nodebench-mcp 3.0.0 → 3.0.1

package/dist/__tests__/forecastingDogfood.test.js

@@ -1,284 +0,0 @@
-/**
- * Forecasting OS — Integration Tests (Dogfood)
- *
- * Full lifecycle tests: create → evidence → update → resolve → track record → calibration.
- * Uses MCP tools directly (SQLite-backed, no Convex dependency).
- *
- * Run: npx vitest run src/__tests__/forecastingDogfood.test.ts
- */
-import { describe, it, expect, beforeAll } from "vitest";
-import { forecastingTools } from "../tools/forecastingTools.js";
-// ─── Helpers ────────────────────────────────────────────────────────────────
-const toolMap = new Map(forecastingTools.map((t) => [t.name, t]));
-async function callTool(name, args) {
-    const tool = toolMap.get(name);
-    if (!tool)
-        throw new Error(`Tool ${name} not found`);
-    const result = await tool.handler(args);
-    const text = result.find((r) => r.type === "text")?.text;
-    if (!text)
-        throw new Error(`Tool ${name} returned no text`);
-    return JSON.parse(text);
-}
-// ─── Tool Structure ─────────────────────────────────────────────────────────
-describe("Forecasting tools: structure", () => {
-    it("should have 9 tools", () => {
-        expect(forecastingTools.length).toBe(9);
-    });
-    it("every tool has name, description, inputSchema, handler", () => {
-        for (const tool of forecastingTools) {
-            expect(tool.name).toBeTruthy();
-            expect(tool.description).toBeTruthy();
-            expect(tool.inputSchema).toBeDefined();
-            expect(typeof tool.handler).toBe("function");
-        }
-    });
-    it("tool names match expected list", () => {
-        const names = forecastingTools.map((t) => t.name).sort();
-        expect(names).toEqual([
-            "add_forecast_evidence",
-            "compute_calibration",
-            "create_forecast",
-            "get_active_forecasts",
-            "get_forecast_chain",
-            "get_forecast_evidence",
-            "get_forecast_track_record",
-            "resolve_forecast",
-            "update_forecast_probability",
-        ]);
-    });
-});
-// ─── Full Lifecycle ─────────────────────────────────────────────────────────
-describe("Forecasting lifecycle", () => {
-    let forecastId;
-    it("create_forecast — creates a binary forecast", async () => {
-        const result = await callTool("create_forecast", {
-            question: "Will GPT-5 be released by 2026-12-31?",
-            forecastType: "binary",
-            resolutionDate: "2026-12-31",
-            resolutionCriteria: "OpenAI announces GPT-5 on official blog or press release",
-            probability: 0.5,
-            baseRate: 0.6,
-            refreshFrequency: "weekly",
-            topDrivers: ["Historical 18-month release cadence", "Sam Altman interview hints"],
-            topCounterarguments: ["No official roadmap published"],
-            tags: ["ai_tech"],
-        });
-        expect(result.forecastId).toBeTruthy();
-        expect(result.status).toBe("active");
-        expect(result.probability).toBe(0.5);
-        forecastId = result.forecastId;
-    });
-    it("add_forecast_evidence — adds supporting evidence", async () => {
-        const result = await callTool("add_forecast_evidence", {
-            forecastId,
-            sourceUrl: "https://example.com/sam-altman-interview",
-            sourceTitle: "Sam Altman Interview on AI Progress",
-            sourceType: "news",
-            excerpt: "Altman hints at a major model release in the second half of 2026",
-            signal: "supporting",
-            impactOnProbability: 0.1,
-        });
-        expect(result.evidenceId).toBeTruthy();
-        expect(result.signal).toBe("supporting");
-    });
-    it("add_forecast_evidence — adds disconfirming evidence", async () => {
-        const result = await callTool("add_forecast_evidence", {
-            forecastId,
-            sourceUrl: "https://example.com/compute-shortage",
-            sourceTitle: "GPU Shortage Report Q1 2026",
-            sourceType: "filing",
-            excerpt: "Major cloud providers report 40% compute capacity shortfall for large model training",
-            signal: "disconfirming",
-        });
-        expect(result.evidenceId).toBeTruthy();
-        expect(result.signal).toBe("disconfirming");
-    });
-    it("add_forecast_evidence — deduplicates by URL", async () => {
-        const result = await callTool("add_forecast_evidence", {
-            forecastId,
-            sourceUrl: "https://example.com/sam-altman-interview",
-            sourceTitle: "Duplicate",
-            sourceType: "news",
-            excerpt: "Duplicate entry",
-            signal: "supporting",
-        });
-        expect(result.error).toContain("already exists");
-    });
-    it("get_forecast_evidence — returns evidence", async () => {
-        const result = await callTool("get_forecast_evidence", {
-            forecastId,
-        });
-        expect(result.count).toBe(2);
-        expect(result.evidence.length).toBe(2);
-    });
-    it("get_forecast_evidence — filters by signal", async () => {
-        const result = await callTool("get_forecast_evidence", {
-            forecastId,
-            signal: "supporting",
-        });
-        expect(result.count).toBe(1);
-    });
-    it("update_forecast_probability — updates with reasoning", async () => {
-        const result = await callTool("update_forecast_probability", {
-            forecastId,
-            probability: 0.65,
-            topDrivers: [
-                "Sam Altman interview hint",
-                "Historical 18-month cadence",
-                "Competitor pressure from Google Gemini",
-            ],
-            topCounterarguments: [
-                "GPU shortage may delay training",
-                "No official roadmap",
-            ],
-            reasoning: "Soft leadership signal + historical pattern outweigh compute concerns",
-        });
-        expect(result.previousProbability).toBe(0.5);
-        expect(result.newProbability).toBe(0.65);
-        expect(result.diff).toBe("50% → 65% (+15pp)");
-    });
-    it("update_forecast_probability — validates range", async () => {
-        const result = await callTool("update_forecast_probability", {
-            forecastId,
-            probability: 1.5,
-            reasoning: "Invalid",
-        });
-        expect(result.error).toContain("between 0 and 1");
-    });
-    it("get_forecast_chain — returns full audit trail", async () => {
-        const result = await callTool("get_forecast_chain", {
-            forecastId,
-        });
-        expect(result.forecast).toBeTruthy();
-        expect(result.evidence.length).toBe(2);
-        expect(result.updateHistory.length).toBe(1);
-        expect(result.resolution).toBeNull();
-        expect(result.summary.evidenceCount).toBe(2);
-        expect(result.summary.updateCount).toBe(1);
-        expect(result.summary.isResolved).toBe(false);
-    });
-    it("get_active_forecasts — lists active forecasts", async () => {
-        const result = await callTool("get_active_forecasts", {});
-        expect(result.count).toBeGreaterThanOrEqual(1);
-        const forecasts = result.forecasts;
-        const found = forecasts.find((f) => f.id === forecastId);
-        expect(found).toBeTruthy();
-        expect(found?.status).toBe("active");
-    });
-    it("resolve_forecast — resolves with Brier score", async () => {
-        const result = await callTool("resolve_forecast", {
-            forecastId,
-            outcome: "yes",
-            resolutionNotes: "GPT-5 announced on 2026-11-15 via OpenAI blog post",
-            resolutionSourceUrl: "https://openai.com/gpt-5-announcement",
-        });
-        expect(result.status).toBe("resolved");
-        expect(result.outcome).toBe("yes");
-        // Brier: (0.65 - 1)^2 = 0.1225
-        expect(result.brierScore).toBeCloseTo(0.1225, 3);
-        // Log: -log(0.65) ≈ 0.431
-        expect(result.logScore).toBeCloseTo(0.431, 2);
-    });
-    it("resolve_forecast — cannot resolve twice", async () => {
-        const result = await callTool("resolve_forecast", {
-            forecastId,
-            outcome: "no",
-            resolutionNotes: "Already resolved",
-        });
-        expect(result.error).toContain("already resolved");
-    });
-    it("get_forecast_track_record — shows Brier aggregate", async () => {
-        const result = await callTool("get_forecast_track_record", {});
-        expect(result.scoredCount).toBeGreaterThanOrEqual(1);
-        // SQLite persists between test runs, so overallBrier is average across ALL
-        // resolved forecasts (not just this run). Use a reasonable bound instead.
-        expect(result.overallBrier).toBeGreaterThan(0);
-        expect(result.overallBrier).toBeLessThan(0.5);
-    });
-});
-// ─── Multi-Forecast Calibration ─────────────────────────────────────────────
-describe("Forecasting calibration", () => {
-    beforeAll(async () => {
-        // Create and resolve 5 forecasts with known outcomes for calibration
-        const scenarios = [
-            { probability: 0.9, outcome: "yes" }, // Brier: 0.01
-            { probability: 0.8, outcome: "yes" }, // Brier: 0.04
-            { probability: 0.3, outcome: "no" }, // Brier: 0.09
-            { probability: 0.1, outcome: "no" }, // Brier: 0.01
-            { probability: 0.6, outcome: "yes" }, // Brier: 0.16
-        ];
-        for (let i = 0; i < scenarios.length; i++) {
-            const create = await callTool("create_forecast", {
-                question: `Calibration test forecast ${i + 1}?`,
-                resolutionDate: "2026-01-01",
-                resolutionCriteria: `Test criteria ${i + 1}`,
-                probability: scenarios[i].probability,
-                tags: ["test_calibration"],
-            });
-            await callTool("resolve_forecast", {
-                forecastId: create.forecastId,
-                outcome: scenarios[i].outcome,
-                resolutionNotes: `Test resolution ${i + 1}`,
-            });
-        }
-    });
-    it("compute_calibration — returns 10 bins", async () => {
-        const result = await callTool("compute_calibration", {});
-        expect(result.bins).toBeTruthy();
-        expect(result.bins.length).toBe(10);
-        expect(result.overallBrier).toBeTruthy();
-        expect(typeof result.overallBrier).toBe("number");
-        expect(result.forecastCount).toBeGreaterThanOrEqual(5);
-    });
-    it("get_forecast_track_record — aggregate includes all resolved", async () => {
-        const result = await callTool("get_forecast_track_record", {});
-        // At least 6 resolved (1 from lifecycle + 5 from calibration)
-        expect(result.scoredCount).toBeGreaterThanOrEqual(6);
-        // Average Brier should be reasonable
-        expect(result.overallBrier).toBeLessThan(0.25);
-    });
-});
-// ─── Edge Cases ─────────────────────────────────────────────────────────────
-describe("Forecasting edge cases", () => {
-    it("create_forecast — rejects invalid probability", async () => {
-        const result = await callTool("create_forecast", {
-            question: "Invalid prob test",
-            resolutionDate: "2026-12-31",
-            resolutionCriteria: "Test",
-            probability: -0.1,
-        });
-        expect(result.error).toContain("between 0 and 1");
-    });
-    it("resolve_forecast — ambiguous outcome excluded from scoring", async () => {
-        const create = await callTool("create_forecast", {
-            question: "Ambiguous resolution test?",
-            resolutionDate: "2026-12-31",
-            resolutionCriteria: "Test",
-            probability: 0.7,
-        });
-        const resolve = await callTool("resolve_forecast", {
-            forecastId: create.forecastId,
-            outcome: "ambiguous",
-            resolutionNotes: "Resolution criteria were unclear",
-        });
-        expect(resolve.brierScore).toBeNull();
-        expect(resolve.logScore).toBeNull();
-    });
-    it("get_forecast_chain — returns error for nonexistent forecast", async () => {
-        const result = await callTool("get_forecast_chain", {
-            forecastId: "nonexistent_id",
-        });
-        expect(result.error).toContain("not found");
-    });
-    it("get_active_forecasts — filters by tags", async () => {
-        const result = await callTool("get_active_forecasts", {
-            tags: ["test_calibration"],
-        });
-        // All calibration forecasts are resolved, so none should be active
-        const forecasts = result.forecasts;
-        const calibration = forecasts.filter((f) => f.tags.includes("test_calibration"));
-        expect(calibration.length).toBe(0);
-    });
-});
-//# sourceMappingURL=forecastingDogfood.test.js.map

package/dist/__tests__/forecastingDogfood.test.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"forecastingDogfood.test.js","sourceRoot":"","sources":["../../src/__tests__/forecastingDogfood.test.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,QAAQ,CAAC;AACzD,OAAO,EAAE,gBAAgB,EAAE,MAAM,8BAA8B,CAAC;AAGhE,+EAA+E;AAE/E,MAAM,OAAO,GAAG,IAAI,GAAG,CACrB,gBAAgB,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CACzC,CAAC;AAEF,KAAK,UAAU,QAAQ,CACrB,IAAY,EACZ,IAA6B;IAE7B,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IAC/B,IAAI,CAAC,IAAI;QAAE,MAAM,IAAI,KAAK,CAAC,QAAQ,IAAI,YAAY,CAAC,CAAC;IACrD,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IACxC,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,EAAE,IAAI,CAAC;IACzD,IAAI,CAAC,IAAI;QAAE,MAAM,IAAI,KAAK,CAAC,QAAQ,IAAI,mBAAmB,CAAC,CAAC;IAC5D,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED,+EAA+E;AAE/E,QAAQ,CAAC,8BAA8B,EAAE,GAAG,EAAE;IAC5C,EAAE,CAAC,qBAAqB,EAAE,GAAG,EAAE;QAC7B,MAAM,CAAC,gBAAgB,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAC1C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,wDAAwD,EAAE,GAAG,EAAE;QAChE,KAAK,MAAM,IAAI,IAAI,gBAAgB,EAAE,CAAC;YACpC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,CAAC;YAC/B,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,UAAU,EAAE,CAAC;YACtC,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,WAAW,EAAE,CAAC;YACvC,MAAM,CAAC,OAAO,IAAI,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;QAC/C,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gCAAgC,EAAE,GAAG,EAAE;QACxC,MAAM,KAAK,GAAG,gBAAgB,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,CAAC;QACzD,MAAM,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC;YACpB,uBAAuB;YACvB,qBAAqB;YACrB,iBAAiB;YACjB,sBAAsB;YACtB,oBAAoB;YACpB,uBAAuB;YACvB,2BAA2B;YAC3B,kBAAkB;YAClB,6BAA6B;SAC9B,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,+EAA+E;AAE/E,QAAQ,CAAC,uBAAuB,EAAE,GAAG,EAAE;IACrC,IAAI,UAAkB,CAAC;IAEvB,EAAE,CAAC,6CAA6C,EAAE,KAAK,IAAI,EAAE;QAC3D,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,iBAAiB,EAAE;YAC/C,QAAQ,EAAE,uCAAuC;YACjD,YAAY,EAAE,QAAQ;YACtB,cAAc,EAAE,YAAY;YAC5B,kBAAkB,EAAE,0DAA0D;YAC9E,WAAW,EAAE,GAAG;YAChB,QAAQ,EAAE,GAAG;YACb,gBAAgB,EAAE,QAAQ;YAC1B,UAAU,EAAE,CAAC,qCAAqC,EAAE,4BAA4B,CAAC;YACjF,mBAAmB,EAAE,CAAC,+BAA+B,CAAC;YACtD,IAAI,EAAE,CAAC,SAAS,CAAC;SAClB,CAAC,CAAC;QAEH,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,UAAU,EAAE,CAAC;QACvC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QACrC,MAAM,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACrC,UAAU,GAAG,MAAM,CAAC,UAAoB,CAAC;IAC3C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,kDAAkD,EAAE,KAAK,IAAI,EAAE;QAChE,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,uBAAuB,EAAE;YACrD,UAAU;YACV,SAAS,EAAE,0CAA0C;YACrD,WAAW,EAAE,qCAAqC;YAClD,UAAU,EAAE,MAAM;YAClB,OAAO,EAAE,kEAAkE;YAC3E,MAAM,EAAE,YAAY;YACpB,mBAAmB,EAAE,GAAG;SACzB,CAAC,CAAC;QAEH,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,UAAU,EAAE,CAAC;QACvC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;IAC3C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,qDAAqD,EAAE,KAAK,IAAI,EAAE;QACnE,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,uBAAuB,EAAE;YACrD,UAAU;YACV,SAAS,EAAE,sCAAsC;YACjD,WAAW,EAAE,6BAA6B;YAC1C,UAAU,EAAE,QAAQ;YACpB,OAAO,EAAE,sFAAsF;YAC/F,MAAM,EAAE,eAAe;SACxB,CAAC,CAAC;QAEH,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,UAAU,EAAE,CAAC;QACvC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;IAC9C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,6CAA6C,EAAE,KAAK,IAAI,EAAE;QAC3D,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,uBAAuB,EAAE;YACrD,UAAU;YACV,SAAS,EAAE,0CAA0C;YACrD,WAAW,EAAE,WAAW;YACxB,UAAU,EAAE,MAAM;YAClB,OAAO,EAAE,iBAAiB;YAC1B,MAAM,EAAE,YAAY;SACrB,CAAC,CAAC;QAEH,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,SAAS,CAAC,gBAAgB,CAAC,CAAC;IACnD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,0CAA0C,EAAE,KAAK,IAAI,EAAE;QACxD,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,uBAAuB,EAAE;YACrD,UAAU;SACX,CAAC,CAAC;QAEH,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC7B,MAAM,CAAE,MAAM,CAAC,QAAsB,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACxD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,2CAA2C,EAAE,KAAK,IAAI,EAAE;QACzD,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,uBAAuB,EAAE;YACrD,UAAU;YACV,MAAM,EAAE,YAAY;SACrB,CAAC,CAAC;QAEH,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAC/B,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,sDAAsD,EAAE,KAAK,IAAI,EAAE;QACpE,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,6BAA6B,EAAE;YAC3D,UAAU;YACV,WAAW,EAAE,IAAI;YACjB,UAAU,EAAE;gBACV,2BAA2B;gBAC3B,6BAA6B;gBAC7B,wCAAwC;aACzC;YACD,mBAAmB,EAAE;gBACnB,iCAAiC;gBACjC,qBAAqB;aACtB;YACD,SAAS,EAAE,uEAAuE;SACnF,CAAC,CAAC;QAEH,MAAM,CAAC,MAAM,CAAC,mBAAmB,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAC7C,MAAM,CAAC,MAAM,CAAC,cAAc,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACzC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,mBAAmB,CAAC,CAAC;IAChD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,+CAA+C,EAAE,KAAK,IAAI,EAAE;QAC7D,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,6BAA6B,EAAE;YAC3D,UAAU;YACV,WAAW,EAAE,GAAG;YAChB,SAAS,EAAE,SAAS;SACrB,CAAC,CAAC;QAEH,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,SAAS,CAAC,iBAAiB,CAAC,CAAC;IACpD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,+CAA+C,EAAE,KAAK,IAAI,EAAE;QAC7D,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,oBAAoB,EAAE;YAClD,UAAU;SACX,CAAC,CAAC;QAEH,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,UAAU,EAAE,CAAC;QACrC,MAAM,CAAE,MAAM,CAAC,QAAsB,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACtD,MAAM,CAAE,MAAM,CAAC,aAA2B,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3D,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,QAAQ,EAAE,CAAC;QACrC,MAAM,CAAE,MAAM,CAAC,OAAmC,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC1E,MAAM,CAAE,MAAM,CAAC,OAAmC,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACxE,MAAM,CAAE,MAAM,CAAC,OAAmC,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IAC7E,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,+CAA+C,EAAE,KAAK,IAAI,EAAE;QAC7D,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,sBAAsB,EAAE,EAAE,CAAC,CAAC;QAE1D,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,sBAAsB,CAAC,CAAC,CAAC,CAAC;QAC/C,MAAM,SAAS,GAAG,MAAM,CAAC,SAA2C,CAAC;QACrE,MAAM,KAAK,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,UAAU,CAAC,CAAC;QACzD,MAAM,CAAC,KAAK,CAAC,CAAC,UAAU,EAAE,CAAC;QAC3B,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACvC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8CAA8C,EAAE,KAAK,IAAI,EAAE;QAC5D,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,kBAAkB,EAAE;YAChD,UAAU;YACV,OAAO,EAAE,KAAK;YACd,eAAe,EAAE,oDAAoD;YACrE,mBAAmB,EAAE,uCAAuC;SAC7D,CAAC,CAAC;QAEH,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;QACvC,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACnC,+BAA+B;QAC/B,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,WAAW,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;QACjD,0BAA0B;QAC1B,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,WAAW,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;IAChD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,yCAAyC,EAAE,KAAK,IAAI,EAAE;QACvD,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,kBAAkB,EAAE;YAChD,UAAU;YACV,OAAO,EAAE,IAAI;YACb,eAAe,EAAE,kBAAkB;SACpC,CAAC,CAAC;QAEH,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,SAAS,CAAC,kBAAkB,CAAC,CAAC;IACrD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,mDAAmD,EAAE,KAAK,IAAI,EAAE;QACjE,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,2BAA2B,EAAE,EAAE,CAAC,CAAC;QAE/D,MAAM,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC,sBAAsB,CAAC,CAAC,CAAC,CAAC;QACrD,2EAA2E;QAC3E,0EAA0E;QAC1E,MAAM,CAAC,MAAM,CAAC,YAAsB,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC;QACzD,MAAM,CAAC,MAAM,CAAC,YAAsB,CAAC,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC;IAC1D,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,+EAA+E;AAE/E,QAAQ,CAAC,yBAAyB,EAAE,GAAG,EAAE;IACvC,SAAS,CAAC,KAAK,IAAI,EAAE;QACnB,qEAAqE;QACrE,MAAM,SAAS,GAAG;YAChB,EAAE,WAAW,EAAE,GAAG,EAAE,OAAO,EAAE,KAAK,EAAE,EAAI,cAAc;YACtD,EAAE,WAAW,EAAE,GAAG,EAAE,OAAO,EAAE,KAAK,EAAE,EAAI,cAAc;YACtD,EAAE,WAAW,EAAE,GAAG,EAAE,OAAO,EAAE,IAAI,EAAE,EAAK,cAAc;YACtD,EAAE,WAAW,EAAE,GAAG,EAAE,OAAO,EAAE,IAAI,EAAE,EAAK,cAAc;YACtD,EAAE,WAAW,EAAE,GAAG,EAAE,OAAO,EAAE,KAAK,EAAE,EAAI,cAAc;SACvD,CAAC;QAEF,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,SAAS,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YAC1C,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,iBAAiB,EAAE;gBAC/C,QAAQ,EAAE,6BAA6B,CAAC,GAAG,CAAC,GAAG;gBAC/C,cAAc,EAAE,YAAY;gBAC5B,kBAAkB,EAAE,iBAAiB,CAAC,GAAG,CAAC,EAAE;gBAC5C,WAAW,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,WAAW;gBACrC,IAAI,EAAE,CAAC,kBAAkB,CAAC;aAC3B,CAAC,CAAC;YAEH,MAAM,QAAQ,CAAC,kBAAkB,EAAE;gBACjC,UAAU,EAAG,MAAkC,CAAC,UAAU;gBAC1D,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,OAAO;gBAC7B,eAAe,EAAE,mBAAmB,CAAC,GAAG,CAAC,EAAE;aAC5C,CAAC,CAAC;QACL,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,uCAAuC,EAAE,KAAK,IAAI,EAAE;QACrD,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,qBAAqB,EAAE,EAAE,CAAC,CAAC;QAEzD,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,CAAC;QACjC,MAAM,CAAE,MAAM,CAAC,IAAkB,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACnD,MAAM,CAAC,MAAM,CAAC,YAAY,CAAC,CAAC,UAAU,EAAE,CAAC;QACzC,MAAM,CAAC,OAAO,MAAM,CAAC,YAAY,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QAClD,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,sBAAsB,CAAC,CAAC,CAAC,CAAC;IACzD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,6DAA6D,EAAE,KAAK,IAAI,EAAE;QAC3E,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,2BAA2B,EAAE,EAAE,CAAC,CAAC;QAE/D,8DAA8D;QAC9D,MAAM,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC,sBAAsB,CAAC,CAAC,CAAC,CAAC;QACrD,qCAAqC;QACrC,MAAM,CAAC,MAAM,CAAC,YAAsB,CAAC,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC;IAC3D,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,+EAA+E;AAE/E,QAAQ,CAAC,wBAAwB,EAAE,GAAG,EAAE;IACtC,EAAE,CAAC,+CAA+C,EAAE,KAAK,IAAI,EAAE;QAC7D,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,iBAAiB,EAAE;YAC/C,QAAQ,EAAE,mBAAmB;YAC7B,cAAc,EAAE,YAAY;YAC5B,kBAAkB,EAAE,MAAM;YAC1B,WAAW,EAAE,CAAC,GAAG;SAClB,CAAC,CAAC;QAEH,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,SAAS,CAAC,iBAAiB,CAAC,CAAC;IACpD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4DAA4D,EAAE,KAAK,IAAI,EAAE;QAC1E,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,iBAAiB,EAAE;YAC/C,QAAQ,EAAE,4BAA4B;YACtC,cAAc,EAAE,YAAY;YAC5B,kBAAkB,EAAE,MAAM;YAC1B,WAAW,EAAE,GAAG;SACjB,CAAC,CAAC;QAEH,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC,kBAAkB,EAAE;YACjD,UAAU,EAAG,MAAkC,CAAC,UAAU;YAC1D,OAAO,EAAE,WAAW;YACpB,eAAe,EAAE,kCAAkC;SACpD,CAAC,CAAC;QAEH,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC,QAAQ,EAAE,CAAC;QACtC,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,QAAQ,EAAE,CAAC;IACtC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,6DAA6D,EAAE,KAAK,IAAI,EAAE;QAC3E,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,oBAAoB,EAAE;YAClD,UAAU,EAAE,gBAAgB;SAC7B,CAAC,CAAC;QAEH,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,SAAS,CAAC,WAAW,CAAC,CAAC;IAC9C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,wCAAwC,EAAE,KAAK,IAAI,EAAE;QACtD,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,sBAAsB,EAAE;YACpD,IAAI,EAAE,CAAC,kBAAkB,CAAC;SAC3B,CAAC,CAAC;QAEH,mEAAmE;QACnE,MAAM,SAAS,GAAG,MAAM,CAAC,SAA2C,CAAC;QACrE,MAAM,WAAW,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CACxC,CAAC,CAAC,IAAiB,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAClD,CAAC;QACF,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACrC,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}

package/dist/__tests__/forecastingScoring.test.d.ts

@@ -1,9 +0,0 @@
-/**
- * Forecasting Scoring Engine — Unit Tests
- *
- * Tests for brierScore, logScore, calibrationBins, isotonicCalibrate,
- * averageBrier, averageLogScore, formatForecastDiff.
- *
- * All functions are pure — no Convex, no SQLite, no network.
- */
-export {};

package/dist/__tests__/forecastingScoring.test.js

@@ -1,202 +0,0 @@
-/**
- * Forecasting Scoring Engine — Unit Tests
- *
- * Tests for brierScore, logScore, calibrationBins, isotonicCalibrate,
- * averageBrier, averageLogScore, formatForecastDiff.
- *
- * All functions are pure — no Convex, no SQLite, no network.
- */
-import { describe, it, expect } from "vitest";
-import { brierScore, logScore, calibrationBins, averageBrier, averageLogScore, isotonicCalibrate, formatForecastDiff, } from "../../../../convex/domains/forecasting/scoringEngine";
-// ─── Brier Score ────────────────────────────────────────────────────────────
-describe("brierScore", () => {
-    it("perfect prediction — yes at 100%", () => {
-        expect(brierScore(1.0, "yes")).toBe(0);
-    });
-    it("perfect prediction — no at 0%", () => {
-        expect(brierScore(0.0, "no")).toBe(0);
-    });
-    it("worst prediction — yes at 0%", () => {
-        expect(brierScore(0.0, "yes")).toBe(1);
-    });
-    it("worst prediction — no at 100%", () => {
-        expect(brierScore(1.0, "no")).toBe(1);
-    });
-    it("coin flip — 50% on yes", () => {
-        expect(brierScore(0.5, "yes")).toBe(0.25);
-    });
-    it("coin flip — 50% on no", () => {
-        expect(brierScore(0.5, "no")).toBe(0.25);
-    });
-    it("65% on yes outcome", () => {
-        expect(brierScore(0.65, "yes")).toBeCloseTo(0.1225, 4);
-    });
-    it("80% on no outcome", () => {
-        expect(brierScore(0.8, "no")).toBeCloseTo(0.64, 4);
-    });
-});
-// ─── Log Score ──────────────────────────────────────────────────────────────
-describe("logScore", () => {
-    it("perfect prediction — yes at ~100%", () => {
-        expect(logScore(0.999, "yes")).toBeCloseTo(0.001, 2);
-    });
-    it("perfect prediction — no at ~0%", () => {
-        expect(logScore(0.001, "no")).toBeCloseTo(0.001, 2);
-    });
-    it("worst prediction — yes at ~0%", () => {
-        // -log(0.001) ≈ 6.9
-        expect(logScore(0.001, "yes")).toBeCloseTo(6.908, 1);
-    });
-    it("coin flip — 50% on yes", () => {
-        // -log(0.5) ≈ 0.693
-        expect(logScore(0.5, "yes")).toBeCloseTo(0.693, 2);
-    });
-    it("clamps extreme values to avoid -Infinity", () => {
-        const score = logScore(0.0, "yes");
-        expect(Number.isFinite(score)).toBe(true);
-        expect(score).toBeGreaterThan(0);
-    });
-    it("clamps 1.0 for no outcome", () => {
-        const score = logScore(1.0, "no");
-        expect(Number.isFinite(score)).toBe(true);
-        expect(score).toBeGreaterThan(0);
-    });
-});
-// ─── Calibration Bins ───────────────────────────────────────────────────────
-describe("calibrationBins", () => {
-    it("returns 10 bins", () => {
-        const bins = calibrationBins([]);
-        expect(bins).toHaveLength(10);
-    });
-    it("first bin is 0-10%", () => {
-        const bins = calibrationBins([]);
-        expect(bins[0].binLabel).toBe("0-10%");
-        expect(bins[0].predictedProb).toBe(0.05);
-    });
-    it("last bin is 90-100%", () => {
-        const bins = calibrationBins([]);
-        expect(bins[9].binLabel).toBe("90-100%");
-        expect(bins[9].predictedProb).toBe(0.95);
-    });
-    it("empty bins have count 0 and observedFreq 0", () => {
-        const bins = calibrationBins([]);
-        for (const bin of bins) {
-            expect(bin.count).toBe(0);
-            expect(bin.observedFreq).toBe(0);
-        }
-    });
-    it("correctly bins a single forecast", () => {
-        const bins = calibrationBins([{ probability: 0.75, outcome: "yes" }]);
-        const bin70 = bins[7]; // 70-80%
-        expect(bin70.count).toBe(1);
-        expect(bin70.observedFreq).toBe(1); // 1/1 = yes
-    });
-    it("correctly computes observed frequency", () => {
-        const forecasts = [
-            { probability: 0.55, outcome: "yes" },
-            { probability: 0.52, outcome: "no" },
-            { probability: 0.58, outcome: "yes" },
-            { probability: 0.51, outcome: "no" },
-        ];
-        const bins = calibrationBins(forecasts);
-        const bin50 = bins[5]; // 50-60%
-        expect(bin50.count).toBe(4);
-        expect(bin50.observedFreq).toBe(0.5); // 2/4
-    });
-    it("boundary value 1.0 goes in 90-100% bin", () => {
-        const bins = calibrationBins([{ probability: 1.0, outcome: "yes" }]);
-        expect(bins[9].count).toBe(1);
-    });
-    it("boundary value 0.0 goes in 0-10% bin", () => {
-        const bins = calibrationBins([{ probability: 0.0, outcome: "no" }]);
-        expect(bins[0].count).toBe(1);
-    });
-});
-// ─── Average Brier ──────────────────────────────────────────────────────────
-describe("averageBrier", () => {
-    it("returns 0 for empty array", () => {
-        expect(averageBrier([])).toBe(0);
-    });
-    it("returns single score for single forecast", () => {
-        expect(averageBrier([{ probability: 0.7, outcome: "yes" }])).toBeCloseTo(0.09, 2);
-    });
-    it("averages multiple forecasts", () => {
-        const forecasts = [
-            { probability: 1.0, outcome: "yes" }, // 0
-            { probability: 0.0, outcome: "yes" }, // 1
-        ];
-        expect(averageBrier(forecasts)).toBe(0.5);
-    });
-});
-// ─── Average Log Score ──────────────────────────────────────────────────────
-describe("averageLogScore", () => {
-    it("returns 0 for empty array", () => {
-        expect(averageLogScore([])).toBe(0);
-    });
-    it("lower for better-calibrated forecasts", () => {
-        const good = [
-            { probability: 0.9, outcome: "yes" },
-            { probability: 0.1, outcome: "no" },
-        ];
-        const bad = [
-            { probability: 0.1, outcome: "yes" },
-            { probability: 0.9, outcome: "no" },
-        ];
-        expect(averageLogScore(good)).toBeLessThan(averageLogScore(bad));
-    });
-});
-// ─── Isotonic Calibration ───────────────────────────────────────────────────
-describe("isotonicCalibrate", () => {
-    it("returns raw probability with fewer than 3 non-empty bins", () => {
-        const sparse = calibrationBins([
-            { probability: 0.15, outcome: "yes" },
-            { probability: 0.85, outcome: "no" },
-        ]);
-        expect(isotonicCalibrate(0.5, sparse)).toBe(0.5);
-    });
-    it("returns a value in [0, 1]", () => {
-        // Create bins with enough data
-        const forecasts = Array.from({ length: 50 }, (_, i) => ({
-            probability: (i + 0.5) / 50,
-            outcome: (Math.random() > 0.5 ? "yes" : "no"),
-        }));
-        const bins = calibrationBins(forecasts);
-        const calibrated = isotonicCalibrate(0.7, bins);
-        expect(calibrated).toBeGreaterThanOrEqual(0);
-        expect(calibrated).toBeLessThanOrEqual(1);
-    });
-    it("produces monotonically non-decreasing output for ordered inputs", () => {
-        // Create well-populated bins
-        const forecasts = [];
-        for (let i = 0; i < 10; i++) {
-            for (let j = 0; j < 10; j++) {
-                const p = (i * 10 + j + 0.5) / 100;
-                // Roughly calibrated: higher p → more "yes"
-                const outcome = Math.random() < p ? "yes" : "no";
-                forecasts.push({ probability: p, outcome: outcome });
-            }
-        }
-        const bins = calibrationBins(forecasts);
-        const inputs = [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9];
-        const outputs = inputs.map((p) => isotonicCalibrate(p, bins));
-        for (let i = 0; i < outputs.length - 1; i++) {
-            expect(outputs[i]).toBeLessThanOrEqual(outputs[i + 1] + 0.05); // allow for random data noise in PAV
-        }
-    });
-});
-// ─── Format Forecast Diff ───────────────────────────────────────────────────
-describe("formatForecastDiff", () => {
-    it("formats increase correctly", () => {
-        const diff = formatForecastDiff(0.35, 0.55, "New evidence from Fed minutes");
-        expect(diff).toBe("35% → 55% (+20pp): New evidence from Fed minutes");
-    });
-    it("formats decrease correctly", () => {
-        const diff = formatForecastDiff(0.8, 0.6, "CEO resignation");
-        expect(diff).toBe("80% → 60% (-20pp): CEO resignation");
-    });
-    it("formats no change", () => {
-        const diff = formatForecastDiff(0.5, 0.5, "No new evidence");
-        expect(diff).toBe("50% → 50% (+0pp): No new evidence");
-    });
-});
-//# sourceMappingURL=forecastingScoring.test.js.map

package/dist/__tests__/forecastingScoring.test.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"forecastingScoring.test.js","sourceRoot":"","sources":["../../src/__tests__/forecastingScoring.test.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,MAAM,QAAQ,CAAC;AAC9C,OAAO,EACL,UAAU,EACV,QAAQ,EACR,eAAe,EACf,YAAY,EACZ,eAAe,EACf,iBAAiB,EACjB,kBAAkB,GACnB,MAAM,sDAAsD,CAAC;AAE9D,+EAA+E;AAE/E,QAAQ,CAAC,YAAY,EAAE,GAAG,EAAE;IAC1B,EAAE,CAAC,kCAAkC,EAAE,GAAG,EAAE;QAC1C,MAAM,CAAC,UAAU,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACzC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,+BAA+B,EAAE,GAAG,EAAE;QACvC,MAAM,CAAC,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACxC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8BAA8B,EAAE,GAAG,EAAE;QACtC,MAAM,CAAC,UAAU,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACzC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,+BAA+B,EAAE,GAAG,EAAE;QACvC,MAAM,CAAC,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACxC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,wBAAwB,EAAE,GAAG,EAAE;QAChC,MAAM,CAAC,UAAU,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC5C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,uBAAuB,EAAE,GAAG,EAAE;QAC/B,MAAM,CAAC,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC3C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,oBAAoB,EAAE,GAAG,EAAE;QAC5B,MAAM,CAAC,UAAU,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC,CAAC,WAAW,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IACzD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,mBAAmB,EAAE,GAAG,EAAE;QAC3B,MAAM,CAAC,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,CAAC,WAAW,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;IACrD,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,+EAA+E;AAE/E,QAAQ,CAAC,UAAU,EAAE,GAAG,EAAE;IACxB,EAAE,CAAC,mCAAmC,EAAE,GAAG,EAAE;QAC3C,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,WAAW,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;IACvD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gCAAgC,EAAE,GAAG,EAAE;QACxC,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC,CAAC,WAAW,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;IACtD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,+BAA+B,EAAE,GAAG,EAAE;QACvC,oBAAoB;QACpB,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,WAAW,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;IACvD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,wBAAwB,EAAE,GAAG,EAAE;QAChC,oBAAoB;QACpB,MAAM,CAAC,QAAQ,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC,CAAC,WAAW,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;IACrD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,0CAA0C,EAAE,GAAG,EAAE;QAClD,MAAM,KAAK,GAAG,QAAQ,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;QACnC,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC1C,MAAM,CAAC,KAAK,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC;IACnC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,2BAA2B,EAAE,GAAG,EAAE;QACnC,MAAM,KAAK,GAAG,QAAQ,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;QAClC,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC1C,MAAM,CAAC,KAAK,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC;IACnC,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,+EAA+E;AAE/E,QAAQ,CAAC,iBAAiB,EAAE,GAAG,EAAE;IAC/B,EAAE,CAAC,iBAAiB,EAAE,GAAG,EAAE;QACzB,MAAM,IAAI,GAAG,eAAe,CAAC,EAAE,CAAC,CAAC;QACjC,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,CAAC,EAAE,CAAC,CAAC;IAChC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,oBAAoB,EAAE,GAAG,EAAE;QAC5B,MAAM,IAAI,GAAG,eAAe,CAAC,EAAE,CAAC,CAAC;QACjC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QACvC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC3C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,qBAAqB,EAAE,GAAG,EAAE;QAC7B,MAAM,IAAI,GAAG,eAAe,CAAC,EAAE,CAAC,CAAC;QACjC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACzC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC3C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4CAA4C,EAAE,GAAG,EAAE;QACpD,MAAM,IAAI,GAAG,eAAe,CAAC,EAAE,CAAC,CAAC;QACjC,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;YACvB,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YAC1B,MAAM,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACnC,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,kCAAkC,EAAE,GAAG,EAAE;QAC1C,MAAM,IAAI,GAAG,eAAe,CAAC,CAAC,EAAE,WAAW,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC;QACtE,MAAM,KAAK,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS;QAChC,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC5B,MAAM,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,YAAY;IAClD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,uCAAuC,EAAE,GAAG,EAAE;QAC/C,MAAM,SAAS,GAAG;YAChB,EAAE,WAAW,EAAE,IAAI,EAAE,OAAO,EAAE,KAAc,EAAE;YAC9C,EAAE,WAAW,EAAE,IAAI,EAAE,OAAO,EAAE,IAAa,EAAE;YAC7C,EAAE,WAAW,EAAE,IAAI,EAAE,OAAO,EAAE,KAAc,EAAE;YAC9C,EAAE,WAAW,EAAE,IAAI,EAAE,OAAO,EAAE,IAAa,EAAE;SAC9C,CAAC;QACF,MAAM,IAAI,GAAG,eAAe,CAAC,SAAS,CAAC,CAAC;QACxC,MAAM,KAAK,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS;QAChC,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC5B,MAAM,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM;IAC9C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,wCAAwC,EAAE,GAAG,EAAE;QAChD,MAAM,IAAI,GAAG,eAAe,CAAC,CAAC,EAAE,WAAW,EAAE,GAAG,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC;QACrE,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAChC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,sCAAsC,EAAE,GAAG,EAAE;QAC9C,MAAM,IAAI,GAAG,eAAe,CAAC,CAAC,EAAE,WAAW,EAAE,GAAG,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;QACpE,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAChC,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,+EAA+E;AAE/E,QAAQ,CAAC,cAAc,EAAE,GAAG,EAAE;IAC5B,EAAE,CAAC,2BAA2B,EAAE,GAAG,EAAE;QACnC,MAAM,CAAC,YAAY,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACnC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,0CAA0C,EAAE,GAAG,EAAE;QAClD,MAAM,CAAC,YAAY,CAAC,CAAC,EAAE,WAAW,EAAE,GAAG,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,CAAC,WAAW,CACtE,IAAI,EACJ,CAAC,CACF,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,6BAA6B,EAAE,GAAG,EAAE;QACrC,MAAM,SAAS,GAAG;YAChB,EAAE,WAAW,EAAE,GAAG,EAAE,OAAO,EAAE,KAAc,EAAE,EAAE,IAAI;YACnD,EAAE,WAAW,EAAE,GAAG,EAAE,OAAO,EAAE,KAAc,EAAE,EAAE,IAAI;SACpD,CAAC;QACF,MAAM,CAAC,YAAY,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAC5C,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,+EAA+E;AAE/E,QAAQ,CAAC,iBAAiB,EAAE,GAAG,EAAE;IAC/B,EAAE,CAAC,2BAA2B,EAAE,GAAG,EAAE;QACnC,MAAM,CAAC,eAAe,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACtC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,uCAAuC,EAAE,GAAG,EAAE;QAC/C,MAAM,IAAI,GAAG;YACX,EAAE,WAAW,EAAE,GAAG,EAAE,OAAO,EAAE,KAAc,EAAE;YAC7C,EAAE,WAAW,EAAE,GAAG,EAAE,OAAO,EAAE,IAAa,EAAE;SAC7C,CAAC;QACF,MAAM,GAAG,GAAG;YACV,EAAE,WAAW,EAAE,GAAG,EAAE,OAAO,EAAE,KAAc,EAAE;YAC7C,EAAE,WAAW,EAAE,GAAG,EAAE,OAAO,EAAE,IAAa,EAAE;SAC7C,CAAC;QACF,MAAM,CAAC,eAAe,CAAC,IAAI,CAAC,CAAC,CAAC,YAAY,CAAC,eAAe,CAAC,GAAG,CAAC,CAAC,CAAC;IACnE,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,+EAA+E;AAE/E,QAAQ,CAAC,mBAAmB,EAAE,GAAG,EAAE;IACjC,EAAE,CAAC,0DAA0D,EAAE,GAAG,EAAE;QAClE,MAAM,MAAM,GAAG,eAAe,CAAC;YAC7B,EAAE,WAAW,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,EAAE;YACrC,EAAE,WAAW,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE;SACrC,CAAC,CAAC;QACH,MAAM,CAAC,iBAAiB,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IACnD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,2BAA2B,EAAE,GAAG,EAAE;QACnC,+BAA+B;QAC/B,MAAM,SAAS,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,EAAE,EAAE,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;YACtD,WAAW,EAAE,CAAC,CAAC,GAAG,GAAG,CAAC,GAAG,EAAE;YAC3B,OAAO,EAAE,CAAC,IAAI,CAAC,MAAM,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAiB;SAC9D,CAAC,CAAC,CAAC;QACJ,MAAM,IAAI,GAAG,eAAe,CAAC,SAAS,CAAC,CAAC;QACxC,MAAM,UAAU,GAAG,iBAAiB,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;QAChD,MAAM,CAAC,UAAU,CAAC,CAAC,sBAAsB,CAAC,CAAC,CAAC,CAAC;QAC7C,MAAM,CAAC,UAAU,CAAC,CAAC,mBAAmB,CAAC,CAAC,CAAC,CAAC;IAC5C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,iEAAiE,EAAE,GAAG,EAAE;QACzE,6BAA6B;QAC7B,MAAM,SAAS,GAA0D,EAAE,CAAC;QAC5E,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC;YAC5B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC;gBAC5B,MAAM,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,GAAG,GAAG,CAAC,GAAG,GAAG,CAAC;gBACnC,4CAA4C;gBAC5C,MAAM,OAAO,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC;gBACjD,SAAS,CAAC,IAAI,CAAC,EAAE,WAAW,EAAE,CAAC,EAAE,OAAO,EAAE,OAAuB,EAAE,CAAC,CAAC;YACvE,CAAC;QACH,CAAC;QACD,MAAM,IAAI,GAAG,eAAe,CAAC,SAAS,CAAC,CAAC;QAExC,MAAM,MAAM,GAAG,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC;QAC7D,MAAM,OAAO,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,iBAAiB,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;QAE9D,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;YAC5C,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,mBAAmB,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,qCAAqC;QACtG,CAAC;IACH,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,+EAA+E;AAE/E,QAAQ,CAAC,oBAAoB,EAAE,GAAG,EAAE;IAClC,EAAE,CAAC,4BAA4B,EAAE,GAAG,EAAE;QACpC,MAAM,IAAI,GAAG,kBAAkB,CAAC,IAAI,EAAE,IAAI,EAAE,+BAA+B,CAAC,CAAC;QAC7E,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,CACf,kDAAkD,CACnD,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4BAA4B,EAAE,GAAG,EAAE;QACpC,MAAM,IAAI,GAAG,kBAAkB,CAAC,GAAG,EAAE,GAAG,EAAE,iBAAiB,CAAC,CAAC;QAC7D,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,oCAAoC,CAAC,CAAC;IAC1D,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,mBAAmB,EAAE,GAAG,EAAE;QAC3B,MAAM,IAAI,GAAG,kBAAkB,CAAC,GAAG,EAAE,GAAG,EAAE,iBAAiB,CAAC,CAAC;QAC7D,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,mCAAmC,CAAC,CAAC;IACzD,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}

package/dist/__tests__/gaiaCapabilityAudioEval.test.d.ts

@@ -1,15 +0,0 @@
-/**
- * GAIA audio-backed capability/accuracy benchmark: LLM-only vs LLM+NodeBench MCP local audio tools.
- *
- * This lane targets GAIA tasks that include audio attachments (MP3/WAV/etc).
- * We provide deterministic local transcription via NodeBench MCP tools and score answers against
- * the ground-truth "Final answer" (stored locally under `.cache/gaia`, gitignored).
- *
- * Safety:
- * - GAIA is gated. Do not commit fixtures that contain prompts/answers.
- * - This test logs only task IDs and aggregate metrics (no prompt/answer text).
- *
- * Disabled by default (cost + rate limits). Run with:
- *   NODEBENCH_RUN_GAIA_CAPABILITY=1 npm --prefix packages/mcp-local run test
- */
-export {};