({ query: question, domain, limit: 10 })).slice(0, 30);
     const schemaContext = schemaContextWithAllowedSqlContext(schemaContextWithinQuestionScope(input.schemaContext ?? [], input.contextPack, scopedContextPack), scopedContextPack);
     const catalogRoute = input.contextPack?.routeDecision;
-    const questionPlan = input.contextPack?.questionPlan?.requestedShape
+    const baseQuestionPlan = input.contextPack?.questionPlan?.requestedShape
         ? input.contextPack.questionPlan
         : buildAnalysisQuestionPlan(question, input.followUp);
+    const questionPlan = questionPlanWithResolvedMemberBindings(baseQuestionPlan, schemaContext);
     // Retrieval may surface a high-trust block because its source tables and
     // vocabulary overlap the question even when its output contract does not.
     // Keep such candidates in the audit trail, but do not put their SQL or a stale
@@ -546,7 +640,15 @@ async function runAnswerLoop(input) {
             kg,
         })
         : null;
-    const shouldTryCertifiedRoute = shouldUseCertifiedRoute(catalogRoute, intent);
+    const explicitlyRequestedCertifiedBlock = effectiveBlockHints.some((hint) => {
+        const node = kg.getNode(`block:${hint}`);
+        return Boolean(node
+            && node.status === 'certified'
+            && objectNameInQuestion(question, node)
+            && /\b(run|use|open|show|execute|certified|saved|block)\b/i.test(question));
+    });
+    const shouldTryCertifiedRoute = shouldUseCertifiedRoute(catalogRoute, intent)
+        || explicitlyRequestedCertifiedBlock;
     const catalogCertifiedHit = shouldTryCertifiedRoute
         ? certifiedHitFromContextPack(input.contextPack, kg)
         : null;
@@ -613,7 +715,10 @@ async function runAnswerLoop(input) {
         let executionError;
         if (artifactHit.node.kind === 'block' && input.executeCertifiedBlock) {
             try {
-                result = await input.executeCertifiedBlock(artifactHit.node, { question });
+                result = await input.executeCertifiedBlock(artifactHit.node, {
+                    question,
+                    ...certifiedInvocationInputs(artifactHit.node, questionPlan),
+                });
                 result = trimResultToRequestedTopN(result, questionPlan);
             }
             catch (err) {
@@ -928,6 +1033,8 @@ async function runAnswerLoop(input) {
     let governedMetricAnswer = false;
     const semanticBridgeToolCalls = [];
     let semanticBridgeAnswer;
+    let semanticRuntimeFailure;
+    let semanticRuntimeCompiledAnswer = false;
     if (input.semanticLayer && semanticMetricMatch) {
         semanticBridgeAnswer = composeSemanticQueryForQuestion({
             semanticLayer: input.semanticLayer,
@@ -939,6 +1046,44 @@ async function runAnswerLoop(input) {
             ...(input.semanticDriver ? { driver: input.semanticDriver } : {}),
             ...(input.semanticTableMapping ? { tableMapping: input.semanticTableMapping } : {}),
         });
+        // Exact scalar metrics need no AI member selection. If the native compiler
+        // cannot express a derived/ratio/cumulative metric, hand the identifier to
+        // the host's shared dbt Cloud/MetricFlow adapter immediately.
+        if (!semanticBridgeAnswer
+            && input.semanticQueryCompiler
+            && questionPlan.requestedShape.dimensions.length === 0
+            && questionPlan.dimensionTerms.length === 0
+            && questionPlan.requestedShape.filters.length === 0
+            && questionPlan.timeTerms.length === 0) {
+            const matchedName = input.semanticLayer.listMetrics().find((metric) => metric.name === semanticMetricMatch.metric.name
+                || metric.name.endsWith(`.${semanticMetricMatch.metric.name}`)
+                || semanticMetricMatch.metric.name.endsWith(`.${metric.name}`))?.name;
+            if (matchedName) {
+                const selection = { metrics: [matchedName], dimensions: [] };
+                try {
+                    const compiled = await input.semanticQueryCompiler(selection);
+                    semanticBridgeAnswer = composeSemanticQueryFromCompiledMembers({
+                        semanticLayer: input.semanticLayer,
+                        question,
+                        selection,
+                        sql: compiled.sql,
+                    });
+                    if (semanticBridgeAnswer) {
+                        semanticRuntimeCompiledAnswer = compiled.engine !== 'native';
+                        semanticBridgeToolCalls.push({
+                            name: 'compile_semantic_query',
+                            status: 'checked',
+                            inputSummary: `metric: ${matchedName}`,
+                            outputSummary: `Compiled through ${compiled.engine}`,
+                            order: semanticBridgeToolCalls.length + 1,
+                        });
+                    }
+                }
+                catch (error) {
+                    semanticRuntimeFailure = error instanceof Error ? error.message : String(error);
+                }
+            }
+        }
         // Lane-2 LLM fallback: the deterministic token matcher missed, but a metric
         // matched so the semantic layer IS relevant. Spend ONE call to have the model
         // pick members (the query_semantic_model contract); the compiler still owns the
@@ -953,13 +1098,37 @@ async function runAnswerLoop(input) {
                 reasoningEffort: input.reasoningEffort,
             });
             if (selection) {
-                const composed = composeSemanticQueryFromMembers({
+                let composed = composeSemanticQueryFromMembers({
                     semanticLayer: input.semanticLayer,
                     question,
                     selection,
                     ...(input.semanticDriver ? { driver: input.semanticDriver } : {}),
                     ...(input.semanticTableMapping ? { tableMapping: input.semanticTableMapping } : {}),
                 });
+                if (!composed && input.semanticQueryCompiler) {
+                    try {
+                        const compiled = await input.semanticQueryCompiler(selection);
+                        composed = composeSemanticQueryFromCompiledMembers({
+                            semanticLayer: input.semanticLayer,
+                            question,
+                            selection,
+                            sql: compiled.sql,
+                        });
+                        if (composed) {
+                            semanticRuntimeCompiledAnswer = compiled.engine !== 'native';
+                            semanticBridgeToolCalls.push({
+                                name: 'compile_semantic_query',
+                                status: 'checked',
+                                inputSummary: `metrics: ${selection.metrics.join(', ')}${selection.dimensions?.length ? `; by ${selection.dimensions.join(', ')}` : ''}`,
+                                outputSummary: `Compiled through ${compiled.engine}`,
+                                order: semanticBridgeToolCalls.length + 1,
+                            });
+                        }
+                    }
+                    catch (error) {
+                        semanticRuntimeFailure = error instanceof Error ? error.message : String(error);
+                    }
+                }
                 // Coverage guard: if the question asked for a breakdown but the LLM
                 // selection produced none, the governed answer would silently DROP the
                 // requested grouping (governed-but-wrong). Fall through to Lane-3
@@ -984,6 +1153,36 @@ async function runAnswerLoop(input) {
             }
         }
     }
+    if (!semanticBridgeAnswer && semanticRuntimeFailure) {
+        const text = `The governed semantic metric was found, but its semantic runtime could not compile the request: ${semanticRuntimeFailure}`;
+        return {
+            kind: 'no_answer',
+            sourceTier: 'no_answer',
+            certification: 'analyst_review_required',
+            reviewStatus: 'none',
+            confidence: 0,
+            text,
+            answer: text,
+            refusalCode: 'modeling_gap',
+            refusalDetails: { code: 'semantic_runtime_required', message: text },
+            citations: [],
+            memoryContext: input.memoryContext,
+            evidence: buildNoAnswerEvidence({
+                question,
+                reason: text,
+                artifactHits,
+                businessHits,
+                semanticHits,
+                manifestHits,
+                considered,
+                memoryContext: input.memoryContext ?? [],
+                toolCalls: semanticBridgeToolCalls,
+            }),
+            contextPack: input.contextPack,
+            considered,
+            providerUsed: provider.name,
+        };
+    }
     const metricFirst = semanticMetricMatch
         ? buildGovernedMetricFirstSql({
             metric: semanticMetricMatch.metric,
@@ -1000,6 +1199,7 @@ async function runAnswerLoop(input) {
     let contextLedger = createContextLedger({
         contextPack: input.contextPack,
         schemaContext,
+        dialect: input.semanticDriver,
     });
     // W2.2 — certified-block adaptation lane. When a certified block is context-only
     // ONLY because the question adds exactly one filter whose value maps to a column
@@ -1042,6 +1242,7 @@ async function runAnswerLoop(input) {
             kg,
             driver: input.semanticDriver,
             tableMapping: input.semanticTableMapping,
+            semanticQueryCompiler: input.semanticQueryCompiler,
             onCompiled: (record) => compiledSemanticRecords.push(record),
         }),
     ];
@@ -1156,6 +1357,7 @@ async function runAnswerLoop(input) {
             if (trust.tier === 'semantic_metric' && trust.compiled) {
                 governedMetricAnswer = true;
                 const compiled = trust.compiled;
+                semanticRuntimeCompiledAnswer = compiled.engine === 'metricflow-cli' || compiled.engine === 'dbt-cloud';
                 const governedNote = `Answered from governed semantic metric${compiled.metrics.length === 1 ? '' : 's'} ${compiled.metrics.join(', ')}${compiled.dimensions.length > 0 ? ` by ${compiled.dimensions.join(', ')}` : ''} (compiled via the semantic layer). Reusable block certification remains a separate review.`;
                 parsed = { ...parsed, text: parsed.text ? `${parsed.text}\n\n${governedNote}` : governedNote };
                 proposalToolCalls.push({
@@ -1352,13 +1554,19 @@ async function runAnswerLoop(input) {
         && questionPlan.requestedShape.filters.length === 0
         && !questionPlan.requestedShape.topN;
     let contextValidation;
-    const initialValidation = contextLedger.validateSql(parsed.sql, {
-        question,
-        intent,
-        filterValues: input.followUp?.filters,
-        trustedFilterValues: trustedFollowUpFilterValues(input.followUp),
-        memberBindings: input.followUp?.memberBindings,
-    });
+    // dbt Cloud/MetricFlow SQL has already been compiled against the selected
+    // semantic environment. A thin local retrieval pack cannot safely reject it
+    // and replace it with a guessed leaf-measure SQL. Warehouse execution remains
+    // the final binder/runtime check.
+    const initialValidation = semanticRuntimeCompiledAnswer
+        ? { ok: true, warnings: ['SQL compiled by the configured semantic runtime.'] }
+        : contextLedger.validateSql(parsed.sql, {
+            question,
+            intent,
+            filterValues: input.followUp?.filters,
+            trustedFilterValues: trustedFollowUpFilterValues(input.followUp),
+            memberBindings: input.followUp?.memberBindings,
+        });
     const rankedGrainGap = missingRankedGrainOutput(questionPlan, parsed.sql, semanticBridgeAnswer?.dimensions);
     contextValidation = initialValidation.ok && rankedGrainGap
         ? {
@@ -1654,7 +1862,7 @@ async function runAnswerLoop(input) {
         };
     }
     const dbtFirstJoinSafety = input.manifest
-        ? evaluateDbtFirstGeneratedSql(parsed.sql, input.manifest, input.domainContext?.purpose, input.domainContext)
+        ? evaluateDbtFirstGeneratedSql(parsed.sql, input.manifest, input.domainContext?.purpose, input.domainContext, input.semanticDriver)
         : undefined;
     if (dbtFirstJoinSafety && !dbtFirstJoinSafety.safe) {
         // Business-language text for the chat surface; the machine policy message
@@ -3162,6 +3370,15 @@ function applyParsedProposalMetadata(target, source) {
 function buildCertifiedBlockDqlArtifact(block, result) {
     if (block.kind !== 'block')
         return undefined;
+    const parameters = block.parameters ?? [];
+    const parameterValues = Object.fromEntries((result?.parameters ?? []).map((entry) => [entry.name, entry.value]));
+    const parameterContract = renderCertifiedParameterContract(block);
+    const artifactRuntime = {
+        ...(parameters.length > 0 ? { parameters } : {}),
+        ...(Object.keys(parameterValues).length > 0 ? { parameterValues } : {}),
+        persistence: 'saved',
+        trustState: 'certified',
+    };
     const sql = block.sql?.trim() ?? result?.sql?.trim() ?? block.examples?.find((example) => example.sql?.trim())?.sql?.trim();
     if (!sql) {
         // The block answered but its SQL was not inlined in the index (navigation /
@@ -3177,6 +3394,7 @@ function buildCertifiedBlockDqlArtifact(block, result) {
             source: `// Certified DQL block "${escapeDqlArtifactString(block.name)}"`
                 + `${block.sourcePath ? `\n// source: ${block.sourcePath}` : ''}`
                 + `\n// SQL is not inlined in the index — open the source file to view or edit the query.`,
+            ...artifactRuntime,
         };
     }
     const domain = block.domain ?? 'misc';
@@ -3186,7 +3404,7 @@ function buildCertifiedBlockDqlArtifact(block, result) {
     domain = "${escapeDqlArtifactString(domain)}"
     type = "custom"
     status = "certified"${block.owner ? `\n    owner = "${escapeDqlArtifactString(block.owner)}"` : ''}
-    description = """${escapeDqlArtifactTripleString(description)}"""${sourcePathComment}
+    description = """${escapeDqlArtifactTripleString(description)}"""${sourcePathComment}${parameterContract}
 
     query = """
         ${sql.replace(/"""/g, '\\"\\"\\"').split('\n').join('\n        ')}
@@ -3198,8 +3416,51 @@ function buildCertifiedBlockDqlArtifact(block, result) {
         name: block.name,
         sourcePath: block.sourcePath,
         source,
+        compiledSql: result?.sql ?? sql,
+        ...artifactRuntime,
     };
 }
+function renderCertifiedParameterContract(block) {
+    const parameters = block.parameters ?? [];
+    if (parameters.length === 0)
+        return '';
+    const lines = ['\n\n    params {'];
+    for (const parameter of parameters) {
+        const defaultValue = parameter.default === undefined ? '' : ` = ${renderCertifiedDqlParameterValue(parameter.default)}`;
+        lines.push(`      ${parameter.name}: ${parameter.type}${defaultValue}`);
+    }
+    lines.push('    }', '    parameterPolicy {');
+    for (const parameter of parameters)
+        lines.push(`      ${parameter.name} = "${parameter.policy}"`);
+    lines.push('    }');
+    const explicitBindings = new Map((block.filterBindings ?? []).map((entry) => [entry.filter, entry.binding]));
+    const bindings = parameters.flatMap((parameter) => {
+        const explicit = explicitBindings.get(parameter.name);
+        if (explicit)
+            return [{ name: parameter.name, value: explicit }];
+        if (parameter.binding?.kind === 'semantic_filter')
+            return [{ name: parameter.name, value: parameter.binding.field }];
+        if (parameter.binding?.kind === 'limit')
+            return [{ name: parameter.name, value: 'limit' }];
+        return [];
+    });
+    if (bindings.length > 0) {
+        lines.push('    filterBindings {');
+        for (const binding of bindings)
+            lines.push(`      ${binding.name} = "${escapeDqlArtifactString(binding.value)}"`);
+        lines.push('    }');
+    }
+    return lines.join('\n');
+}
+function renderCertifiedDqlParameterValue(value) {
+    if (Array.isArray(value))
+        return `[${value.map(renderCertifiedDqlParameterValue).join(', ')}]`;
+    if (typeof value === 'number' && Number.isFinite(value))
+        return String(value);
+    if (typeof value === 'boolean')
+        return value ? 'true' : 'false';
+    return `"${escapeDqlArtifactString(String(value ?? ''))}"`;
+}
 function buildGeneratedSqlDqlArtifact(input) {
     const sql = input.sql?.trim();
     if (!sql)