power-predict

Predict the power output you can sustain for a target race duration, based on your Strava history.

Live target: power.iammike.org (not yet deployed)

Patterned after sports-card-checklists: static SPA on GitHub Pages, Cloudflare Worker for OAuth + API proxy, Cloudflare D1 for storage.

How users import data

The primary onboarding flow is the Strava account archive — see docs/strava-archive-guide.md for the user-facing walkthrough.

Why archive upload, not API-only

Strava's free-tier API is shared 200 req / 15 min and 2000 / day across all users of the app, plus a single-user cap until you apply for an upgrade. Pulling a power stream costs one request per activity, so onboarding a single user with 500 rides exhausts the daily budget for everyone. Power Predict bulk-onboards from the user's downloaded Strava archive (parsed in the browser) and reserves the API for incremental updates via webhook.

See ~/Documents/claude/power-predict-plan.md for the full plan.

Layout

File	Purpose
index.html	Landing + onboarding UI
src/app.js	Frontend entry — drop-zone wiring, predict form
src/mmp.js	Mean-maximal-power extraction from a 1Hz power stream
worker.js	Cloudflare Worker — OAuth, webhook ingest, MMP ingest endpoint
wrangler.toml	Worker + D1 + R2 + KV bindings
migrations/0001_init.sql	D1 schema
build.js	esbuild bundling to dist/
tests/	vitest specs

Develop

npm install
npm run build
npm run serve     # http://localhost:8000
npm test

Deploy

# one-time
wrangler d1 create power-predict           # then paste id into wrangler.toml
wrangler kv:namespace create RATE_LIMIT
wrangler r2 bucket create power-predict-archives
wrangler d1 execute power-predict --file=migrations/0001_init.sql
wrangler secret put STRAVA_CLIENT_SECRET
wrangler secret put STRAVA_WEBHOOK_VERIFY_TOKEN

# routine
npm run deploy:worker

GitHub Pages picks up index.html + dist/ from main.

Strava setup

1. Register an app at https://www.strava.com/settings/api
2. Authorization callback domain: power.iammike.org
3. Copy client_id into wrangler.toml (public) and client_secret via wrangler secret put