When to open:When you finish a project and want to do a retrospective / want to read others' failure lessons / want to conduct team-level retrospective training. This wiki integrates yitang's system + Wang Kai's practical experience + Xiaohongshu pitfalls.
One-sentence core:Truman: "Projects will most likely die, 90% barely make money" + "Failure is 30 times more common than success" (Wang Huiwen) —— Retrospective is not post-event punishment; it turns "money spent on failure" into "assets for the future".
0. Core Definition
- Retrospective = Post-event systematic analysis of "why it happened this way" + "how to improve next time"
- AAR(After Action Review) = 4-question military retrospective method
- 4-layer retrospective = Decision / Parameter / Transfer / Capability
1. Truman's In-Depth Retrospective System (Exclusive)
Yitang / First Lesson on In-Depth Retrospective:
1.1 5-Step Retrospective Method
1. 还原决策当时(我当时知道什么?基于什么判断?)
2. 拆开关键节点(哪几个节点决定结果?)
3. 评估实际 vs 预期(差多少?为什么差?)
4. 提炼可迁移规律(下次类似场景怎么办?)
5. 写进个人/团队的"判断库"
1.2 5 Retrospective Principles
- Focus on facts, not people(Avoid turning it into a criticism session)
- Fact-based(Not conclusion-based)
- Emotion-free(Wait until things cool down before doing it)
- Open and transparent(For all participants)
- Focus on patterns(Not single events)
1.3 4 Levels of Retrospective
| Level | Focus | Output |
|---|---|---|
| L1 Single-event retrospective | Why this time didn't go well | Fix the current problem |
| L2 Decision retrospective | Whether the decision itself was correct | Revise the decision framework |
| L3 Parameter retrospective | Whether key parameters were misestimated | Revise estimation methods |
| L4 Capability retrospective | Whether team capabilities were sufficient | Revise team building |
1.4 AAR 4 Questions (Military Retrospective Method)
- What was expected to happen?
- What actually happened?
- Why was there a difference?
- What to do next time?
1.5 Classic Quotes
- Wang Xing:"Do not use tactical diligence to cover up strategic laziness"
- Liu Chuanzhi:"Retrospective is earning back the money you spent in the past"
- Chairman Mao:"Learn war through war"
1.6 Candy Case: Yu Lu's 5-Year In-Depth Retrospective of Someet
- Full 5-year startup retrospective (from excitement to shutdown)
- Review of multiple decision nodes
- Mismatch between team capabilities and business challenges
- Conflict between long-term value and short-term growth
2. Wang Kai's 11 Firsthand Product Failure Lessons (Exclusive Full Version)
Wang Kai's Startup Methodology and Business Judgment Framework in the AI Era §12 + Wang Kai's Product Failure Retrospective Collection:
2.1 List of Firsthand Products
| Project | Time | Result | Key Lesson |
|---|---|---|---|
| enjoy Food E-commerce | Before 2018 | Company experience | Raised 100 million USD but the founding team made no money |
| Rishiqing | Before 2018 | Company experience | Painful internet SaaS transformation |
| AI E-commerce Guide Site | 2023 | Shut down | Early token costs could not be balanced |
| FastFacts(SEC announcement crawler) | 2023 | Shut down | Revenue ceiling of 300,000 RMB; quantitative teams were 5 seconds faster than us |
| MJ for Feishu | 2023 | Open source | Good traffic but no profit |
| AiPathly(AI job application report) | 2024 | Shut down | Mismatch of doing SEO for an innovative product |
| Quanta Quest(C-end wiki AI search) | 2024 | Shut down | Blocked by data permission issues |
| Squirrel Quick Read(PC official account reading + AI) | 2024 | Continuously operating | Limited domestic C-end scale but good retention |
| AI Smoking Cessation Product | 2025-2026 | Continuously operating | "Treat AI as a coach" |
| RSI AI Investment Strategy | 2025-2026 | 2-4% monthly return | See Wang Kai's RSI AI Investment Strategy and "Good Asset, Good Price, Good Volatility" Framework |
| KeepRule | 2026 | Demo launched | AI Coding self-running database |
2.2 Common Lessons
- Incorrect cost calculation(Most cases)
- Died within 3 months after launch(Indie developers should not give a project 1 year of patience)
- Ostrich mentality(Thinking that making a good product is enough, without thinking about traffic/payment)
2.3 6 Typical Ways to Fail
- WhatsApp push bot(Wang Kai's firsthand retrospective):Chose the wrong track; overseas push is directly covered by IMs
- AI e-commerce guide site:Wrong content generation model,token consumption could not be balanced
- MJ for Feishu:Open source brought spread but could not support the team
- Quanta Quest:Notion/Gmail Data permission application blocked; security agency intervention in evaluation dragged out time
- AiPathly:Doing SEO for innovative products is a complete mismatch(Search volume had not even been established)
- Domestic SaaS category:Low average deal size + heavy guanxi/FB culture + renewal rate is meaningless
2.4 Disciplines for Indie Developers
- Making a product is like making a chair:When there are no chairs in the world yet, don't worry about comfort when making the first one
- Big companies give 1 year of patience; indie developers decide life or death in 3 months
- After verifying that "someone wants the chair", then add armrests
- Projects will most likely die, 90% barely make money
- Don't have an ostrich mentality:Burying your head in building without thinking about traffic, payment, expansion
- Launch at least 2-3 projects at the same time, after finishing, maintain profitable ones at low cost and move on to the next
3. Detailed Retrospective of WhatsApp Push Bot (Wang Kai Exclusive Full Version)
Wang Kai's Product Failure Retrospective Collection:
3.1 Project Background
- Launched in 2023
- Wanted to build "overseas WeChat group bot + push"
- Pivoted to Telegram immediately after 6 hours of validation
3.2 4 Core Mistakes
- Time zone(Domestic user acquisition, overseas release, time misalignment)
- Zombie accounts(Used a large number of fake accounts for overseas promotion)
- Domestic traffic(Thought domestic traffic could influence overseas markets)
- Product preparation(Launched before infrastructure was ready)
3.3 Summary of 7 Project Failure Lessons
- Fundamental differences between overseas IM ecosystems and WeChat ecosystem(WhatsApp / Telegram / Discord / Signal each have their own user groups)
- Decision tree for IM selection for overseas products:Look at user groups / look at laws / look at ecosystem maturity
- Data permission review dragged out time(Quanta Quest)
- AI model costs could not be balanced(Early tokens were too expensive in 2023)
- SEO is not suitable for innovative products(AiPathly)
- Revenue ceiling too low(FastFacts 300,000 RMB / quantitative team 5-second advantage)
- Open source brought spread but could not support the team(MJ for Feishu)
4. Xiaohongshu Vibecoding Pitfalls (Exclusive)
Xiaohongshu / Vibecoding Practical Collection (Games/Tools/Mini Programs) §2 + newtype · AI Coding Practice and Toolstack Evolution §9.3:
4.1 0-80 vs 80-100 Life-or-Death Line
"Vibe coding gets a product from 0 to 80, but 80 to 100 is the state where the product truly becomes usable"(Interview with the author of "Ci Le Ge Ci")
80→100 Difficulties:
- Non-code logic(Dictionary integration + image-text matching = dirty work)
- Review time(2-month mini-game review, and the mini program code could not pass)
- User experience(80→100 is the real life-or-death dividing line)
- Interest > Technology(Development completion depends on self-drive,no monetary reward in the early stage))
4.2 Lao Huang's Vibe Coding Cleanup Prophecy
newtype · AI Coding Practice and Toolstack Evolution §9.3 (Post 1789, 2025-08):
"After you pass the validation period of building products with Vibe Coding and start to scale up for real,you will very likely need such a service (Vibe Coding cleanup).Because code written by AI may contain many hidden pitfalls. It's fine when the scale is small. When the system becomes more complex and has more features, these pitfalls will emerge."
Lao Huang also stepped on these pitfalls himself:
- Prompt House user login verification, sub-page flow Suddenly became unstable after adding more sub-pages(Repeated refreshing)
- Spent a lot of time troubleshooting with Codex
- Lesson:Did not prepare architecture for SEO sub-pages + kept adding new things without periodic tidying
4.3 Anti-patterns (What Vibe Coding Cannot Replace)
newtype · AI Coding Practice and Toolstack Evolution §9.4:
"With Vibe Coding, you don't have to write code, but the requirements in other aspects are not lowered at all. "
Meaning:
- Architectural thinking still needs to be done
- Periodic refactoring still needs to be done
- SEO / performance / security / user experience still need to be done
- Otherwise AI will accumulate more and more "barely running" code → eventually explode
5. 4-Layer Retrospective + Validation ROI (yitang Exclusive)
5.1 4-Layer Retrospective
| Layer | Focus |
|---|---|
| L1 Decision retrospective | Was the decision itself correct? |
| L2 Parameter retrospective | Were key parameters misestimated? |
| L3 Transfer retrospective | Can this pattern be transferred to other scenarios? |
| L4 Capability retrospective | Were team capabilities sufficient? |
5.2 "Validation ROI" Judgment for Retrospectives
Key question:How much can this retrospective earn back in the future?
- L1 Single problem fix → Low ROI
- L4 Capability building → High ROI (long-term)
Truman's mindset:Retrospective is earning back the money you spent in the past(Liu Chuanzhi's golden quote).
6. 6 Major Categories of Failure Modes (Comprehensive Classification)
Integrated from yitang + Wang Kai + Xiaohongshu + Lao Huang:
| Category | Manifestation | Typical Case |
|---|---|---|
| Track mismatch | Chose the wrong track (doing SEO for innovative products / doing WhatsApp for overseas push) | AiPathly / WhatsApp push bot |
| Cost calculation mismatch | Token costs unbalanced / revenue ceiling / insufficient ARPU | AI e-commerce guide site / FastFacts |
| Infrastructure blocked | Data permissions / platform review / legal compliance | Quanta Quest / mini-game review |
| 0-80 vs 80-100 | Vibe Coding makes 0-80 fast, but you die at 80-100 | Prompt House sub-pages / Ci Le Ge Ci |
| Ostrich mentality | Not thinking about traffic / payment / expansion | (Repeatedly warned by Wang Kai) |
| Tactical diligence covers up strategic laziness | Team works overtime every day but the direction is wrong | (Wang Xing's golden quote) |
7. How to Choose — Retrospective Path by Identity
| Who you are | Primary method | Starting point |
|---|---|---|
| Want to do a single-project retrospective | AAR 4 questions | Expected / Actual / Difference / Next time |
| Want to do a team-level retrospective | yitang 5 steps + 5 principles + 4 levels | Restore → Break down nodes → Evaluate → Extract → Archive |
| Want to read real failure cases | Wang Kai's 11 projects | 6 typical ways to fail + 7 lessons |
| Want to avoid Vibe Coding pitfalls | Xiaohongshu + Lao Huang | 0-80 vs 80-100 + cleanup prophecy + anti-patterns |
| Want to do long-term capability building | L4 Capability retrospective | Don't fix current problems, revise team building |
General Disciplines:
- Projects will most likely die, 90% barely make money(Wang Kai)
- Failure is 30 times more common than success(Wang Huiwen)
- Retrospective is earning back the money you spent in the past(Liu Chuanzhi)
- Do not use tactical diligence to cover up strategic laziness(Wang Xing)
- 3 months decides life or death(Indie development)
- 80→100 is the real life-or-death line(Xiaohongshu)
- Vibe Coding cannot replace architectural thinking(Lao Huang)
- Incorrect cost calculation is the number one cause of death for AI products(Wang Kai)
8. Reference List
- Yitang / First Lesson on In-Depth Retrospective —— 5-step method + 5 principles + 4 levels + AAR / Yu Lu's 5-year retrospective of Someet
- Wang Kai's Product Failure Retrospective Collection —— Detailed retrospectives of WhatsApp push / FastFacts / AiPathly / Quanta Quest
- Wang Kai's Startup Methodology and Business Judgment Framework in the AI Era §12 —— List of 11 firsthand products + common lessons + 6 ways to fail
- Wang Kai's RSI AI Investment Strategy and "Good Asset, Good Price, Good Volatility" Framework —— Negative example of buying Duolingo ten times in a row / lesson of high-priced ASML orders
- Xiaohongshu / Vibecoding Practical Collection (Games/Tools/Mini Programs) §2 —— Ci Le Ge Ci interview / 0-80 vs 80-100 life-or-death line
- newtype · AI Coding Practice and Toolstack Evolution §9.3 + §9.4 —— Vibe Coding Cleanup Prophecy / Anti-patterns
Related skill:
yitang-deep-retrospective—— In-Depth Post-Mortemyitang-flash-inspiration-soul—— Flash of Inspiration (Comparison)
Related topic wiki:
- Entrepreneurship Methodology (0 to 1) (Cross-Author Thematic Synthesis) —— Entrepreneurship Methodology (Batch 1)
- Lean Startup / MVP / Key Assumptions (Cross-Author Thematic Synthesis) —— Lean Startup (This Batch, Batch 7)
- ROI / Decision Evaluation System (Cross-Author Thematic Synthesis) —— ROI Decision-Making (Batch 6)
Sources and Related Materials
- Yitang / First Lesson on In-Depth Post-Mortems
- Wang Kai's Collection of Product Failure Post-Mortems
- Wang Kai's Entrepreneurship Methodology and Business Judgment Framework in the AI Era
- Wang Kai's RSI AI Investment Strategy and the "Good Asset, Good Price, Good Volatility" Framework
- Xiaohongshu / Vibecoding Hands-On Collection (Games/Tools/Mini Programs)
- newtype · AI Coding Practice and Toolstack Evolution
- Entrepreneurship Methodology (0 to 1) (Cross-Author Thematic Synthesis)
- Lean Startup / MVP / Key Assumptions (Cross-Author Thematic Synthesis)
- ROI / Decision Evaluation System (Cross-Author Thematic Synthesis)