experiment(confidence-engine): complete openai reconstruction apparatus

This commit is contained in:
2026-09-06 10:06:06 +01:00
parent a45dd903cf
commit 215c783d11
5 changed files with 281 additions and 7 deletions
+75
View File
@@ -1,9 +1,13 @@
import { describe, expect, it, vi } from "vitest";
import {
__resetChatSupportForTests,
createOpenAIStrictSchema,
createOpenAIReconstructionProvider,
getProvider,
normaliseOpenAITransportResponse,
} from "@/lib/llm/provider.js";
import { reconstructionV2Schema } from "@/lib/reconstruction/schema.js";
import { z } from "zod";
describe("OllamaLlmProvider chat capability detection", () => {
it("uses the configured model for the chat probe and keeps the chat path", async () => {
@@ -121,6 +125,76 @@ describe("OllamaLlmProvider chat capability detection", () => {
});
describe("OpenAI reconstruction provider experiment seam", () => {
it("projects canonical optional fields as required but nullable", () => {
const nativeSchema = z.toJSONSchema(reconstructionV2Schema);
const projectedSchema = createOpenAIStrictSchema(
nativeSchema,
reconstructionV2Schema,
);
function resolveLocalRef(schema, root = projectedSchema) {
if (!schema.$ref) return schema;
return schema.$ref
.replace(/^#\//, "")
.split("/")
.reduce((value, key) => value?.[key], root);
}
function acceptsNull(schema, root = projectedSchema) {
const resolved = resolveLocalRef(schema, root);
return (
resolved?.type === "null" ||
(Array.isArray(resolved?.type) && resolved.type.includes("null")) ||
[...(resolved?.anyOf ?? []), ...(resolved?.oneOf ?? [])].some((branch) =>
acceptsNull(branch, root),
)
);
}
function assertAllPropertiesRequired(schema, root = projectedSchema) {
const resolved = resolveLocalRef(schema, root);
if (resolved?.properties) {
expect(resolved.required).toEqual(Object.keys(resolved.properties));
Object.values(resolved.properties).forEach((property) => assertAllPropertiesRequired(property, root));
}
(resolved?.anyOf ?? []).forEach((branch) => assertAllPropertiesRequired(branch, root));
(resolved?.oneOf ?? []).forEach((branch) => assertAllPropertiesRequired(branch, root));
if (resolved?.items) assertAllPropertiesRequired(resolved.items, root);
}
assertAllPropertiesRequired(projectedSchema);
const transition = projectedSchema.properties.reconstruction.properties.unexplainedTransitions.items;
expect(transition.required).toContain("entity");
expect(acceptsNull(transition.properties.entity)).toBe(true);
const inputClassification = resolveLocalRef(
projectedSchema.properties.inputClassification,
);
expect(inputClassification.required).toContain("secondaryTypes");
expect(acceptsNull(inputClassification.properties.secondaryTypes)).toBe(true);
expect(acceptsNull(transition.properties.id)).toBe(false);
});
it("removes only optional transport null placeholders", () => {
const schema = z.object({
optionalText: z.string().optional(),
nullableText: z.string().nullable(),
requiredText: z.string(),
children: z.array(z.object({ optionalChild: z.string().optional() })),
});
const nativeSchema = z.toJSONSchema(schema);
expect(normaliseOpenAITransportResponse({
optionalText: null,
nullableText: null,
requiredText: null,
children: [{ optionalChild: null }],
}, schema, nativeSchema)).toEqual({
nullableText: null,
requiredText: null,
children: [{}],
});
});
it("uses the Responses API with the canonical strict reconstruction schema", async () => {
const fetchSpy = vi.fn().mockResolvedValue({
ok: true,
@@ -158,6 +232,7 @@ describe("OpenAI reconstruction provider experiment seam", () => {
expect(schemaText).toContain("secondaryTypes");
expect(schemaText).toContain("confidence");
expect(schemaText).toContain("importance");
expect(request.text.format.schema).not.toEqual(z.toJSONSchema(reconstructionV2Schema));
expect(result).toEqual({
response: { reconstruction: "result" },
providerApiPath: "/v1/responses",
@@ -9,7 +9,7 @@
* with a deterministic inline double (zero live model calls).
*/
import { describe, expect, it, beforeAll, afterAll } from "vitest";
import { describe, expect, it, beforeAll, afterAll, vi } from "vitest";
import { execFile } from "child_process";
import { writeFile, unlink } from "fs/promises";
import { join, dirname } from "path";
@@ -159,6 +159,62 @@ describe("start-case-experiment-helper.cjs apparatus", () => {
// ── D — Success output ──────────────────────────────────────────
describe("D — success output", () => {
it("selects the existing OpenAI provider through the helper injection seam", async () => {
const { runStartCaseExperiment } = require(helperPath);
const previousProvider = process.env.START_CASE_EXPERIMENT_PROVIDER;
const previousKey = process.env.OPENAI_API_KEY;
process.env.START_CASE_EXPERIMENT_PROVIDER = "openai";
process.env.OPENAI_API_KEY = "test-key";
try {
const result = await runStartCaseExperiment(
{ scenario: "OpenAI provider selection scenario" },
null,
{
startCase: async (_input, dependencies) => {
expect(typeof dependencies.reconstructionProvider?.generateReconstruction).toBe("function");
expect(dependencies.reconstructionModelName).toBe("gpt-5.6-terra");
return {
success: true,
situationGraph: { nodes: [], edges: [] },
assessment: null,
selectedQuestion: null,
summary: "test",
};
},
},
);
expect(result.success).toBe(true);
} finally {
if (previousProvider === undefined) delete process.env.START_CASE_EXPERIMENT_PROVIDER;
else process.env.START_CASE_EXPERIMENT_PROVIDER = previousProvider;
if (previousKey === undefined) delete process.env.OPENAI_API_KEY;
else process.env.OPENAI_API_KEY = previousKey;
}
});
it("fails before startCase when OpenAI is selected without a key", async () => {
const { runStartCaseExperiment } = require(helperPath);
const previousProvider = process.env.START_CASE_EXPERIMENT_PROVIDER;
const previousKey = process.env.OPENAI_API_KEY;
process.env.START_CASE_EXPERIMENT_PROVIDER = "openai";
delete process.env.OPENAI_API_KEY;
const startCase = vi.fn();
try {
await expect(
runStartCaseExperiment({ scenario: "Missing key scenario" }, null, { startCase }),
).rejects.toThrow("START_CASE_EXPERIMENT_PROVIDER=openai requires OPENAI_API_KEY");
expect(startCase).not.toHaveBeenCalled();
} finally {
if (previousProvider === undefined) delete process.env.START_CASE_EXPERIMENT_PROVIDER;
else process.env.START_CASE_EXPERIMENT_PROVIDER = previousProvider;
if (previousKey === undefined) delete process.env.OPENAI_API_KEY;
else process.env.OPENAI_API_KEY = previousKey;
}
});
it("forwards an explicit reconstruction provider through the startCase path", async () => {
const { runStartCaseExperiment } = require(helperPath);
const reconstructionProvider = { generateReconstruction() {} };