Seedscore | Impact Assessment System — What's in this Package

Research documentation for Seedscore, an IAT Tool used for Accountable AI audit created for the Smol Gardens Project.

📋 What You Have

A complete, production-ready system for measuring and publishing the Human, Systems, and Environmental Impact of AI-assisted work.

The Core Problem

Organizations building with AI often ask:

"Is this actually beneficial for our team?" (Human Impact)
"Is this sustainable and resilient?" (Systems Impact)
"What's the environmental cost?" (Environmental Impact)

Existing frameworks don't adequately address all three dimensions together, and there's no standard way to compare impact across projects and organizations.

The Solution

A 15-question scorecard that generates comparable, publishable impact assessments. Think of it like:

A sustainability report for AI projects
A skills audit for team capabilities
A carbon footprint with human-workflow comparison

📦 Files in This Package

1. impact-assessment.html ⭐ START HERE

The working tool you use to assess projects

Single-file HTML (no dependencies, no server needed)
Runs entirely in your browser
4 tabs: Project Setup → Assessment → Dashboard → Report
Input: Project metadata + 13 questions
Output: Three-pillar scorecard + exportable JSON

How to use:

Open in any web browser
Enter project details (LLM, API calls, team size, etc.)
Answer 15 questions (5 min each)
Get scorecard and export results
Share the JSON report with your team

2. QUICK_START.md

For people who just want to use the tool

5-minute overview of what the tool does
How to interpret scores
Real example walkthrough
FAQ section
Common troubleshooting

Who reads this: Project managers, product owners, anyone running an assessment

3. IMPLEMENTATION_GUIDE.md

For people building on top of this or deploying it

Full technical architecture
The 13 questions with detailed rubrics
Data schema & JSON structure
Environmental calculation formulas
5-phase roadmap (MVP → persistence → benchmarking → integration → analytics)
Customization examples (adding questions, adjusting weights, etc.)
Publishing & citation guidelines

Who reads this: Engineers, architects, governance leads, open-source contributors

4. DATA_SCHEMA.json

Machine-readable schema for integrations

JSON Schema definitions for Project, Assessment, ScoreCard, Report
Complete example objects
Emissions data by LLM model
Human comparison calculations

Who uses this: API developers, database architects, data science teams

🎯 Quick Navigation

I want to assess my first project: → Open impact-assessment.html in your browser

I want to understand what I'm doing: → Read QUICK_START.md (20 min)

I want to customize or extend this: → Read IMPLEMENTATION_GUIDE.md (detailed technical guide)

I want to build an API or backend: → Reference DATA_SCHEMA.json for data structures

🚀 Getting Started (Next 30 Minutes)

Step 1: Download

Download all 4 files from this package. Put them in a folder.

Step 2: Run Your First Assessment

Double-click impact-assessment.html
Fill in project details (2 min)
Answer 13 questions (10 min)
View your scorecard (5 min)
Export JSON results (1 min)

Step 3: Share Results

Share the exported JSON with your team. Ask: "What do you think? Do you agree with these scores?"

Step 4: Plan Improvements

Based on your scores, identify the lowest pillar and plan improvements.

📊 The Assessment at a Glance

15 Questions, 3 Pillars

Human Impact (5 questions)

Core purpose alignment
Labor & skills uplift
Reusability by others
Cognitive sovereignty (users in control)
Collaboration & fairness

Systems Impact (5 questions)

Speed vs. cost tradeoff
Quality & trust
Vendor lock-in risk
Environmental resource use
System resilience & maintainability

Environment (3 questions)

Model efficiency
Impact tracking
Comparison to human workflow

Scoring: 0–2.0 per pillar

0 = Not Met (concern)
1 = Partially Met (progress)
2 = Fully Met (strong alignment)

Output: Three-Pillar Scorecard

Human:        1.4/2.0  ✓ Good
Systems:      0.8/2.0  ⚠ Needs work (vendor lock-in)
Environment:  0.7/2.0  ⚠ Needs work (improve measurement)
Overall:      1.0/2.0

🔬 Example: Real Project Assessment

Project: Content Generation Pipeline

LLM: Claude 3.5 Sonnet
API Calls (30 days): 1,500
Team: 3 people
Duration: 30 days
Outcome: 50% faster content, better consistency

Scores:

Human:        1.4   (well-aligned, but training needed)
Systems:      0.8   (moderate, but vendor risk)
Environment:  0.7   (saves 96× CO₂ vs. manual work!)

Key Finding: AI approach is much greener than human labor, but team needs more training on interpreting outputs.

🌍 Environmental Metrics

The tool estimates CO₂ footprint using:

Inference call count
Average tokens per call (~1,200)
Model-specific emissions (pre-loaded for Claude, GPT-4, Gemini, etc.)

Example:

1,500 calls × 1,200 tokens × 0.0000138 kg CO₂/token ÷ 1,000 = 0.025 kg CO₂

vs. Human alternative:
3 people × 30 days × 8 hours × 0.02 kg CO₂/hour = 2.4 kg CO₂

AI is 96× more efficient!

📈 5-Phase Implementation Roadmap

Phase 1: MVP ✓ (DONE)

Single-file HTML tool
15 questions, three pillars
JSON export

Phase 2: Persistence (1–2 weeks)

LocalStorage to save assessments
Multi-project tracking
Basic trending

Phase 3: Benchmarking (2–4 weeks)

Backend database
Aggregate results across orgs
Public comparison dashboard

Phase 4: Integration (4–8 weeks)

Claude.ai sidebar extension
Slack bot
GitHub Actions for CI/CD

Phase 5: Analytics (8–12 weeks)

Trend analysis (is Human Impact improving?)
Correlation analysis (System Impact vs Human Impact)
Governance rules & policies

🔧 How to Customize

Add Your Own Questions

Edit the questions array in the HTML:

const questions = [
  {
    id: 'myq1',
    category: 'human',  // or 'systems' or 'environment'
    title: 'Your question here?',
    context: 'Why it matters...',
    options: [
      { value: 0, label: 'Not Met', desc: '...' },
      { value: 1, label: 'Partially Met', desc: '...' },
      { value: 2, label: 'Fully Met', desc: '...' }
    ]
  }
];

Weight Pillars Differently

Change the overall calculation:

// Default: equal weight
overallScore = (human + systems + environment) / 3;

// Custom: Environment matters most (50%)
overallScore = (human × 0.25) + (systems × 0.25) + (environment × 0.5);

Add LLM-Specific Emissions Data

const emissionsData = {
  'Claude 3.5 Sonnet': 0.0000138,
  'GPT-4': 0.0000195,
  'Your Custom Model': 0.000012,
};

📚 What to Do With Your Results

Internally

Share the JSON report with stakeholders
Track scores over time (quarterly reassessments)
Identify trends (are we improving on human agency?)
Set organizational thresholds (e.g., "must score ≥1.0 before launch")

Externally (Contribute to Community)

Export your reports as JSON
Upload to a public GitHub repo (e.g., ai-impact-reports)
Add metadata: industry, project type, outcomes
Link to this methodology so others can replicate
Share findings in blog posts, papers, or conferences

For Research

Aggregate anonymized assessments to:

Identify which dimensions are most commonly weak
Find best practices in high-scoring projects
Benchmark by industry/model/company size
Inform policy & governance

❓ Common Questions

Q: Do I need a server? A: No. The HTML tool runs entirely in your browser. No login, no cloud required.

Q: Can I customize the questions? A: Yes. Edit the HTML questions array. Keep the standard 13 for comparability.

Q: How do I compare scores with other projects? A: Export the JSON and build a simple dashboard. The DATA_SCHEMA.json defines the format.

Q: Should every AI project be assessed? A: Recommend assessing all "significant" AI projects (>$1k cost, >10 team hours, customer-facing, new vendor).

Q: Can I integrate this with my existing tools? A: Yes. The data schema is straightforward JSON. APIs, bots, and integrations can consume it.

Q: How accurate are the environmental estimates? A: Rough but useful for comparison. For precise measurement, integrate Code Carbon.

🎓 Key Concepts

Concept	Meaning	Example
Human Impact	Does AI strengthen or weaken the team?	Team learns more about AI outputs (good) vs. blindly trusts them (bad)
Systems Impact	Is the architecture sustainable?	Multi-vendor architecture (resilient) vs. single vendor (fragile)
Environment	What's the carbon footprint?	AI uses 96× less energy than human work (very good)
Cognitive Sovereignty	Users understand & critique outputs	Team debates AI decisions vs. auto-accepts them
Vendor Lock-in	Dependency on single provider	Easy to switch models (good) vs. hard to migrate (bad)
Reusability	Can others use this work?	Open, documented, easy to adopt vs. proprietary, bespoke
Systemic Fragility	How easily does the system break?	Well-designed, tested vs. brittle, hard to maintain

🤝 Contributing

This is an open framework. Contributions welcome:

New question sets (for specific industries, use cases)
Integration examples (Slack bots, API servers, dashboards)
Emissions data (more precise per-model estimates)
Translations (make available in multiple languages)
Case studies (share your assessments and findings)

📄 License & Citation

License: CC BY-SA 4.0 (Creative Commons Attribution-ShareAlike)

You can use, modify, and distribute freely
You must give credit to the original creators
Derivative works must use the same license

How to cite:

@tool{seedscore-impact-assessment-2026,
  title={Seedscore Impact Assessment Tool: Measuring Human, Systems, and Environmental Impact of AI-enabled Work},
  author={[Krizia Fernando/Open Civic Tech]},
  year={2026},
  url={https://github.com/femmecubator/seedscore},
  license={CC BY-SA 4.0}
}

🚦 Next Steps

For Your First Assessment (Today)

✅ Open impact-assessment.html
✅ Fill in your most recent project
✅ Answer 13 questions (~10 min)
✅ Export and review results
✅ Share with your team

For Your Organization (This Week)

✅ Run 2–3 more assessments on different projects
✅ Collect feedback on questions
✅ Identify patterns (e.g., low Systems Impact across all projects?)
✅ Share findings in team meeting

For Long-Term Impact (This Month)

✅ Set quarterly assessment schedule
✅ Define organizational thresholds ("all new AI projects must score ≥1.0")
✅ Build simple tracking dashboard (Google Sheets, Airtable, etc.)
✅ Plan Phase 2 improvements (persistence, benchmarking, etc.)
✅ Contribute findings to public dataset

💬 Support

Questions or issues?

Check QUICK_START.md FAQ
Review IMPLEMENTATION_GUIDE.md for technical details
File an issue on GitHub

Want to contribute?

Fork the repo
Submit a pull request
Share your assessments & findings

Want to use commercially?

You can! CC BY-SA 4.0 allows commercial use
Just credit the original methodology

🎉 You're Ready!

You now have everything needed to: ✅ Assess your AI projects
✅ Understand impact across three dimensions
✅ Make data-driven decisions
✅ Share findings with your team and community
✅ Contribute to the collective knowledge about AI impact

Start with seedscore.html — open it now!

Last Updated: December 2024
Version: 1.0 MVP
Status: Production-ready for testing

🌱 Let's measure impact thoughtfully.

Name		Name	Last commit message	Last commit date
docs/plans		docs/plans
site		site
MVP_PLAN_WEEK1-2.md		MVP_PLAN_WEEK1-2.md
QUICK_START.md		QUICK_START.md
README.md		README.md
SMOL-GARDENS.md		SMOL-GARDENS.md

Uh oh!

Latest commit

History

Folders and files

Repository files navigation