OSS Investment Scorecard

开源项目投资评估框架 · AI周期专版

A structured, weighted scoring framework for USD-denominated VC funds evaluating open source projects during the AI technology acceleration cycle.

Built from practice, not theory — calibrated against real deals including vLLM/Inferact ($150M @ $800M) and Hugging Face ($235M @ $4.5B).

About the author: Lucy Chen is an EIR (Entrepreneur in Residence) at Zoo Capital, a Singapore-based VC fund with USD $2B+ AUM focused on open-source AI investing. She has 15+ years in product commercialization and capital strategy, with two prior startup exits. This framework was developed from direct evaluation work — not synthesized from secondary sources.

📢 V1.2 Update: From "Scoring Rubric" to "Diligence Protocol"

March 2026 Update: We have officially released V1.2 of the framework. This update is a structural leap designed to eliminate bias when evaluating early-stage projects (Seed/Series A) where public data is often scarce.

Key Improvements in V1.2:

• Mandatory Fact Sheet: A 7-item pre-evaluation gate including competitive benchmarking.
• Indirect Signal Inference: Guidelines for using SOC2, billing complexity, and hiring signals to estimate traction when ARR is not public.
• Project Age Calibration: Prioritizing Velocity over absolute levels for projects <12 months old.
• Narrative Pivot Exemptions: Distinguishing market-following pivots from failure-driven ones.

👉 View V1.2 Full Framework (SKILL.md) | View V1.2 Release Notes

📖 Framework Overview

Score Thresholds:

• 🟢 8.5–10.0 → Strongly Recommend
• 🟡 7.0–8.4 → Recommend with Conditions
• 🟠 5.5–6.9 → Watch / Track (re-evaluate in 6-9 months)
• 🔴 < 5.5 → Pass

One-Vote Vetoes: 6 conditions that trigger automatic Pass regardless of total score. See SKILL.md for full details.

🛡️ Optional Module: Star Health & Anti-Fraud Protocol

In the AI-cycle, GitHub stars have become a highly manipulable vanity metric. To ensure the integrity of Dimension A (Ecosystem Health), analysts may optionally trigger the Star Health Protocol to detect "surface-level" popularity vs. real developer utility.

When to use:

• Projects in hyper-hyped sectors (e.g., AI Agents, RAG wrappers).
• Projects with >20% MoM star growth but low issue activity.
• Pre-due diligence for Seed/Series A rounds.

Core Health Metrics:

Integration Note: If the Star Health Protocol is active and returns "Weak" signals, the score for Dimension A should be penalized by 1.0 - 2.0 points, regardless of absolute star count.

❓ Frequently Asked Questions

Q: What metrics actually matter when evaluating an open source AI project for investment — beyond GitHub stars?

Stars measure visibility, not investability. The framework weights four higher-signal metrics instead:

• PR velocity (commits per week, time-to-merge) — measures active development health
• Production dependents — how many real projects import this as a dependency
• Contributor diversity — single-maintainer projects fail at scale regardless of star count
• Revenue quality — ARR from enterprise contracts outweighs usage-based or donation income

A project with 2,000 stars and 40 production dependents is more investable than one with 40,000 stars and 3 dependents. LLaMA-Factory has 68K stars and scores borderline Pass on this framework.

Q: What are the automatic disqualifiers — red flags that kill a deal regardless of total score?

Six One-Vote Vetoes trigger an automatic Pass regardless of weighted score:

1. Core IP not owned by the entity — contributor IP not properly assigned
2. License incompatibility — copyleft obligations that block commercial use
3. Single-maintainer bus factor with no succession plan
4. No path to US market access — critical for USD-denominated exit
5. Active litigation or unresolved patent claims on foundational technology
6. Fabricated benchmarks — any evidence of manipulated performance claims

A project scoring 8.2/10 on dimensions still receives a Pass if any veto applies.

Q: How does this framework handle projects where ARR or revenue isn't public?

V1.2 introduced indirect signal inference for early-stage projects (Seed/Series A):

• SOC2 Type II certification → signals enterprise sales motion is underway
• Billing complexity in public API docs → indicates commercial tiers exist
• Senior GTM/sales hires on LinkedIn → revenue pursuit without disclosed numbers
• Enterprise case studies on website → production adoption even without ARR disclosure

"No public revenue data" ≠ "no revenue." The framework explicitly distinguishes these.

Q: How do open source AI projects monetize — and which models are most investable?

Revenue quality hierarchy (highest to lowest investability):

1. ARR from enterprise contracts — predictable, sticky, highest multiple
2. Usage-based cloud revenue — scales with adoption
3. Dual-license commercial tier — proven OSS model (HashiCorp, Elastic)
4. Professional services / consulting — doesn't scale, weak signal
5. Donations / sponsorships only — red flag, no commercial validation

Projects like Unsloth (8.10/10) score high despite no disclosed ARR because 150M+ downloads and YC backing signal imminent commercial traction.

Q: What's the difference between a "Watch" verdict and a "Pass"?

• Watch (5.5–6.9): Real signals exist but 1–2 critical elements are missing — usually PMF evidence or GTM capability. Re-evaluate in 6–9 months. These are pipeline entries, not rejections.
• Pass (<5.5): Fundamental structural issues — missing moat, no community traction, or a One-Vote Veto applies.

DeerFlow (6.15, Watch ⚠️Corp) is Watch because ByteDance backing structurally limits exit potential — but the technology is real. WFGY (5.80, Watch) is Watch because PMF is unproven, not because the technical thesis is wrong.

Q: How does this compare to standard VC due diligence for SaaS or traditional software?

Three gaps this framework fills that standard VC frameworks miss:

1. Community health as a first-class signal — 25% weight on ecosystem health reflects that open source distribution is the moat, not a feature.
2. Corp flag discipline — projects backed by Alibaba, ByteDance, or Ant Group are flagged (⚠️Corp) because corporate ownership structurally constrains exit optionality regardless of technology quality.
3. Calibrated anchors — every score is relative to vLLM (8.9/10, $800M valuation) and Hugging Face (8.35/10, $4.5B valuation). Without anchors, scores are opinions.

📁 Files

🚀 How to Use

Option A — Use with Claude AI

1. Download oss-investment-scorecard.skill
2. Go to Claude.ai → Settings → Skills → Upload
3. Ask Claude: "Evaluate [project name] for open source VC investment"
4. Claude will apply the full framework automatically

Option B — Manual Evaluation

1. Open template/evaluation-template.md
2. Fill in each dimension with your research
3. Calculate weighted score
4. Submit your evaluation (see below)

Option C — Use with Any LLM Agent

Works with GPT-4, Gemini, OpenClaw, Cursor, Manus, or any agent that accepts a system prompt.

1. Open SKILL.md in this repository
2. Copy everything from line 17 onwards (skip the YAML header between the --- markers at the top)
3. Paste into your agent's system prompt or context window
4. Ask: "Evaluate [project name] for open source VC investment"

📬 Submit Your Evaluation — Connect with Investors & Founders

Why submit?

This repository is maintained by Lucy Chen, EIR (Entrepreneur in Residence) at Zoo Capital, a Singapore-based VC fund with USD $2B+ AUM, focused on broad open-source project investing.

When you submit an evaluation, two things happen:

1. Your evaluation becomes part of the public record — other investors and founders can see which projects have been assessed
2. You get optionally connected — if you're an investor looking for deal flow, or a founder wanting investor feedback, Lucy can introduce relevant parties

Who should submit:

• 🔍 Investors who evaluated a project and want deal-sharing partners or co-investors
• 🏗️ Founders who want their project professionally scored and introduced to investors
• 📊 Analysts building open-source investment theses

How to Submit

→ Submit via GitHub Issue

Or reach Lucy directly:
📧 ossinvestor.2026@gmail.com
💼 LinkedIn: linkedin.com/in/lucycxy
📘 Facebook: facebook.com/lucy.chen.908347
🌐 Fund: zoocap.com

📊 Evaluated Projects (Community Submissions)

Investment Scores Overview (Top Projects)

Potential vs. Independence Matrix

quadrantChart
    title "Independence vs. Investment Potential (Numbered Map)"
    x-axis "Low Independence" --> "High Independence"
    y-axis "Low Potential" --> "High Potential"
    quadrant-1 "Invest Track"
    quadrant-2 "Watch & Verify"
    quadrant-3 "Pass"
    quadrant-4 "Corp Asset"
    "1": [0.90, 0.92]
    "2": [0.85, 0.88]
    "3": [0.88, 0.85]
    "4": [0.82, 0.82]
    "5": [0.80, 0.78]
    "6": [0.55, 0.76]
    "7": [0.65, 0.76]
    "8": [0.52, 0.68]
    "9": [0.72, 0.62]
    "10": [0.75, 0.60]
    "11": [0.15, 0.58]
    "12": [0.70, 0.55]
    "13": [0.60, 0.52]
    "14": [0.45, 0.51]
    "15": [0.78, 0.80]

Legend:

1. vLLM | 2. HuggingFace | 3. Unsloth | 4. LMCache | 5. Hindsight | 6. DeepAgents | 7. AReaL | 8. AgentScope | 9. TradingAgents | 10. Hermes | 11. DeerFlow | 12. WFGY | 13. MiroFish | 14. Aryn/Sycamore | 15. Infisical Agent Vault

This table is updated as community submissions are reviewed. Submit yours →

🤝 Contributing

• Improve the framework: Open a PR with proposed changes to SKILL.md
• Add a case study: Submit a scored evaluation via Issue
• Translate: Chinese/English versions both welcome

📄 License

MIT — use freely, attribution appreciated.

Maintained by Lucy Chen · Zoo Capital · Last updated: April 2026 (v1.2 Roadmap)