confidence-engine/tests/graph/episode-reasoning-input.test.js

import { describe, expect, it } from "vitest";
import { buildEpisodeAwareGraphPrompt } from "@/lib/graph/prompt-builder-episode.js";
import { buildGraphUpdatePrompt } from "@/lib/graph/prompt-builder.js";
import { makeEdge, makeGraph, makeNode } from "@/lib/graph/schema.js";

// ── Test fixture: multi-turn prepared episode ───────────────

function makeEpisodeFixture() {
  const situationGraph = makeGraph({
    centralStatement: "Market share declining in core segment",
    nodes: [
      makeNode({
        id: "core-shrink-unknown",
        label: "Core segment shrink rate",
        description: "Need to determine if core market is genuinely shrinking or if this is measurement artifact",
        kind: "unknown",
        status: "unknown",
        confidence: "medium",
      }),
      makeNode({
        id: "competitor-move-obs",
        label: "Competitor pricing move",
        description: "Competitor X reduced prices by 15%",
        kind: "observation",
        status: "supported",
        confidence: "high",
      }),
    ],
    edges: [
      makeEdge({
        id: "e1",
        fromNodeId: "competitor-move-obs",
        toNodeId: "core-shrink-unknown",
        relationship: "supports",
        confidence: "medium",
        description: "Observation informs unknown",
      }),
    ],
    activeUnknownNodeId: "core-shrink-unknown",
    resolvedNodeIds: [],
    currentSummary: "Nodes: 1 observation, 1 unknown | Edges: 1 total | Unknowns: 1 unresolved",
  });

  const turns = [
    {
      contributionId: "contrib-a01",
      sequence: 1,
      question: "Has the core market volume actually decreased, or is it a measurement artifact?",
      answer: "Our internal data shows core volume dropped from 12,400 to 11,800 units over Q3.",
    },
    {
      contributionId: "contrib-a02",
      sequence: 2,
      question: "Does Competitor X's pricing explain the full decline?",
      answer: "Competitor X's price cut correlates with a 40% shift in our B2B accounts. We estimate ~75% of the decline maps to this competitive move.",
    },
    {
      contributionId: "contrib-b01",
      sequence: 3,
      question: "Is core quality perception affected?",
      answer: "We have no evidence of quality degradation — returns rate is stable at 0.3%.",
    },
  ];

  const eligibleCanonicalFindings = [
    {
      findingId: "find-core-drop",
      contributionId: "contrib-a01",
      proposition: "Core market volume declined from 12,400 to 11,800 units in Q3",
      sourceObservation: "Raw internal data shows drop from 12,400 to 11,800 units",
      endorsement: null, // working premise — not explicit agreement
    },
    {
      findingId: "find-competitive-shift",
      contributionId: "contrib-a02",
      proposition: "Competitor X pricing accounts for ~75% of core decline via B2B account shift",
      sourceObservation: "Model inference linking price to volume",
      endorsement: "agree", // explicitly endorsed by user
    },
    {
      findingId: "find-quality-stable",
      contributionId: "contrib-b01",
      proposition: "Core quality perception remains stable — returns rate unchanged at 0.3%",
      sourceObservation: "Original observation from Q2 analysis",
      endorsement: null, // working premise — not explicit agreement
    },
  ];

  const excludedFindingProvenance = [
    {
      findingId: "find-excluded-supply",
      contributionId: "contrib-a01",
      proposition: "SUPPLY-chain disruption partially contributed to volume loss (excluded from current reasoning)",
      sourceObservation: "Initial hypothesis later determined irrelevant",
      disposition: "not_relevant",
    },
  ];

  return { situationGraph, targetNodeId: "core-shrink-unknown", turns, eligibleCanonicalFindings, excludedFindingProvenance };
}

// ── Helper: extract provider-active content (everything before excluded section) ──

function getProviderContent(prompt) {
  const excludedSection = prompt.indexOf("## Excluded Findings");
  return excludedSection > 0 ? prompt.slice(0, excludedSection) : prompt;
}

// ====================================================================
// Case 1: Ordered multi-turn evidence
// ====================================================================

describe("Episode reasoning input — ordered multi-turn evidence", () => {
  const prompt = buildEpisodeAwareGraphPrompt({ episode: makeEpisodeFixture() });

  it("includes all three turns in sequence order", () => {
    expect(prompt).toContain("Turn 1 [contributionId: contrib-a01]");
    expect(prompt).toContain("Turn 2 [contributionId: contrib-a02]");
    expect(prompt).toContain("Turn 3 [contributionId: contrib-b01]");
    // Verify ordering: Turn 1 appears before Turn 2 before Turn 3
    const t1 = prompt.indexOf("Turn 1 [contributionId: contrib-a01]");
    const t2 = prompt.indexOf("Turn 2 [contributionId: contrib-a02]");
    const t3 = prompt.indexOf("Turn 3 [contributionId: contrib-b01]");
    expect(t1).toBeLessThan(t2);
    expect(t2).toBeLessThan(t3);
  });

  it("includes verbatim answers from each turn in provider-active content", () => {
    const providerContent = getProviderContent(prompt);
    expect(providerContent).toContain("core volume dropped from 12,400 to 11,800 units over Q3");
    expect(providerContent).toContain("40% shift in our B2B accounts");
    expect(providerContent).toContain("returns rate is stable at 0.3%");
  });

  it("includes ordered investigation turns header", () => {
    expect(prompt).toContain("### Ordered Investigation Turns (ordered context)");
  });
});

// ====================================================================
// Case 2: Raw evidence distinction
// ====================================================================

describe("Episode reasoning input — raw evidence distinction", () => {
  const prompt = buildEpisodeAwareGraphPrompt({ episode: makeEpisodeFixture() });
  const providerContent = getProviderContent(prompt);

  it("clearly labels answers as verbatim (raw evidence)", () => {
    expect(providerContent).toContain("Verbatim Answer");
    expect(providerContent).toContain("not raw user statements");
  });

  it("labels eligible findings as canonical propositions derived from evidence", () => {
    expect(providerContent).toContain("Eligible Canonical Findings Derived During Investigation");
    expect(providerContent).toContain("canonical propositions derived from the investigation turns above");
  });

  it("does NOT conflate Finding proposition with raw user statement", () => {
    // The prompt must distinguish, not blur:
    expect(providerContent).not.toContain("user said");
    expect(providerContent).not.toContain("raw evidence" + "prop"); // would blur the distinction
    // And should clearly separate the two concepts:
    expect(providerContent).toContain("Proposition:");
    expect(providerContent).toContain("Verbatim Answer:");
  });

  it("marks source of each proposition as derived, not raw", () => {
    expect(providerContent).toContain("derived from the investigation");
  });
});

// ====================================================================
// Case 3: Null versus agree distinction
// ====================================================================

describe("Episode reasoning input — null versus agree endorsement", () => {
  const prompt = buildEpisodeAwareGraphPrompt({ episode: makeEpisodeFixture() });
  const providerContent = getProviderContent(prompt);

  it("null is labeled as working premise, not explicit agreement", () => {
    // Both findings with null endorsement must show the working premise label
    expect(providerContent).toContain("null (working premise, not explicit agreement)");
  });

  it("agree is labeled as explicitly endorsed proposition", () => {
    expect(providerContent).toContain("agree (explicitly endorsed proposition)");
  });

  it("both semantics are distinguishable in provider-active content", () => {
    // The two distinct labels must appear
    const nullLabel = "null (working premise, not explicit agreement)";
    const agreeLabel = "agree (explicitly endorsed proposition)";
    expect(providerContent).toContain(nullLabel);
    expect(providerContent).toContain(agreeLabel);
    // And the endorsement semantics section exists:
    expect(providerContent).toContain("### Endorsement Semantics");
  });
});

// ====================================================================
// Case 4: Excluded evidence absent from provider-active content
// ====================================================================

describe("Episode reasoning input — excluded evidence absent", () => {
  const prompt = buildEpisodeAwareGraphPrompt({ episode: makeEpisodeFixture() });
  const providerContent = getProviderContent(prompt);

  it("excluded proposition does NOT appear in provider-active content", () => {
    // The excluded finding's proposition must be absent from the provider-visible section
    expect(providerContent).not.toContain("SUPPLY-chain disruption");
  });

  it("prepared episode retains excluded provenance (data layer)", () => {
    // Excluded findings are preserved in the prepared episode contract,
    // they just don't appear in provider-active reasoning content.
    const episode = makeEpisodeFixture();
    expect(episode.excludedFindingProvenance).toHaveLength(1);
    expect(episode.excludedFindingProvenance[0].proposition).toContain("SUPPLY-chain disruption");
  });

  it("excluded proposition absent from both prompt and provider content", () => {
    // Since excluded findings are intentionally NOT serialized into the prompt,
    // the excluded proposition is absent everywhere in the generated prompt.
    expect(prompt).not.toContain("SUPPLY-chain disruption");
    const providerContent = getProviderContent(prompt);
    expect(providerContent).not.toContain("SUPPLY-chain disruption");
  });
});

// ====================================================================
// Case 5: Corrected proposition — provider receives corrected form
// ====================================================================

describe("Episode reasoning input — corrected proposition", () => {
  function makeCorrectedEpisode() {
    const situationGraph = makeGraph({
      centralStatement: "Baseline scenario",
      nodes: [makeNode({ id: "a-unknown", label: "A Unknown", kind: "unknown", status: "unknown" })],
      edges: [],
      activeUnknownNodeId: "a-unknown",
      resolvedNodeIds: [],
      currentSummary: "Test",
    });

    return {
      situationGraph,
      targetNodeId: "a-unknown",
      turns: [
        { contributionId: "contrib-c01", sequence: 1, question: "Is A true?", answer: "Yes." },
      ],
      eligibleCanonicalFindings: [
        {
          findingId: "find-corrected",
          contributionId: "contrib-c01",
          proposition: "Corrected canonical text — user fixed the observation",
          sourceObservation: "Original model observation that was incorrect",
          endorsement: null,
        },
      ],
      excludedFindingProvenance: [],
    };
  }

  const prompt = buildEpisodeAwareGraphPrompt({ episode: makeCorrectedEpisode() });
  const providerContent = getProviderContent(prompt);

  it("provider-active evidence uses the corrected canonical proposition", () => {
    expect(providerContent).toContain("Corrected canonical text — user fixed the observation");
  });

  it("sourceObservation does NOT need to appear in provider-active content", () => {
    // The prompt builder intentionally omits sourceObservation from active reasoning
    // because corrected proposition is the current canonical authority.
    expect(providerContent).not.toContain("Original model observation that was incorrect");
  });

  it("corrected Finding appears under eligible findings, not excluded", () => {
    expect(prompt).toContain("## Eligible Canonical Findings Derived During Investigation");
  });
});

// ====================================================================
// Case 6: Existing GraphUpdateProposal contract still targeted
// ====================================================================

describe("Episode reasoning input — proposal contract preserved", () => {
  const prompt = buildEpisodeAwareGraphPrompt({ episode: makeEpisodeFixture() });

  it("required JSON fields are present", () => {
    expect(prompt).toContain("addedNodes");
    expect(prompt).toContain("updatedNodes");
    expect(prompt).toContain("addedEdges");
    expect(prompt).toContain("removedEdgeIds");
    expect(prompt).toContain("resolvedUnknownNodeIds");
    expect(prompt).toContain("affectedNodeIds");
    expect(prompt).toContain("selectedQuestion");
    expect(prompt).toContain("answerMeaning");
    expect(prompt).toContain("structuralActionRequired");
  });

  it("enum values present", () => {
    expect(prompt).toContain("observation | reported_claim | metric | state | transition | relationship | assumption | unknown | conclusion | option");
    expect(prompt).toContain("known | unknown | provisional | supported | weakened | contradicted | resolved");
  });

  it("proposal rules are intact", () => {
    expect(prompt).toContain("Propose changes only. Never return a replacement graph.");
    expect(prompt).toContain("answerMeaning.userSupportedMeaning must state only what the user's answer directly supports.");
    expect(prompt).toContain("MUST express its effect through structural mutation");
  });

  it("Decision Sufficiency Rule is present", () => {
    expect(prompt).toContain("## Decision Sufficiency Rule");
  });

  it("output contract reminder is present", () => {
    expect(prompt).toContain("Return one JSON object only");
    expect(prompt).toContain("Never include a full graph");
  });
});

// ====================================================================
// Case 7: Single-turn regression — existing path intact
// ====================================================================

describe("Episode reasoning input — single-turn regression", () => {
  it("buildGraphUpdatePrompt still accepts previousQuestion/answer and produces same structure", () => {
    const prompt = buildGraphUpdatePrompt({
      situationGraph: makeGraph({
        centralStatement: "test",
        nodes: [makeNode({ id: "n1", label: "A", kind: "unknown", status: "unknown" })],
        edges: [],
        activeUnknownNodeId: "n1",
        resolvedNodeIds: [],
        currentSummary: "test graph",
      }),
      previousQuestion: "What is A?",
      answer: "It's B.",
    });

    // Essential single-turn sections must still be present:
    expect(prompt).toContain("## Previous Selected Question");
    expect(prompt).toContain("What is A?");
    expect(prompt).toContain("## User Answer");
    expect(prompt).toContain("It's B.");

    // GraphUpdateProposal contract fields must still be present:
    expect(prompt).toContain("addedNodes");
    expect(prompt).toContain("updatedNodes");
    expect(prompt).toContain("answerMeaning");

    // Episode-aware sections must NOT appear in single-turn path:
    expect(prompt).not.toContain("## Ordered Investigation Turns");
    expect(prompt).not.toContain("## Eligible Canonical Findings Derived During Investigation");
    expect(prompt).not.toContain("Completed Focused Investigation Evidence");
  });

  it("buildGraphUpdatePrompt still contains existing rules", () => {
    const prompt = buildGraphUpdatePrompt({
      situationGraph: makeGraph({
        centralStatement: "test",
        nodes: [makeNode({ id: "n1", label: "A", kind: "unknown", status: "unknown" })],
        edges: [],
        activeUnknownNodeId: "n1",
        resolvedNodeIds: [],
        currentSummary: "test graph",
      }),
      previousQuestion: "Q?",
      answer: "A.",
    });

    // These are established rules that must survive unchanged:
    expect(prompt).toContain("MUST express its effect through structural mutation");
    expect(prompt).toContain("rule #6");
    expect(prompt).toContain("Do not add duplicate unknowns");
  });
});