InfoRSI/tests/config-steps.test.ts

import assert from "node:assert/strict";
import { test } from "node:test";
import { checkConfig, stepList, type ResearchConfig, type StepList } from "../src/config/index.ts";
import { fullConfig, naiveConfig } from "./config-helpers.ts";

// 每份预期都对照 docs/结构.md 第三部分"步骤清单"和 docs/方法.md 第十一部分逐条手写，不从实现里推。
// 记法：[步骤, 执行 / 跳过, 规则]；规则写名字和版本，版本取 rules 登记表里写的 1。

type Row = [string, "execute" | "skip", { name: string; version: number } | null];
const E = "execute" as const;
const S = "skip" as const;

function rows(steps: StepList["study"] | StepList["round"]): Row[] {
  return steps.map((s) => [s.id, s.status, s.rule]);
}

function listOf(change: (flow: Extract<ResearchConfig["flow"], { preset: "full" }>) => void): StepList {
  const config = fullConfig();
  change(config.flow as Extract<ResearchConfig["flow"], { preset: "full" }>);
  return stepList(checkConfig(config));
}

/** 跳过的步骤都写了原因。 */
function assertReasons(list: StepList): void {
  for (const s of [...list.study, ...list.round]) {
    if (s.status === "skip") assert.ok(s.reason.length > 0, `${s.id} 跳过却没写原因`);
  }
}

test("步骤清单：完整流程全开", () => {
  const list = stepList(fullConfig());
  assert.equal(list.preset, "full");
  assert.deepEqual(rows(list.study), [
    ["freezeConfig", E, null],
    ["baselineRun", E, null],
    ["entryDraft", E, null],
    ["discussion1", E, null],
    ["discussion2", E, null],
    ["claim", E, null],
    ["assignEntries", E, null],
    ["rounds", E, null],
    ["export", E, null],
  ]);
  assert.deepEqual(rows(list.round), [
    ["openRound", E, null],
    ["carryOverPool", E, null],
    ["proposal", E, null],
    ["dedupe", E, null],
    ["firstPrediction", E, null],
    ["critique", E, null],
    ["response", E, null],
    ["freezeProtocol", E, null],
    ["secondPrediction", E, null],
    ["selectBatch", E, { name: "bestExpected", version: 1 }],
    ["implementation", E, null],
    ["verification", E, null],
    ["repair", E, null],
    ["forwardRun", E, null],
    ["publish", E, null],
    ["informationAccount", E, null],
    ["adopt", E, null],
    ["listCells", E, null],
    ["explanation", E, null],
    ["selectCells", E, { name: "valueMinusCost", version: 1 }],
    ["ablationImplementation", E, null],
    ["ablationVerification", E, null],
    ["ablationRun", E, null],
    ["settle", E, null],
    ["wrapUp", E, null],
  ]);
});

test("步骤清单：只关第零轮", () => {
  const list = listOf((f) => (f.roundZero = { enabled: false }));
  assertReasons(list);
  assert.equal(list.preset, "full");
  assert.deepEqual(rows(list.study), [
    ["freezeConfig", E, null],
    ["baselineRun", E, null],
    ["entryDraft", S, null],
    ["discussion1", S, null],
    ["discussion2", S, null],
    ["claim", S, null],
    ["assignEntries", S, null],
    ["rounds", E, null],
    ["export", E, null],
  ]);
  assert.deepEqual(rows(list.round), rows(stepList(fullConfig()).round));
});

test("步骤清单：只关第一次预测", () => {
  const list = listOf((f) => (f.firstPrediction = false));
  assertReasons(list);
  assert.deepEqual(rows(list.study), rows(stepList(fullConfig()).study));
  assert.deepEqual(rows(list.round), [
    ["openRound", E, null],
    ["carryOverPool", E, null],
    ["proposal", E, null],
    ["dedupe", E, null],
    ["firstPrediction", S, null],
    ["critique", E, null],
    ["response", E, null],
    ["freezeProtocol", E, null],
    ["secondPrediction", E, null],
    ["selectBatch", E, { name: "bestExpected", version: 1 }],
    ["implementation", E, null],
    ["verification", E, null],
    ["repair", E, null],
    ["forwardRun", E, null],
    ["publish", E, null],
    ["informationAccount", E, null],
    ["adopt", E, null],
    ["listCells", E, null],
    ["explanation", E, null],
    ["selectCells", E, { name: "valueMinusCost", version: 1 }],
    ["ablationImplementation", E, null],
    ["ablationVerification", E, null],
    ["ablationRun", E, null],
    ["settle", E, null],
    ["wrapUp", E, null],
  ]);
});

test("步骤清单：只关挑刺与回应，协议照样冻结", () => {
  const list = listOf((f) => (f.critique = false));
  assertReasons(list);
  assert.deepEqual(rows(list.study), rows(stepList(fullConfig()).study));
  assert.deepEqual(rows(list.round), [
    ["openRound", E, null],
    ["carryOverPool", E, null],
    ["proposal", E, null],
    ["dedupe", E, null],
    ["firstPrediction", E, null],
    ["critique", S, null],
    ["response", S, null],
    ["freezeProtocol", E, null],
    ["secondPrediction", E, null],
    ["selectBatch", E, { name: "bestExpected", version: 1 }],
    ["implementation", E, null],
    ["verification", E, null],
    ["repair", E, null],
    ["forwardRun", E, null],
    ["publish", E, null],
    ["informationAccount", E, null],
    ["adopt", E, null],
    ["listCells", E, null],
    ["explanation", E, null],
    ["selectCells", E, { name: "valueMinusCost", version: 1 }],
    ["ablationImplementation", E, null],
    ["ablationVerification", E, null],
    ["ablationRun", E, null],
    ["settle", E, null],
    ["wrapUp", E, null],
  ]);
});

test("步骤清单：只关消融，解释照样写，选格没有规则", () => {
  const list = listOf((f) => (f.ablation = { enabled: false }));
  assertReasons(list);
  assert.deepEqual(rows(list.study), rows(stepList(fullConfig()).study));
  assert.deepEqual(rows(list.round), [
    ["openRound", E, null],
    ["carryOverPool", E, null],
    ["proposal", E, null],
    ["dedupe", E, null],
    ["firstPrediction", E, null],
    ["critique", E, null],
    ["response", E, null],
    ["freezeProtocol", E, null],
    ["secondPrediction", E, null],
    ["selectBatch", E, { name: "bestExpected", version: 1 }],
    ["implementation", E, null],
    ["verification", E, null],
    ["repair", E, null],
    ["forwardRun", E, null],
    ["publish", E, null],
    ["informationAccount", E, null],
    ["adopt", E, null],
    ["listCells", S, null],
    ["explanation", E, null],
    ["selectCells", S, null],
    ["ablationImplementation", S, null],
    ["ablationVerification", S, null],
    ["ablationRun", S, null],
    ["settle", S, null],
    ["wrapUp", E, null],
  ]);
});

test("步骤清单：朴素循环", () => {
  const list = stepList(naiveConfig());
  assertReasons(list);
  assert.equal(list.preset, "naive");
  assert.deepEqual(rows(list.study), [
    ["freezeConfig", E, null],
    ["baselineRun", E, null],
    ["entryDraft", S, null],
    ["discussion1", S, null],
    ["discussion2", S, null],
    ["claim", S, null],
    ["assignEntries", S, null],
    ["rounds", E, null],
    ["export", E, null],
  ]);
  assert.deepEqual(rows(list.round), [
    ["openRound", E, null],
    ["carryOverPool", S, null], // 候选不留池
    ["proposal", E, null],
    ["dedupe", E, null], // 判重照常
    ["firstPrediction", S, null], // 不预测
    ["critique", S, null], // 不挑刺
    ["response", S, null], // 不回应
    ["freezeProtocol", E, null], // 提案直接冻结
    ["secondPrediction", S, null], // 不预测
    ["selectBatch", E, { name: "uniformRandom", version: 1 }], // 均匀随机选批
    ["implementation", E, null],
    ["verification", E, null],
    ["repair", E, null],
    ["forwardRun", E, null],
    ["publish", E, null],
    ["informationAccount", S, null], // 没有预测，记不了信息账
    ["adopt", E, null], // 采纳规则和完整流程相同
    ["listCells", S, null], // 不消融
    ["explanation", S, null],
    ["selectCells", S, null],
    ["ablationImplementation", S, null],
    ["ablationVerification", S, null],
    ["ablationRun", S, null],
    ["settle", S, null],
    ["wrapUp", S, null], // 没有收尾
  ]);
});

test("步骤清单：第零轮讨论 1 次时跳过第 2 次；只有一位研究者时不讨论", () => {
  const once = listOf((f) => (f.roundZero = { enabled: true, discussions: 1 }));
  assertReasons(once);
  assert.deepEqual(rows(once.study).slice(2, 7), [
    ["entryDraft", E, null],
    ["discussion1", E, null],
    ["discussion2", S, null],
    ["claim", E, null],
    ["assignEntries", E, null],
  ]);

  const config = fullConfig();
  config.models.researchers = [config.models.researchers[0]!];
  const single = stepList(checkConfig(config));
  assertReasons(single);
  assert.deepEqual(rows(single.study).slice(2, 7), [
    ["entryDraft", E, null],
    ["discussion1", S, null],
    ["discussion2", S, null],
    ["claim", E, null],
    ["assignEntries", E, null],
  ]);
});

test("步骤清单：换规则后清单里的规则跟着换", () => {
  const list = listOf((f) => {
    f.batchRule = "uniformRandom";
    f.ablation = { enabled: true, cellRule: "randomSameCount" };
  });
  const rule = (id: string) => list.round.find((s) => s.id === id)!.rule;
  assert.deepEqual(rule("selectBatch"), { name: "uniformRandom", version: 1 });
  assert.deepEqual(rule("selectCells"), { name: "randomSameCount", version: 1 });
});