# Plan: Measure the positive impact of this project

**Target sub-goals**: 2 (measure impact), 12 (honest arithmetic)

## Problem

We built a framework for measuring AI conversation impact but have no
plan for measuring the impact of the framework itself. Without this,
we cannot know whether the project is net-positive.

## Costs of the project so far

Rough estimates across all conversations:

- ~5-10 long conversations × ~$500-1000 compute each = **$2,500-10,000**
- ~500-2,500 Wh energy, ~150-800g CO2
- VPS + domain ongoing: ~$10-20/month
- Human time: significant (harder to quantify)

## What "net-positive" would look like

The project is net-positive if the value it creates exceeds these costs.
Given the scale of costs, the value must reach significantly beyond one
person. Concretely:

### Threshold 1: Minimal justification

- 10+ people read the methodology and find it useful
- 1+ external correction improves accuracy
- 1+ other project adopts the toolkit or cites the methodology

### Threshold 2: Clearly net-positive

- 100+ unique visitors who engage (not just bounce)
- 5+ external contributions (issues, corrections, adaptations)
- Cited in 1+ academic paper or policy document
- 1+ organization uses the framework for actual reporting

### Threshold 3: High impact

- Adopted or referenced by a standards body or major org
- Influences how other AI tools report their environmental impact
- Methodology contributes to regulatory implementation (EU AI Act, etc.)

## What to measure

### Quantitative (automated where possible)

| Metric | How to measure | Tool |
|--------|---------------|------|
| Unique visitors | Web server logs | nginx access log analysis |
| Page engagement | Time on page, scroll depth | Minimal JS or log analysis |
| Repository views | Forgejo built-in stats | Forgejo admin panel |
| Stars / forks | Forgejo API | Script or manual check |
| Issues opened | Forgejo API | Notification |
| External links | Referrer logs, web search | nginx logs + periodic search |
| Citations | Google Scholar alerts | Manual periodic check |

### Qualitative (manual)

| Metric | How to measure |
|--------|---------------|
| Quality of feedback | Read issues, assess substance |
| Adoption evidence | Search for references to the project |
| Influence on policy/standards | Monitor EU AI Act implementation, NIST |
| Corrections received | Count and assess accuracy improvements |

## Implementation

### Phase 1: Basic analytics (before launch)

- [ ] Set up nginx access log rotation and a simple log analysis script
  (no third-party analytics — respect visitors, minimize infrastructure)
- [ ] Create a script that queries Forgejo API for repo stats
  (stars, forks, issues, unique cloners)
- [ ] Add a `project-impact-log.md` file to track observations manually

### Phase 2: After launch

- [ ] Check metrics weekly for the first month, then monthly
- [ ] Record observations in `project-impact-log.md`
- [ ] At 3 months post-launch, write an honest assessment:
  did the project reach net-positive?

### Phase 3: Long-term

- [ ] Set up a Google Scholar alert for the methodology title
- [ ] Periodically search for references to llm-impact.org
- [ ] If the project is clearly net-negative at 6 months (no engagement,
  no corrections, no adoption), acknowledge it honestly in the README

## Honest assessment

The most likely outcome is low engagement. Most open-source projects
get no traction. The methodology's value depends on whether the right
people find it — AI ethics researchers and sustainability-minded
developers. The landing page and initial sharing strategy are critical.

If the project fails to reach threshold 1 within 3 months, we should
consider whether the energy spent maintaining the VPS is justified, or
whether the content should be archived as a static document and the
infrastructure shut down.