Token Economy Tracker

Live dashboard → https://charlesliuyx.github.io/Token-Economy-Tracker/

A daily auto-updating AI monetization dashboard. The build artifact is a pure static, single-file HTML page (Chart.js inlined, no CDN, works offline), refreshed once a day by GitHub Actions and served on GitHub Pages.

What it tracks

1. Frontier Lab ARR (live estimate) — big live-ticking ARR numbers for Anthropic / OpenAI, built from publicly disclosed anchors + curve fitting + extrapolation with confidence bands.
2. Token demand & adoption — OpenRouter model token usage (per-model, model comparison, network-wide total, weekly Top Models), Vercel AI Gateway share (Token / $ Spend Top 10 + trends), SDK downloads (npm / PyPI), expert quotes, and Epoch AI Top Sites.
3. GPU rental prices — $/GPU-hr cards for H100 / H200 / B200 / A100 / RTX 5090, plus price trend charts.
4. AI datacenter buildout — summary cards from the Epoch AI Frontier Data Centers dataset (site count / IT power GW / H100-equivalents), cumulative industry growth, and a daily news feed.

How it works

GitHub Actions cron (daily)
        │
        ▼
scripts/fetch/*        one independent fetcher per source; a single source
        │              failing never blocks the run
        ▼
data/                  append-only daily JSON snapshots + latest.json per source
        │
        ▼
scripts/build.py       renders everything into one file
        │
        ▼
site/index.html        single-file static dashboard (GitHub Pages / open locally)

Design principles: fetch and build are fully decoupled (either half can be rerun independently), history is append-only (any chart can be redrawn from raw snapshots at any time), and there is no server or runtime network request.

Stack: Python 3.12 (requests / pyyaml / numpy / jinja2 — deliberately minimal) and vanilla JS with a vendored Chart.js inlined at build time.

Data sources

Source	Kind	What it provides
OpenRouter	unofficial API	Model token usage + weekly rankings
Vercel AI Gateway	unofficial API	Gateway token / spend leaderboards
npm + PyPI	official APIs	AI SDK download counts
Epoch AI	official dataset (CC-BY)	Frontier data centers dataset
GPU price indexes	web	GPU rental $/GPU-hr
Google News RSS	unofficial API	Datacenter news feed (domain-whitelisted)
Manual anchors	git-maintained	Publicly disclosed ARR data points

Each source is registered in data/sources.yml and documented in docs/sources/.

Quick start

make venv     # create virtualenv and install dependencies
make update   # run all fetchers (each source independently)
make build    # data/ -> site/index.html
make serve    # preview locally
make test     # run the test suite

make fetch-<source>   # debug a single fetcher, e.g. make fetch-openrouter

Hard rules

• data/**/daily/ snapshots are immutable (append-only history).
• site/index.html is a build artifact — never edit it by hand; only make build may generate it.
• No external CDNs or runtime network requests; the page must open offline.
• Behavior changes must update the corresponding docs in the same PR.
• A new data source ships as a trio: data/sources.yml entry + docs/sources/<source>.md + a response sample.

Commit message convention

Format: <type>: <summary> — written in English, imperative mood, no trailing period, summary ≤ 72 characters. Add a body only when the why isn't obvious.

Type	Use for
feat	New panel, fetcher, or user-visible capability
fix	Bug fix in pipeline, build, or frontend
data	Data snapshot updates or manual anchor additions
docs	Documentation only
build	Build script, Makefile, or dependency changes
ci	GitHub Actions workflow changes
refactor	Code restructuring with no behavior change
test	Test-only changes
chore	Anything else (housekeeping)

Examples:

feat: add GPU price trend panel
data: add 2 ARR anchors (YipitData, as of 2026-06-30)
fix: handle empty OpenRouter rankings response

Automated daily commits from CI use data: YYYY-MM-DD.

Documentation

Docs follow progressive disclosure: start at AGENTS.md (entry point with a task routing table), then docs/INDEX.md for the full map — plan, architecture, conventions, runbook, ARR methodology, per-source docs, and ADRs. Most internal docs are written in Chinese.