test: verify ambiguity handling across domains

2026-08-02 16:17:37 +01:00
parent 51ce356218
commit b84989b96a
3 changed files with 370 additions and 0 deletions
@@ -0,0 +1,132 @@
+import { describe, expect, it } from "vitest";
+import {
+  formulateQuestion,
+  formulateTieResolutionQuestion,
+} from "@/lib/graph/question-formulator.js";
+import {
+  explainUnknownSelection,
+  selectActiveUnknownCandidate,
+} from "@/lib/graph/utils.js";
+import { ambiguityGeneralisationFixtures } from "@/tests/fixtures/ambiguity-generalisation.js";
+
+function neutraliseUnknownLabels(graph) {
+  let counter = 0;
+  return {
+    ...graph,
+    nodes: graph.nodes.map((node) => {
+      if (node.kind !== "unknown") return { ...node };
+      counter += 1;
+      return {
+        ...node,
+        label: `Unknown ${String.fromCharCode(64 + counter)}`,
+        description: `Unknown factor ${counter}.`,
+      };
+    }),
+  };
+}
+
+function isSingleQuestion(question) {
+  return (question.match(/\?/g) || []).length === 1;
+}
+
+describe("ambiguity generalisation", () => {
+  it("preserves ambiguity across contradiction scenarios without favouring one explanation", () => {
+    const summary = ambiguityGeneralisationFixtures.map((fixture) => {
+      const explanation = explainUnknownSelection(fixture.graph, []);
+      const selection = selectActiveUnknownCandidate(fixture.graph, []);
+      const neutralExplanation = explainUnknownSelection(
+        neutraliseUnknownLabels(fixture.graph),
+        [],
+      );
+      const tieQuestion = formulateTieResolutionQuestion({
+        graph: fixture.graph,
+      });
+      const representativeUnknown = fixture.graph.nodes.find(
+        (node) => node.kind === "unknown",
+      );
+      const fallbackQuestion = formulateQuestion({
+        node: representativeUnknown,
+        graph: fixture.graph,
+      });
+
+      const lowerQuestion = tieQuestion.question.toLowerCase();
+      for (const term of fixture.disallowedQuestionTerms) {
+        expect(lowerQuestion).not.toContain(term.toLowerCase());
+      }
+
+      expect(explanation.status).toBe("ambiguous");
+      expect(selection.status).toBe("ambiguous");
+      expect(selection.selectedNode).toBeNull();
+      expect(explanation.selectedNodeId).toBeNull();
+      expect(explanation.candidates).toHaveLength(2);
+      expect(explanation.summary.selectedReason).toBe(
+        "No justified distinction between leading unknowns.",
+      );
+      expect(explanation.alphabeticalUsedAsReasoning).toBe(false);
+      expect(neutralExplanation.status).toBe("ambiguous");
+      expect(isSingleQuestion(tieQuestion.question)).toBe(true);
+      expect(tieQuestion.question.toLowerCase()).not.toContain(" and ");
+      expect(tieQuestion.question.toLowerCase()).not.toContain(" or ");
+
+      return {
+        scenario: fixture.scenario,
+        candidateCount: explanation.candidates.length,
+        ambiguityStatus: explanation.status,
+        tieReason: explanation.summary.selectedReason,
+        investigationStrategy: tieQuestion.strategy,
+        question: tieQuestion.question,
+        explanationFavoured: explanation.selectedNodeId !== null,
+      };
+    });
+
+    expect(summary).toMatchInlineSnapshot(`
+      [
+        {
+          "ambiguityStatus": "ambiguous",
+          "candidateCount": 2,
+          "explanationFavoured": false,
+          "investigationStrategy": null,
+          "question": "What changed during the period that could explain why Revenue increased by 18%, but cash in the bank fell over the same period?",
+          "scenario": "Revenue increased by 18%, but cash in the bank fell over the same period.",
+          "tieReason": "No justified distinction between leading unknowns.",
+        },
+        {
+          "ambiguityStatus": "ambiguous",
+          "candidateCount": 2,
+          "explanationFavoured": false,
+          "investigationStrategy": null,
+          "question": "What changed during the period that could explain why Customer satisfaction scores increased, but complaints also increased?",
+          "scenario": "Customer satisfaction scores increased, but complaints also increased.",
+          "tieReason": "No justified distinction between leading unknowns.",
+        },
+        {
+          "ambiguityStatus": "ambiguous",
+          "candidateCount": 2,
+          "explanationFavoured": false,
+          "investigationStrategy": null,
+          "question": "What changed during the period that could explain why Average delivery time decreased by 25%, but order cancellations increased?",
+          "scenario": "Average delivery time decreased by 25%, but order cancellations increased.",
+          "tieReason": "No justified distinction between leading unknowns.",
+        },
+        {
+          "ambiguityStatus": "ambiguous",
+          "candidateCount": 2,
+          "explanationFavoured": false,
+          "investigationStrategy": null,
+          "question": "What changed during the period that could explain why Website traffic doubled, but sales remained unchanged?",
+          "scenario": "Website traffic doubled, but sales remained unchanged.",
+          "tieReason": "No justified distinction between leading unknowns.",
+        },
+        {
+          "ambiguityStatus": "ambiguous",
+          "candidateCount": 2,
+          "explanationFavoured": false,
+          "investigationStrategy": null,
+          "question": "What changed during the period that could explain why Production output increased by 30%, but quality defects also increased?",
+          "scenario": "Production output increased by 30%, but quality defects also increased.",
+          "tieReason": "No justified distinction between leading unknowns.",
+        },
+      ]
+    `);
+  });
+});