mschwar · mschwar · Apr 24, 2026 · Apr 24, 2026
diff --git a/.github/workflows/validate.yml b/.github/workflows/validate.yml
@@ -0,0 +1,23 @@
+name: Validate Plots
+
+on:
+  push:
+  pull_request:
+
+jobs:
+  validate:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - run: python -m pip install --upgrade pip
+      - run: python -m pip install -r requirements.txt
+      - run: python build_all.py
+      - run: python scripts/generate_homepage.py
+      - run: python scripts/generate_readme_links.py
+      - run: python scripts/generate_sitemap.py
+      - run: python scripts/validate_all.py
+      - run: python scripts/check_links.py
+      - run: python scripts/check_accessibility_static.py
diff --git a/README.md b/README.md
@@ -1,169 +1,198 @@
-# Plots
+# Exponential Progress Atlas
 
-Interactive timelines of exponential tech progress – showing growth, compression, and scaling laws enabling modern AI.
+Interactive timelines showing how compute, energy, coordination, memory, and adoption compound into civilizational acceleration.
 
-## Quick Links
+The root inventory is [`plots_manifest.json`](plots_manifest.json). Homepage cards, README links, build ordering, dashboard lanes, and validation all read from that manifest.
 
-- [AI Compute Timeline (Interactive)](ai-compute-timeline/output/ai_compute_timeline_interactive.html)
-- [Adoption Timeline (Interactive)](adoption-timeline/output/adoption_timeline_interactive.html)
-- [Energetic Scaling (Interactive)](energetic-scaling/output/energetic_scaling_interactive.html)
-- [Civilization Scaling (Interactive)](civilization-scaling/output/civilization_scaling_interactive.html)
-- [Energy Leverage (Interactive)](energy-leverage-per-person/output/energy_leverage_interactive.html)
+## Published Inventory (9)
+
+- [AI Compute Timeline](ai-compute-timeline/output/ai_compute_timeline_interactive.html)
+- [Adoption Timeline](adoption-timeline/output/adoption_timeline_interactive.html)
+- [Energetic Scaling](energetic-scaling/output/energetic_scaling_interactive.html)
+- [Civilization Scaling](civilization-scaling/output/civilization_scaling_interactive.html)
+- [Energy Leverage](energy-leverage-per-person/output/energy_leverage_interactive.html)
+- [Model Sizes](model-sizes/output/model_sizes_interactive.html)
+- [AI Benchmark Progress](ai-benchmark-progress/output/benchmark_progress_interactive.html)
+- [Cost to Train](cost-to-train/output/cost_to_train_interactive.html)
+- [Unified Dashboard](dashboard/index.html)
 
 ---
 
-## 1. AI Compute Timeline (1900–2026)
+## 1. AI Compute Timeline
+
+Training compute from early electronic computing to frontier AI, with proxies and speculative projections labeled separately.
 
-Training FLOPs milestones for AI history – from vacuum tubes to frontier models.
+Hero stat: **10^27+ FLOPs**. Data confidence: **mixed**.
 
-- **Interactive**: [View Full Interactive Version (Plotly – Hover/Zoom)](ai-compute-timeline/output/ai_compute_timeline_interactive.html)
+- **Interactive**: [AI Compute Timeline](ai-compute-timeline/output/ai_compute_timeline_interactive.html)
 - **Static**: [PNG](ai-compute-timeline/output/ai_compute_timeline_highres.png) | [SVG](ai-compute-timeline/output/ai_compute_timeline.svg)
+- **Data**: [ai-compute-timeline/data/ai_milestones.csv](ai-compute-timeline/data/ai_milestones.csv)
+- **Metadata**: [ai-compute-timeline/data/meta.json](ai-compute-timeline/data/meta.json)
 - **Details**: [ai-compute-timeline/](ai-compute-timeline/)
 
-![AI Compute Timeline](ai-compute-timeline/output/ai_compute_timeline_highres.png)
-
-*Zoom recommended for 2010+ frontier cluster (10²⁴–10²⁷+ range).*
-
 ---
 
-## 2. Accelerating Paradigms in Computing & Connectivity (1957–2030+)
+## 2. Adoption Timeline
 
-Time to ~50M users adoption – shows exponential compression from years to days.
+Time-to-scale proxies across computing, connectivity, mobile, cloud, and AI paradigms.
 
-- **Interactive**: [View Full Interactive Version (Plotly – Hover/Zoom)](adoption-timeline/output/adoption_timeline_interactive.html)
+Hero stat: **60x faster**. Data confidence: **mixed**.
+
+- **Interactive**: [Adoption Timeline](adoption-timeline/output/adoption_timeline_interactive.html)
 - **Static**: [PNG](adoption-timeline/output/adoption_timeline_highres.png) | [SVG](adoption-timeline/output/adoption_timeline.svg)
+- **Data**: [adoption-timeline/data/tech_adoption.csv](adoption-timeline/data/tech_adoption.csv)
+- **Metadata**: [adoption-timeline/data/meta.json](adoption-timeline/data/meta.json)
 - **Details**: [adoption-timeline/](adoption-timeline/)
 
-![Adoption Timeline](adoption-timeline/output/adoption_timeline_highres.png)
-
-*From ~10 years (1957) to ~60 days (2022) – a 60x+ acceleration, with AI pushing toward near-instant scaling.*
-
 ---
 
-## 3. Energetic Scaling: Biology vs. Technology
+## 3. Energetic Scaling
 
-Neural efficiency vs. body size (Biology) and compute efficiency vs. time (Tech) – both reveal power laws.
+Biology, hardware efficiency, AI training compute, and foraging energetics compared with clean source datasets.
 
-- **Interactive**: [View Full Interactive Version (Plotly – Hover/Zoom)](energetic-scaling/output/energetic_scaling_interactive.html)
+Hero stat: **10^6x+ efficiency**. Data confidence: **mixed**.
+
+- **Interactive**: [Energetic Scaling](energetic-scaling/output/energetic_scaling_interactive.html)
 - **Static**: [PNG](energetic-scaling/output/energetic_scaling_highres.png) | [SVG](energetic-scaling/output/energetic_scaling.svg)
+- **Data**: [energetic-scaling/data/scaling_data.csv](energetic-scaling/data/scaling_data.csv)
+- **Metadata**: [energetic-scaling/data/meta.json](energetic-scaling/data/meta.json)
 - **Details**: [energetic-scaling/](energetic-scaling/)
 
-![Energetic Scaling](energetic-scaling/output/energetic_scaling_highres.png)
-
-*Humans are the biological outlier (EQ~7); AI is the technological outlier (75 quadrillion-fold compute/$ since 1939).*
-
 ---
 
-## 4. Scaling Civilization: Energy, Coordination, Memory, Replication
+## 4. Civilization Scaling
 
-Multi-lane log-time timeline showing how four fundamental metrics have scaled from 1M years ago to 2030+.
+Five civilizational lanes: energy, coordination, memory, replication, and latency over log-time.
 
-- **Interactive**: [View Full Interactive Version (Plotly – Hover/Zoom)](civilization-scaling/output/civilization_scaling_interactive.html)
+Hero stat: **5 lanes**. Data confidence: **mixed**.
+
+- **Interactive**: [Civilization Scaling](civilization-scaling/output/civilization_scaling_interactive.html)
 - **Static**: [PNG](civilization-scaling/output/civilization_scaling_highres.png) | [SVG](civilization-scaling/output/civilization_scaling.svg)
+- **Data**: [civilization-scaling/data/civilization_metrics.csv](civilization-scaling/data/civilization_metrics.csv)
+- **Metadata**: [civilization-scaling/data/meta.json](civilization-scaling/data/meta.json)
 - **Details**: [civilization-scaling/](civilization-scaling/)
 
-![Civilization Scaling](civilization-scaling/output/civilization_scaling_highres.png)
-
-*Log-time compresses ~99% of human existence (pre-writing) into the left side, expanding modern acceleration on the right. Phase flips (Fire → Agriculture → Writing → Printing → Internet → AI) stack to enable exponential progress.*
-
 ---
 
-## 5. Energy Leverage per Person (NEW)
+## 5. Energy Leverage
 
-How much total energy does an average human command compared to the metabolic baseline (~114 W)?
+Per-person energy command relative to the metabolic baseline, with period anchors labeled explicitly.
 
-- **Interactive**: [View Full Interactive Version (Plotly – Hover/Zoom)](energy-leverage-per-person/output/energy_leverage_interactive.html)
+Hero stat: **17x body energy**. Data confidence: **high**.
+
+- **Interactive**: [Energy Leverage](energy-leverage-per-person/output/energy_leverage_interactive.html)
 - **Static**: [PNG](energy-leverage-per-person/output/energy_leverage_highres.png) | [SVG](energy-leverage-per-person/output/energy_leverage.svg)
+- **Data**: [energy-leverage-per-person/data/energy_leverage_datapoints.csv](energy-leverage-per-person/data/energy_leverage_datapoints.csv)
+- **Metadata**: [energy-leverage-per-person/data/meta.json](energy-leverage-per-person/data/meta.json)
 - **Details**: [energy-leverage-per-person/](energy-leverage-per-person/)
 
-![Energy Leverage](energy-leverage-per-person/output/energy_leverage_highres.png)
+---
+
+## 6. Model Sizes
 
-*Humans went from ~1–2× body energy (foragers) to ~17× body energy (modern). The post-1750 coal/steam and post-1950 oil/electricity jumps dominate the visual story.*
+Language model parameter counts over time, separating disclosed counts from estimates and unreleased projections.
+
+Hero stat: **1.5B -> 5T params**. Data confidence: **speculative**.
+
+- **Interactive**: [Model Sizes](model-sizes/output/model_sizes_interactive.html)
+- **Static**: [PNG](model-sizes/output/model_sizes_highres.png) | [SVG](model-sizes/output/model_sizes.svg)
+- **Data**: [model-sizes/data/llm_model_sizes.csv](model-sizes/data/llm_model_sizes.csv)
+- **Metadata**: [model-sizes/data/meta.json](model-sizes/data/meta.json)
+- **Details**: [model-sizes/](model-sizes/)
 
 ---
 
-## Why These Plots?
+## 7. AI Benchmark Progress
 
-| Timeline | Shows | Trend |
-|----------|-------|-------|
-| **Compute** | Training FLOPs (10⁰ → 10²⁷) | Exponential **growth** ↑ |
-| **Adoption** | Time to 50M users | Exponential **compression** ↓ |
-| **Energetic** | Neurons/kg & cps/$ | **Power laws** (log-log linear) |
-| **Civilization** | Energy/Coord/Memory/Repl | **Stacking infrastructure** layers |
-| **Energy Leverage** | Watts/person vs metabolic | **~17× body energy** (modern) |
+Benchmark progress against human baselines across knowledge, coding, software engineering, and reasoning tasks.
 
-Together they illustrate:
-- **Compute**: Raw exponential growth enabling AI scale
-- **Adoption**: Ecosystem acceleration compressing timelines
-- **Energetic**: Fundamental scaling rules – humans and AI as outliers
-- **Civilization**: How infrastructure layers (fire → writing → internet → AI) enable each successive leap
-- **Energy Leverage**: Per-person energy command from foragers (~2×) to modern (~17×)
+Hero stat: **4 benchmark lanes**. Data confidence: **mixed**.
 
-Inspired by Kurzweil, [Epoch AI](https://epochai.org/), [Statista](https://www.statista.com/), [Asymco](http://www.asymco.com/), Herculano-Houzel (neuronal scaling), Kleiber (metabolic 0.75), Kaplan/Charnov (LHT/OFT).
+- **Interactive**: [AI Benchmark Progress](ai-benchmark-progress/output/benchmark_progress_interactive.html)
+- **Static**: [PNG](ai-benchmark-progress/output/benchmark_progress_highres.png) | [SVG](ai-benchmark-progress/output/benchmark_progress.svg)
+- **Data**: [ai-benchmark-progress/data/benchmark_data.csv](ai-benchmark-progress/data/benchmark_data.csv)
+- **Metadata**: [ai-benchmark-progress/data/meta.json](ai-benchmark-progress/data/meta.json)
+- **Details**: [ai-benchmark-progress/](ai-benchmark-progress/)
 
 ---
 
-## Repository Structure
+## 8. Cost to Train
 
-Each plot follows a standardized structure:
+Training cost, FLOPs, and capability over time, showing the efficiency paradox at the frontier.
 
-```
-<plot-name>/
-├── index.html              # Interactive page (uses shared/site.css)
-├── data/
-│   ├── <slug>.csv          # Source data
-│   └── meta.json           # Metadata: title, description, fields, sources
-├── output/
-│   ├── *_interactive.html  # Plotly interactive chart
-│   ├── *_highres.png       # High-res PNG export
-│   └── *.svg               # SVG vector export
-├── src/
-│   ├── *.py                # Matplotlib static generator
-│   └── *_plotly.py         # Plotly interactive generator
-└── README.md
-```
-
-### Shared Assets
+Hero stat: **$/FLOP collapse**. Data confidence: **mixed**.
 
-- `shared/site.css` – Common styles for all pages
-- `shared/site.js` – Navigation bar injection
-- `scripts/validate_all.py` – Validate all plots (run: `python scripts/validate_all.py`)
+- **Interactive**: [Cost to Train](cost-to-train/output/cost_to_train_interactive.html)
+- **Static**: [PNG](cost-to-train/output/cost_to_train_highres.png) | [SVG](cost-to-train/output/cost_to_train.svg)
+- **Data**: [cost-to-train/data/training_costs.csv](cost-to-train/data/training_costs.csv)
+- **Metadata**: [cost-to-train/data/meta.json](cost-to-train/data/meta.json)
+- **Details**: [cost-to-train/](cost-to-train/)
 
 ---
 
-## Development
+## 9. Unified Dashboard
 
-### Install dependencies
+A synchronized overview of the atlas inventory using the same manifest as the homepage, README, build, and validator.
 
-```bash
-pip install -r requirements.txt
-```
+Hero stat: **9 atlas entries**. Data confidence: **mixed**.
 
-### Regenerate all plots
+- **Interactive**: [Unified Dashboard](dashboard/index.html)
+- **Data**: [plots_manifest.json](plots_manifest.json)
+- **Metadata**: [plots_manifest.json](plots_manifest.json)
+- **Details**: [dashboard/](dashboard/)
 
-```bash
-python build_all.py
-```
 
-### Regenerate a specific plot
+---
+
+## Data Contracts
 
-```bash
-cd <plot-name>/src
-python <slug>.py           # Static PNG/SVG
-python <slug>_plotly.py    # Interactive HTML
+- `ai-compute-timeline/data/ai_milestones.csv` uses normalized fields: `year,event,category,value_numeric,value_low,value_high,value_unit,estimate_status,source_id,confidence,display_label,notes`.
+- `adoption-timeline/data/tech_adoption.csv` includes `adoption_metric_type`, `comparability_level`, `source_id`, `confidence`, and notes so unlike adoption proxies are not treated as perfectly comparable.
+- Energetic Scaling keeps comparison-level data in `scaling_data.csv` and splits clean source contracts into `biology_neural_scaling.csv`, `hardware_efficiency.csv`, `ai_training_flops.csv`, and `foraging_lht.csv`.
+
+## Repository Structure
+
+Each plot should follow this structure:
+
+```text
+<plot-name>/
+├── index.html
+├── data/
+│   ├── <slug>.csv
+│   └── meta.json
+├── output/
+│   ├── *_interactive.html
+│   ├── *_highres.png
+│   └── *.svg
+├── src/
+│   ├── *.py
+│   └── *_plotly.py
+└── README.md
 ```
 
-### Validate all plots
+## Development
 
 ```bash
+python -m pip install -r requirements.txt
+python build_all.py
+python scripts/generate_homepage.py
+python scripts/generate_readme_links.py
 python scripts/validate_all.py
+python scripts/check_links.py
+python scripts/check_accessibility_static.py
 ```
 
----
+## Adding a New Plot
+
+1. Create the standard plot directory structure.
+2. Add data, metadata, generator scripts, output paths, and README.
+3. Add the entry to `plots_manifest.json` with `status: "draft"` until outputs and sources pass validation.
+4. Run the build, generators, validators, link checker, and accessibility checker.
+5. Change `status` to `"published"` only when the plot should appear on the homepage and dashboard.
 
-## Contributing
+## Deployment
 
-Ideas, new milestones, or bug reports welcome! See [CONTRIBUTING.md](CONTRIBUTING.md).
+GitHub Pages deploys should run the same validation commands in CI before publishing. A failed build, broken relative link, missing alt text, stale output, or manifest mismatch should block deployment.
 
 ## License
 

diff --git a/adoption-timeline/data/meta.json b/adoption-timeline/data/meta.json
@@ -1,34 +1,51 @@
 {
   "title": "Adoption Timeline",
-  "description": "Time to ~50M users adoption, showing exponential compression from years to days (1957-2026).",
+  "description": "Time-to-scale proxies across technology paradigms. The chart labels metric type and comparability so FORTRAN, ARPANET, iPhone, ChatGPT, and agentic tools are not treated as perfectly equivalent.",
   "fields": {
-    "Year": "Launch year of technology/product",
+    "Year": "Launch year of technology or product",
     "Event": "Name of technology or product",
-    "Category": "Tech category (Hardware, Software, Internet/Web, Mobile, Social/Apps, Cloud, AI/Agentic)",
-    "Days_to_Adoption": "Days to reach ~50M users or equivalent adoption milestone",
-    "Impact": "Historical impact level (Transformative, High, Medium, Speculative)"
+    "Category": "Technology category",
+    "Days_to_Adoption": "Days to reach the stated proxy scale",
+    "Impact": "Historical impact level",
+    "adoption_metric_type": "Metric type: users, developers, devices, accounts, installs, or organizations",
+    "comparability_level": "strict_proxy, comparable_proxy, rough_analogy, or projection",
+    "source_id": "Source identifier for auditable rows",
+    "confidence": "Qualitative confidence",
+    "comparability_notes": "Why the proxy is or is not comparable with other rows",
+    "notes": "Additional audit notes"
   },
   "sources": [
     {
+      "id": "statista",
       "name": "Statista",
       "url": "https://www.statista.com/",
       "accessed": "2026-01",
       "notes": "User growth and adoption data"
     },
     {
+      "id": "asymco",
       "name": "Asymco",
       "url": "http://www.asymco.com/",
       "accessed": "2026-01",
       "notes": "Historical technology adoption curves"
     },
     {
+      "id": "epoch",
       "name": "Epoch AI",
       "url": "https://epochai.org/",
       "accessed": "2026-01",
       "notes": "AI adoption metrics"
+    },
+    {
+      "id": "source_review_needed",
+      "name": "Source review needed",
+      "url": "https://github.com/mschwar/plots",
+      "accessed": "2026-04",
+      "notes": "Projection retained for chart continuity but flagged for source review"
     }
   ],
-  "transformations": "Log10(days) for y-axis. Midpoints used for date ranges. Pre-1990 estimates approximate.",
+  "transformations": "Log10(days) for y-axis. Metric types and comparability levels control marker shape. Trend line is a visual guide, not a causal model.",
   "created": "2026-01",
+  "last_updated": "2026-04-24",
   "author": "mschwar"
 }