@memberjunction/testing-engine 6.1.0-edge.5 → 6.1.0-edge.7

package/dist/oracles/AgentDecisionOracle.js

@@ -0,0 +1,107 @@
+/**
+ * @fileoverview Agent-decision oracles — the centrepiece of the prompt-eval corpus (test plan §3.2).
+ * @module @memberjunction/testing-engine
+ */
+import { normalizeDecision } from '../eval/decision.js';
+import { evaluateCorpusExpectation } from '../eval/corpus.js';
+import { evaluateWellFormed } from '../eval/wellFormed.js';
+function readActual(input) {
+    const actual = input.actualOutput;
+    // A driver that produced nothing at all still has to be judged rather than crash the run —
+    // "no output" is the single most important observation this suite makes.
+    return actual?.turn ? actual : { turn: { text: typeof actual === 'string' ? actual : '' } };
+}
+/**
+ * `agent-decision-match` — did the agent decide the right thing?
+ *
+ * Reads the decision off whichever channel the model used (envelope or native tool calls) and
+ * compares it against an encoding-independent expectation. That normalization is what lets a
+ * baseline cell and a native cell be compared on an identical corpus case: the
+ * yardstick does not move between them, only the model's behavior does.
+ *
+ * The result's `details` carry the components separately — decision kind, action accuracy, param
+ * fidelity, per-matcher outcomes — so §6.1's metrics aggregate out of persisted `ResultDetails`
+ * without re-running anything.
+ *
+ * Configuration is the {@link DecisionExpectation} itself.
+ */
+export class AgentDecisionOracle {
+    constructor() {
+        this.type = 'agent-decision-match';
+    }
+    async evaluate(input, config) {
+        try {
+            const expectation = config;
+            if (!expectation?.kind) {
+                return { oracleType: this.type, passed: false, score: 0, message: "Configuration requires a 'kind' (the expected decision)" };
+            }
+            const actual = readActual(input);
+            const observed = normalizeDecision(actual.turn);
+            // evaluateCorpusExpectation, NOT evaluateDecision: only the former unwraps `anyOf`.
+            // `anyOf` is a corpus-level kind — a case that legitimately admits several decisions —
+            // and evaluateDecision would compare the literal string 'anyOf' against the observed
+            // kind and fail every time, which is a scored failure the model can never avoid.
+            const evaluation = evaluateCorpusExpectation((input.test?.Name ?? 'case'), expectation, observed);
+            return {
+                oracleType: this.type,
+                passed: evaluation.passed,
+                score: evaluation.score,
+                message: evaluation.passed
+                    ? `decided '${observed.kind}' via ${observed.encoding} as expected`
+                    : evaluation.messages.join('; '),
+                details: {
+                    expectedKind: expectation.kind,
+                    observedKind: observed.kind,
+                    encoding: observed.encoding,
+                    decisionKindMatch: evaluation.decisionKindMatch,
+                    actionAccuracy: evaluation.actionAccuracy,
+                    paramFidelity: evaluation.paramFidelity,
+                    paramResults: evaluation.paramResults,
+                    forbiddenViolations: evaluation.forbiddenViolations,
+                    observedActions: observed.actions.map((a) => a.name),
+                    observedSubAgents: observed.subAgents.map((s) => s.name),
+                    // Surfaced so the scorecard can count dual-channel
+                    // turns and concatenated envelopes without re-running anything.
+                    dualChannel: observed.dualChannel ?? false,
+                    payloadChanged: observed.payloadChange !== undefined,
+                    narrationWithCalls: observed.narrationWithCalls ?? false,
+                    placeholderCall: observed.placeholderCall ?? false,
+                    shadowEnvelopeKind: observed.shadowEnvelopeKind,
+                    diagnostic: observed.diagnostic
+                }
+            };
+        }
+        catch (error) {
+            return { oracleType: this.type, passed: false, score: 0, message: `Decision evaluation error: ${error.message}` };
+        }
+    }
+}
+/**
+ * `response-well-formed` — was the output usable at all?
+ *
+ * This is the oracle that measures the malformed/discard class directly — the issue's headline
+ * number. It asks strictly less than {@link AgentDecisionOracle}: a turn can be well-formed and
+ * wrong, but never malformed and right. Keeping them apart is what lets a comparison say *which* of
+ * "malformed rate fell" and "decision accuracy rose" actually happened.
+ *
+ * Configuration is a {@link WellFormedConfig}. Note `requireEnvelope` defaults to **off**: measurement
+ * measured that a correct native tool call carries no text at all on Google, so requiring an
+ * envelope unconditionally would fail every correct native turn.
+ */
+export class ResponseWellFormedOracle {
+    constructor() {
+        this.type = 'response-well-formed';
+    }
+    async evaluate(input, config) {
+        try {
+            const actual = readActual(input);
+            const observed = normalizeDecision(actual.turn);
+            const result = evaluateWellFormed({ decision: observed, finishReason: actual.finishReason, executionError: actual.executionError }, config);
+            return { oracleType: this.type, passed: result.passed, score: result.score, message: result.message, details: result.details };
+        }
+        catch (error) {
+            return { oracleType: this.type, passed: false, score: 0, message: `Well-formedness evaluation error: ${error.message}` };
+        }
+    }
+}
+//# sourceMappingURL=AgentDecisionOracle.js.map

package/dist/oracles/AgentDecisionOracle.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"AgentDecisionOracle.js","sourceRoot":"","sources":["../../src/oracles/AgentDecisionOracle.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAIH,OAAO,EAAE,iBAAiB,EAAgB,MAAM,kBAAkB,CAAC;AAEnE,OAAO,EAAE,yBAAyB,EAA0B,MAAM,gBAAgB,CAAC;AACnF,OAAO,EAAE,kBAAkB,EAAyB,MAAM,oBAAoB,CAAC;AAe/E,SAAS,UAAU,CAAC,KAAkB;IAClC,MAAM,MAAM,GAAG,KAAK,CAAC,YAAkD,CAAC;IACxE,2FAA2F;IAC3F,yEAAyE;IACzE,OAAO,MAAM,EAAE,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,EAAE,IAAI,EAAE,OAAO,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC;AAChG,CAAC;AAED;;;;;;;;;;;;;GAaG;AACH,MAAM,OAAO,mBAAmB;IAAhC;QACa,SAAI,GAAG,sBAAsB,CAAC;IAiD3C,CAAC;IA/CG,KAAK,CAAC,QAAQ,CAAC,KAAkB,EAAE,MAAoB;QACnD,IAAI,CAAC;YACD,MAAM,WAAW,GAAG,MAAwC,CAAC;YAC7D,IAAI,CAAC,WAAW,EAAE,IAAI,EAAE,CAAC;gBACrB,OAAO,EAAE,UAAU,EAAE,IAAI,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,EAAE,OAAO,EAAE,yDAAyD,EAAE,CAAC;YAClI,CAAC;YACD,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,CAAC,CAAC;YACjC,MAAM,QAAQ,GAAG,iBAAiB,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;YAChD,oFAAoF;YACpF,uFAAuF;YACvF,qFAAqF;YACrF,iFAAiF;YACjF,MAAM,UAAU,GAAG,yBAAyB,CACxC,CAAC,KAAK,CAAC,IAAI,EAAE,IAAI,IAAI,MAAM,CAAC,EAAE,WAA2C,EAAE,QAAQ,CAAC,CAAC;YAEzF,OAAO;gBACH,UAAU,EAAE,IAAI,CAAC,IAAI;gBACrB,MAAM,EAAE,UAAU,CAAC,MAAM;gBACzB,KAAK,EAAE,UAAU,CAAC,KAAK;gBACvB,OAAO,EAAE,UAAU,CAAC,MAAM;oBACtB,CAAC,CAAC,YAAY,QAAQ,CAAC,IAAI,SAAS,QAAQ,CAAC,QAAQ,cAAc;oBACnE,CAAC,CAAC,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC;gBACpC,OAAO,EAAE;oBACL,YAAY,EAAE,WAAW,CAAC,IAAI;oBAC9B,YAAY,EAAE,QAAQ,CAAC,IAAI;oBAC3B,QAAQ,EAAE,QAAQ,CAAC,QAAQ;oBAC3B,iBAAiB,EAAE,UAAU,CAAC,iBAAiB;oBAC/C,cAAc,EAAE,UAAU,CAAC,cAAc;oBACzC,aAAa,EAAE,UAAU,CAAC,aAAa;oBACvC,YAAY,EAAE,UAAU,CAAC,YAAY;oBACrC,mBAAmB,EAAE,UAAU,CAAC,mBAAmB;oBACnD,eAAe,EAAE,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;oBACpD,iBAAiB,EAAE,QAAQ,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;oBACxD,mDAAmD;oBACnD,gEAAgE;oBAChE,WAAW,EAAE,QAAQ,CAAC,WAAW,IAAI,KAAK;oBAC1C,cAAc,EAAE,QAAQ,CAAC,aAAa,KAAK,SAAS;oBACpD,kBAAkB,EAAE,QAAQ,CAAC,kBAAkB,IAAI,KAAK;oBACxD,eAAe,EAAE,QAAQ,CAAC,eAAe,IAAI,KAAK;oBAClD,kBAAkB,EAAE,QAAQ,CAAC,kBAAkB;oBAC/C,UAAU,EAAE,QAAQ,CAAC,UAAU;iBAClC;aACJ,CAAC;QACN,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACb,OAAO,EAAE,UAAU,EAAE,IAAI,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,EAAE,OAAO,EAAE,8BAA+B,KAAe,CAAC,OAAO,EAAE,EAAE,CAAC;QACjI,CAAC;IACL,CAAC;CACJ;AAED;;;;;;;;;;;GAWG;AACH,MAAM,OAAO,wBAAwB;IAArC;QACa,SAAI,GAAG,sBAAsB,CAAC;IAe3C,CAAC;IAbG,KAAK,CAAC,QAAQ,CAAC,KAAkB,EAAE,MAAoB;QACnD,IAAI,CAAC;YACD,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,CAAC,CAAC;YACjC,MAAM,QAAQ,GAAG,iBAAiB,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;YAChD,MAAM,MAAM,GAAG,kBAAkB,CAC7B,EAAE,QAAQ,EAAE,QAAQ,EAAE,YAAY,EAAE,MAAM,CAAC,YAAY,EAAE,cAAc,EAAE,MAAM,CAAC,cAAc,EAAE,EAChG,MAA0B,CAC7B,CAAC;YACF,OAAO,EAAE,UAAU,EAAE,IAAI,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,CAAC;QACnI,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACb,OAAO,EAAE,UAAU,EAAE,IAAI,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,EAAE,OAAO,EAAE,qCAAsC,KAAe,CAAC,OAAO,EAAE,EAAE,CAAC;QACxI,CAAC;IACL,CAAC;CACJ"}

package/dist/oracles/TraceSubAgentValidatorOracle.d.ts

@@ -0,0 +1,44 @@
+/**
+ * @fileoverview Sub-agent dispatch oracle — the shipped-but-missing `trace-validate-sub-agents`.
+ * @module @memberjunction/testing-engine
+ */
+import { IOracle } from './IOracle.js';
+import { OracleInput, OracleConfig, OracleResult } from '../types.js';
+/**
+ * Validates which sub-agents an agent run dispatched, and how many iterations it took.
+ *
+ * **This oracle was referenced by shipped metadata before it existed.** Both stock research-agent
+ * tests configure `trace-validate-sub-agents` and weight it at 0.25 and 0.3 respectively; with no
+ * implementation registered, the engine logged `Oracle not found` and a quarter to a third of each
+ * test's weight silently vanished from the score. The tests reported green on the remaining
+ * oracles. That is the defect this closes (test plan §2, gap 5).
+ *
+ * **Traversal.** Sub-agent runs are reached through the **Sub-Agent step's `TargetLogID`**, not
+ * through `AIAgentRun.ParentRunID`. The generated ORM is explicit that ParentRunID "records
+ * parentage but is not a link the tree traverses", and the platform's own run-tree walk uses the
+ * step link — so querying by ParentRunID would find runs the platform does not consider part of
+ * this tree, and miss none that it does. The walk recurses, because a sub-agent may dispatch its
+ * own, and a required agent two levels down was still dispatched by this run.
+ *
+ * Configuration (as the shipped tests already supply it):
+ * - `requiredAgents`: names that must appear
+ * - `forbiddenAgents`: names that must not
+ * - `minIterations`: minimum Prompt steps on the top-level run
+ *
+ * @example
+ * ```json
+ * { "type": "trace-validate-sub-agents", "weight": 0.3,
+ *   "config": { "requiredAgents": ["Database Research Agent", "Web Research Agent"], "minIterations": 1 } }
+ * ```
+ */
+export declare class TraceSubAgentValidatorOracle implements IOracle {
+    readonly type = "trace-validate-sub-agents";
+    /** Guard against a cyclic or pathologically deep run tree; real trees are 2–3 levels. */
+    private static readonly MAX_DEPTH;
+    evaluate(input: OracleInput, config: OracleConfig): Promise<OracleResult>;
+    /** Walks the run tree, collecting dispatched sub-agent names and the parent's iteration count. */
+    private collectFacts;
+    private loadSteps;
+    private loadRun;
+}
+//# sourceMappingURL=TraceSubAgentValidatorOracle.d.ts.map

package/dist/oracles/TraceSubAgentValidatorOracle.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"TraceSubAgentValidatorOracle.d.ts","sourceRoot":"","sources":["../../src/oracles/TraceSubAgentValidatorOracle.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAKH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,WAAW,EAAE,YAAY,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AAGnE;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AACH,qBAAa,4BAA6B,YAAW,OAAO;IACxD,QAAQ,CAAC,IAAI,+BAA+B;IAE5C,yFAAyF;IACzF,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,SAAS,CAAK;IAEhC,QAAQ,CAAC,KAAK,EAAE,WAAW,EAAE,MAAM,EAAE,YAAY,GAAG,OAAO,CAAC,YAAY,CAAC;IAoB/E,kGAAkG;YACpF,YAAY;YA8BZ,SAAS;YAUT,OAAO;CAQxB"}

package/dist/oracles/TraceSubAgentValidatorOracle.js

@@ -0,0 +1,105 @@
+/**
+ * @fileoverview Sub-agent dispatch oracle — the shipped-but-missing `trace-validate-sub-agents`.
+ * @module @memberjunction/testing-engine
+ */
+import { RunView } from '@memberjunction/core';
+import { evaluateSubAgentTrace } from '../eval/subAgentTrace.js';
+/**
+ * Validates which sub-agents an agent run dispatched, and how many iterations it took.
+ *
+ * **This oracle was referenced by shipped metadata before it existed.** Both stock research-agent
+ * tests configure `trace-validate-sub-agents` and weight it at 0.25 and 0.3 respectively; with no
+ * implementation registered, the engine logged `Oracle not found` and a quarter to a third of each
+ * test's weight silently vanished from the score. The tests reported green on the remaining
+ * oracles. That is the defect this closes (test plan §2, gap 5).
+ *
+ * **Traversal.** Sub-agent runs are reached through the **Sub-Agent step's `TargetLogID`**, not
+ * through `AIAgentRun.ParentRunID`. The generated ORM is explicit that ParentRunID "records
+ * parentage but is not a link the tree traverses", and the platform's own run-tree walk uses the
+ * step link — so querying by ParentRunID would find runs the platform does not consider part of
+ * this tree, and miss none that it does. The walk recurses, because a sub-agent may dispatch its
+ * own, and a required agent two levels down was still dispatched by this run.
+ *
+ * Configuration (as the shipped tests already supply it):
+ * - `requiredAgents`: names that must appear
+ * - `forbiddenAgents`: names that must not
+ * - `minIterations`: minimum Prompt steps on the top-level run
+ *
+ * @example
+ * ```json
+ * { "type": "trace-validate-sub-agents", "weight": 0.3,
+ *   "config": { "requiredAgents": ["Database Research Agent", "Web Research Agent"], "minIterations": 1 } }
+ * ```
+ */
+export class TraceSubAgentValidatorOracle {
+    constructor() {
+        this.type = 'trace-validate-sub-agents';
+    }
+    /** Guard against a cyclic or pathologically deep run tree; real trees are 2–3 levels. */
+    static { this.MAX_DEPTH = 6; }
+    async evaluate(input, config) {
+        try {
+            const agentRun = input.targetEntity;
+            if (!agentRun?.ID) {
+                return { oracleType: this.type, passed: false, score: 0, message: 'No agent run entity provided' };
+            }
+            const facts = await this.collectFacts(agentRun.ID, input.contextUser);
+            const result = evaluateSubAgentTrace(facts, config);
+            return { oracleType: this.type, passed: result.passed, score: result.score, message: result.message, details: result.details };
+        }
+        catch (error) {
+            return {
+                oracleType: this.type,
+                passed: false,
+                score: 0,
+                message: `Sub-agent trace validation error: ${error.message}`
+            };
+        }
+    }
+    /** Walks the run tree, collecting dispatched sub-agent names and the parent's iteration count. */
+    async collectFacts(rootRunID, contextUser) {
+        const dispatchedAgents = [];
+        const visited = new Set();
+        const rootSteps = await this.loadSteps(rootRunID, contextUser);
+        // Iterations are the PARENT's prompt steps only — a sub-agent's prompts are its own passes.
+        const iterations = rootSteps.filter((s) => s.StepType === 'Prompt').length;
+        const walk = async (steps, depth) => {
+            if (depth > TraceSubAgentValidatorOracle.MAX_DEPTH) {
+                return;
+            }
+            for (const step of steps) {
+                if (step.StepType !== 'Sub-Agent' || !step.TargetLogID || visited.has(step.TargetLogID)) {
+                    continue;
+                }
+                visited.add(step.TargetLogID);
+                const childRun = await this.loadRun(step.TargetLogID, contextUser);
+                // Fall back to the step name when the child run is unreadable: a dispatch that
+                // happened is a fact worth recording even if the child row cannot be loaded.
+                dispatchedAgents.push(childRun?.Agent?.trim() || step.StepName?.trim() || '(unnamed sub-agent)');
+                if (childRun) {
+                    await walk(await this.loadSteps(childRun.ID, contextUser), depth + 1);
+                }
+            }
+        };
+        await walk(rootSteps, 1);
+        return { dispatchedAgents, iterations };
+    }
+    async loadSteps(agentRunID, contextUser) {
+        const result = await new RunView().RunView({
+            EntityName: 'MJ: AI Agent Run Steps',
+            ExtraFilter: `AgentRunID='${agentRunID}'`,
+            OrderBy: 'StepNumber',
+            ResultType: 'entity_object'
+        }, contextUser);
+        return result.Success ? (result.Results ?? []) : [];
+    }
+    async loadRun(agentRunID, contextUser) {
+        const result = await new RunView().RunView({
+            EntityName: 'MJ: AI Agent Runs',
+            ExtraFilter: `ID='${agentRunID}'`,
+            ResultType: 'entity_object'
+        }, contextUser);
+        return result.Success && result.Results?.length === 1 ? result.Results[0] : null;
+    }
+}
+//# sourceMappingURL=TraceSubAgentValidatorOracle.js.map

package/dist/oracles/TraceSubAgentValidatorOracle.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"TraceSubAgentValidatorOracle.js","sourceRoot":"","sources":["../../src/oracles/TraceSubAgentValidatorOracle.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,EAAE,OAAO,EAAE,MAAM,sBAAsB,CAAC;AAK/C,OAAO,EAAE,qBAAqB,EAAqD,MAAM,uBAAuB,CAAC;AAEjH;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AACH,MAAM,OAAO,4BAA4B;IAAzC;QACa,SAAI,GAAG,2BAA2B,CAAC;IA0EhD,CAAC;IAxEG,yFAAyF;aACjE,cAAS,GAAG,CAAC,AAAJ,CAAK;IAEtC,KAAK,CAAC,QAAQ,CAAC,KAAkB,EAAE,MAAoB;QACnD,IAAI,CAAC;YACD,MAAM,QAAQ,GAAG,KAAK,CAAC,YAA8C,CAAC;YACtE,IAAI,CAAC,QAAQ,EAAE,EAAE,EAAE,CAAC;gBAChB,OAAO,EAAE,UAAU,EAAE,IAAI,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,EAAE,OAAO,EAAE,8BAA8B,EAAE,CAAC;YACvG,CAAC;YAED,MAAM,KAAK,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,EAAE,EAAE,KAAK,CAAC,WAAW,CAAC,CAAC;YACtE,MAAM,MAAM,GAAG,qBAAqB,CAAC,KAAK,EAAE,MAA6B,CAAC,CAAC;YAC3E,OAAO,EAAE,UAAU,EAAE,IAAI,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,CAAC;QACnI,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACb,OAAO;gBACH,UAAU,EAAE,IAAI,CAAC,IAAI;gBACrB,MAAM,EAAE,KAAK;gBACb,KAAK,EAAE,CAAC;gBACR,OAAO,EAAE,qCAAsC,KAAe,CAAC,OAAO,EAAE;aAC3E,CAAC;QACN,CAAC;IACL,CAAC;IAED,kGAAkG;IAC1F,KAAK,CAAC,YAAY,CAAC,SAAiB,EAAE,WAAqB;QAC/D,MAAM,gBAAgB,GAAa,EAAE,CAAC;QACtC,MAAM,OAAO,GAAG,IAAI,GAAG,EAAU,CAAC;QAClC,MAAM,SAAS,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,SAAS,EAAE,WAAW,CAAC,CAAC;QAC/D,4FAA4F;QAC5F,MAAM,UAAU,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,KAAK,QAAQ,CAAC,CAAC,MAAM,CAAC;QAE3E,MAAM,IAAI,GAAG,KAAK,EAAE,KAA+B,EAAE,KAAa,EAAiB,EAAE;YACjF,IAAI,KAAK,GAAG,4BAA4B,CAAC,SAAS,EAAE,CAAC;gBACjD,OAAO;YACX,CAAC;YACD,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;gBACvB,IAAI,IAAI,CAAC,QAAQ,KAAK,WAAW,IAAI,CAAC,IAAI,CAAC,WAAW,IAAI,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,WAAW,CAAC,EAAE,CAAC;oBACtF,SAAS;gBACb,CAAC;gBACD,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;gBAC9B,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,WAAW,EAAE,WAAW,CAAC,CAAC;gBACnE,+EAA+E;gBAC/E,6EAA6E;gBAC7E,gBAAgB,CAAC,IAAI,CAAC,QAAQ,EAAE,KAAK,EAAE,IAAI,EAAE,IAAI,IAAI,CAAC,QAAQ,EAAE,IAAI,EAAE,IAAI,qBAAqB,CAAC,CAAC;gBACjG,IAAI,QAAQ,EAAE,CAAC;oBACX,MAAM,IAAI,CAAC,MAAM,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,EAAE,EAAE,WAAW,CAAC,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC;gBAC1E,CAAC;YACL,CAAC;QACL,CAAC,CAAC;QACF,MAAM,IAAI,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;QAEzB,OAAO,EAAE,gBAAgB,EAAE,UAAU,EAAE,CAAC;IAC5C,CAAC;IAEO,KAAK,CAAC,SAAS,CAAC,UAAkB,EAAE,WAAqB;QAC7D,MAAM,MAAM,GAAG,MAAM,IAAI,OAAO,EAAE,CAAC,OAAO,CAAyB;YAC/D,UAAU,EAAE,wBAAwB;YACpC,WAAW,EAAE,eAAe,UAAU,GAAG;YACzC,OAAO,EAAE,YAAY;YACrB,UAAU,EAAE,eAAe;SAC9B,EAAE,WAAW,CAAC,CAAC;QAChB,OAAO,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;IACxD,CAAC;IAEO,KAAK,CAAC,OAAO,CAAC,UAAkB,EAAE,WAAqB;QAC3D,MAAM,MAAM,GAAG,MAAM,IAAI,OAAO,EAAE,CAAC,OAAO,CAAqB;YAC3D,UAAU,EAAE,mBAAmB;YAC/B,WAAW,EAAE,OAAO,UAAU,GAAG;YACjC,UAAU,EAAE,eAAe;SAC9B,EAAE,WAAW,CAAC,CAAC;QAChB,OAAO,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,OAAO,EAAE,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IACrF,CAAC"}

package/dist/oracles/index.d.ts

@@ -4,6 +4,8 @@
 export * from './IOracle.js';
 export * from './SchemaValidatorOracle.js';
 export * from './TraceValidatorOracle.js';
+export * from './TraceSubAgentValidatorOracle.js';
+export * from './AgentDecisionOracle.js';
 export * from './LLMJudgeOracle.js';
 export * from './ExactMatchOracle.js';
 export * from './SQLValidatorOracle.js';

package/dist/oracles/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/oracles/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,cAAc,WAAW,CAAC;AAC1B,cAAc,yBAAyB,CAAC;AACxC,cAAc,wBAAwB,CAAC;AACvC,cAAc,kBAAkB,CAAC;AACjC,cAAc,oBAAoB,CAAC;AACnC,cAAc,sBAAsB,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/oracles/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,cAAc,WAAW,CAAC;AAC1B,cAAc,yBAAyB,CAAC;AACxC,cAAc,wBAAwB,CAAC;AACvC,cAAc,gCAAgC,CAAC;AAC/C,cAAc,uBAAuB,CAAC;AACtC,cAAc,kBAAkB,CAAC;AACjC,cAAc,oBAAoB,CAAC;AACnC,cAAc,sBAAsB,CAAC"}

package/dist/oracles/index.js

@@ -4,6 +4,8 @@
 export * from './IOracle.js';
 export * from './SchemaValidatorOracle.js';
 export * from './TraceValidatorOracle.js';
+export * from './TraceSubAgentValidatorOracle.js';
+export * from './AgentDecisionOracle.js';
 export * from './LLMJudgeOracle.js';
 export * from './ExactMatchOracle.js';
 export * from './SQLValidatorOracle.js';

package/dist/oracles/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/oracles/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,cAAc,WAAW,CAAC;AAC1B,cAAc,yBAAyB,CAAC;AACxC,cAAc,wBAAwB,CAAC;AACvC,cAAc,kBAAkB,CAAC;AACjC,cAAc,oBAAoB,CAAC;AACnC,cAAc,sBAAsB,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/oracles/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,cAAc,WAAW,CAAC;AAC1B,cAAc,yBAAyB,CAAC;AACxC,cAAc,wBAAwB,CAAC;AACvC,cAAc,gCAAgC,CAAC;AAC/C,cAAc,uBAAuB,CAAC;AACtC,cAAc,kBAAkB,CAAC;AACjC,cAAc,oBAAoB,CAAC;AACnC,cAAc,sBAAsB,CAAC"}

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@memberjunction/testing-engine",
   "type": "module",
-  "version": "6.1.0-edge.5",
+  "version": "6.1.0-edge.7",
   "description": "MemberJunction Testing Framework Engine - Core test execution and evaluation engine supporting multiple test types",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
@@ -17,15 +17,15 @@
     "vitest": "^4.0.18"
   },
   "dependencies": {
-    "@memberjunction/ai": "6.1.0-edge.5",
-    "@memberjunction/ai-agents": "6.1.0-edge.5",
-    "@memberjunction/ai-core-plus": "6.1.0-edge.5",
-    "@memberjunction/ai-prompts": "6.1.0-edge.5",
-    "@memberjunction/aiengine": "6.1.0-edge.5",
-    "@memberjunction/core": "6.1.0-edge.5",
-    "@memberjunction/core-entities": "6.1.0-edge.5",
-    "@memberjunction/global": "6.1.0-edge.5",
-    "@memberjunction/testing-engine-base": "6.1.0-edge.5",
+    "@memberjunction/ai": "6.1.0-edge.7",
+    "@memberjunction/ai-agents": "6.1.0-edge.7",
+    "@memberjunction/ai-core-plus": "6.1.0-edge.7",
+    "@memberjunction/ai-prompts": "6.1.0-edge.7",
+    "@memberjunction/aiengine": "6.1.0-edge.7",
+    "@memberjunction/core": "6.1.0-edge.7",
+    "@memberjunction/core-entities": "6.1.0-edge.7",
+    "@memberjunction/global": "6.1.0-edge.7",
+    "@memberjunction/testing-engine-base": "6.1.0-edge.7",
     "debug": "^4.4.3",
     "rxjs": "^7.8.2",
     "zod": "^3.25.0"