# Offer库模拟项目评分表与修改记录

版本：2026-08 · 适用于产品经理、AI 产品与数据运营模拟项目

> 使用边界：本评分表用于评审模拟练习，不代表企业项目验收，也不构成招聘结果保证。简历和作品集中必须明确写明“独立模拟项目 / Independent practice project”，不得冒充实习、客户项目或真实商业成果。

## 一、提交前必须具备的证据

- 一页项目 brief：目标用户、问题、范围、时间和不做什么。
- 假设与证据表：哪些是公开事实、哪些是练习设定、哪些仍待验证。
- 决策记录：至少三项备选方案、选择依据和被放弃方案。
- 核心交付物：PRD、流程、原型、SQL、指标字典或 Evals，按项目类型选择。
- 验证计划：成功指标、失败类型、测试样本、阈值和后续动作。
- 修改记录：保留 V1、评分、修改动作和 V2，不覆盖旧版本。

## 二、100 分评分表

每项先按 0–4 级评分，再换算为对应权重。0 = 缺失；1 = 只有结论；2 = 基本完整但证据薄弱；3 = 有证据且能解释取舍；4 = 可复核、边界清楚且经历过修改。

| 维度 | 权重 | 评审问题 |
| --- | ---: | --- |
| 问题与证据 | 15 | 问题是否具体？事实、假设和练习设定是否分开？是否有反证？ |
| 用户、范围与约束 | 15 | 用户和场景是否明确？时间、资源、合规和不做范围是否清楚？ |
| 方案与取舍 | 20 | 是否比较过备选方案？关键决策能否追溯到证据和约束？ |
| 交付物正确性 | 20 | PRD、流程、SQL、指标或 Evals 是否内部一致、可执行、可检查？ |
| 验证与迭代 | 15 | 是否定义成功、失败、样本和阈值？反馈后是否形成 V2？ |
| 表达与诚信 | 15 | 能否在 5–10 分钟讲清个人工作、限制和反思？是否明确模拟性质？ |

建议门槛：

- 60 分以下：不要放入作品集，先补齐问题、证据和核心交付物。
- 60–74 分：可继续修改，但面试追问下容易暴露证据或取舍不足。
- 75–84 分：可作为作品集草稿，至少再做一轮讲解和反驳练习。
- 85 分以上：可进入定向投递版本；仍需按具体 JD 调整，而不是一稿通投。

## 三、AI 产品项目附加检查

- 任务成功是否被定义为可观察结果，而不是“模型感觉不错”？
- 是否有最小评测集、评分规则、基线和失败类型？
- 是否记录效果、延迟、成本、安全和人工兜底之间的取舍？
- Prompt、模型或工具不可用时，产品如何失败、提示和恢复？
- 是否避免使用不可公开的数据、模型输出或公司机密？

## 四、数据运营 / SQL 项目附加检查

- 每个指标是否写明分子、分母、时间窗、去重键和数据源？
- SQL 是否能在提供的数据结构上运行，并处理空值、重复和时间边界？
- 相关性、因果和练习假设是否被明确区分？
- 异常诊断是否先排口径和数据链路，再检查分群与外部事件？
- 结论是否对应具体行动、负责人、阈值和复查时间？

## 五、三轮修改记录

| 轮次 | 当前总分 | 最大问题 | 本轮修改 | 修改后证据 | 下一轮日期 |
| --- | ---: | --- | --- | --- | --- |
| V1 → V2 |  |  |  |  |  |
| V2 → V3 |  |  |  |  |  |
| V3 → 投递版 |  |  |  |  |  |

## 六、作品集与简历的诚信写法

推荐标题：

- 独立模拟项目：企业级 AI Agent 平台方案
- Independent practice project: Growth analytics dashboard

推荐描述结构：

1. 练习目标和设定边界；
2. 自己完成的分析与交付物；
3. 方案取舍和验证计划；
4. 根据评分或反馈做出的前后修改；
5. 尚未通过真实业务验证的限制。

不要写：真实上线、服务真实客户、带来收入/DAU 增长、曾在某公司负责——除非这些事实真实、获准披露且可由你本人解释。

---

# OfferKu Simulated Project Rubric and Revision Log

Version: 2026-08 · For PM, AI product, and data operations practice projects

> Boundary: This rubric evaluates simulated practice. It is not company acceptance criteria or a hiring guarantee. Label the work “Independent practice project”; never present it as an internship, client engagement, or live business result.

## Required Evidence Before Review

- One-page brief: target user, problem, scope, timeline, and non-goals.
- Assumption/evidence table separating public facts, exercise inputs, and unknowns.
- Decision log with at least three options, selection criteria, and rejected options.
- Core artifacts appropriate to the project: PRD, flow, prototype, SQL, metric dictionary, or evals.
- Validation plan covering success metrics, failure types, samples, thresholds, and actions.
- Revision log retaining V1, its score, changes, and V2.

## 100-Point Rubric

Score each dimension from 0–4, then convert it to the listed weight. 0 = missing; 1 = unsupported conclusion; 2 = basically complete but weak evidence; 3 = evidence-backed with explainable tradeoffs; 4 = reviewable, bounded, and revised.

| Dimension | Weight | Review question |
| --- | ---: | --- |
| Problem and evidence | 15 | Is the problem specific? Are facts, assumptions, and exercise inputs separated? |
| Users, scope, constraints | 15 | Are the user, scenario, resources, compliance constraints, and non-goals explicit? |
| Options and tradeoffs | 20 | Were alternatives compared, and can decisions be traced to evidence and constraints? |
| Artifact correctness | 20 | Are the PRD, flows, SQL, metrics, or evals internally consistent and checkable? |
| Validation and iteration | 15 | Are success, failure, samples, and thresholds defined? Is there a revised V2? |
| Communication and integrity | 15 | Can you explain contribution, limits, and reflection in 5–10 minutes and label the simulation clearly? |

Suggested thresholds: below 60, do not publish; 60–74, revise; 75–84, portfolio draft plus another challenge round; 85+, tailor for a target JD while retaining explicit limits.

## AI Product Addendum

- Define task success as an observable outcome.
- Include a minimum eval set, rubric, baseline, and failure taxonomy.
- Record tradeoffs across quality, latency, cost, safety, and human fallback.
- Specify failure, messaging, and recovery when a model or tool is unavailable.

## Data Operations / SQL Addendum

- Define numerator, denominator, time window, deduplication key, and source.
- Run SQL against the supplied schema and handle nulls, duplicates, and time boundaries.
- Separate correlation, causality, and exercise assumptions.
- Check definitions and pipelines before segments and external events.
- Connect findings to an action, owner, threshold, and review date.

## Three Revision Rounds

| Round | Current score | Largest gap | Change | New evidence | Next review |
| --- | ---: | --- | --- | --- | --- |
| V1 → V2 |  |  |  |  |  |
| V2 → V3 |  |  |  |  |  |
| V3 → application |  |  |  |  |  |

Safe titles include “Independent practice project: Enterprise AI Agent platform” or “Independent practice project: Growth analytics dashboard.” State the exercise boundary, your artifacts, tradeoffs, revisions, and what has not been validated in a live business.
