docs: rewrite deployment section for two-image setup

- Document web/:web and scraper/:scraper image split
- Full first-time setup flow: pull → discover → RCDB IDs → scrape → up
- Document all scraper env vars (TZ, PARK_HOURS_STALENESS_HOURS, COASTER_STALENESS_HOURS)
- Add manual scrape commands, update workflow, add npm test to dev section

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

README.md

@@ -63,7 +63,7 @@ Scrape operating hours for the full year:
 npm run scrape
 ```
 
-Force a full re-scrape (ignores the 7-day staleness window):
+Force a full re-scrape (ignores the staleness window):
 
 ```bash
 npm run scrape:force
@@ -79,6 +79,12 @@ npm run debug -- --park kingsisland --date 2026-06-15
 
 Output is printed to the terminal and saved to `debug/{parkId}_{date}.txt`.
 
+### Run tests
+
+```bash
+npm test
+```
+
 ### Run the dev server
 
 ```bash
@@ -91,33 +97,93 @@ Open [http://localhost:3000](http://localhost:3000). Navigate weeks with the `
 
 ## Deployment
 
-The app uses Next.js standalone output. The SQLite database is stored in a Docker volume at `/app/data`.
+The app ships as two separate Docker images that share a named volume for the SQLite database:
+
+| Image | Tag | Purpose |
+|-------|-----|---------|
+| Next.js web server | `:web` | Reads DB, serves content. No scraping tools. |
+| Scraper + scheduler | `:scraper` | Nightly data refresh. No web server. |
+
+Images are built and pushed automatically by CI on every push to `main`.
+
+### First-time setup
+
+**1. Pull the images**
+
+```bash
+docker pull gitea.thewrightserver.net/josh/sixflagssupercalendar:web
+docker pull gitea.thewrightserver.net/josh/sixflagssupercalendar:scraper
+```
+
+**2. Discover park API IDs**
+
+This one-time step opens a headless browser for each park to find its internal Six Flags API ID. Run it against the scraper image so Playwright is available:
+
+```bash
+docker run --rm -v thoosie_park_data:/app/data \
+  gitea.thewrightserver.net/josh/sixflagssupercalendar:scraper \
+  npm run discover
+```
+
+**3. Set RCDB IDs for the coaster filter**
+
+Open `data/park-meta.json` in the Docker volume and set `rcdb_id` for each park to the numeric ID from the RCDB URL (e.g. `https://rcdb.com/4529.htm` → `4529`). You can curl it directly from the repo:
+
+```bash
+curl -o /var/lib/docker/volumes/thoosie_park_data/_data/park-meta.json \
+  https://gitea.thewrightserver.net/josh/SixFlagsSuperCalendar/raw/branch/main/data/park-meta.json
+```
+
+**4. Run the initial scrape**
+
+```bash
+docker run --rm -v thoosie_park_data:/app/data \
+  gitea.thewrightserver.net/josh/sixflagssupercalendar:scraper \
+  npm run scrape
+```
+
+**5. Start services**
 
 ```bash
 docker compose up -d
 ```
 
-### Seed the database inside the container
+Both services start. The scraper runs nightly at 3 AM (container timezone, set via `TZ`).
 
-The production image includes Playwright and Chromium, so discovery and scraping run directly against the container's volume:
+### Updating
 
 ```bash
-docker compose exec web npm run discover
-docker compose exec web npm run scrape
+docker compose pull && docker compose up -d
 ```
 
-Or as a one-off against the named volume:
+### Scraper environment variables
+
+Set these in `docker-compose.yml` under the `scraper` service to override defaults:
+
+| Variable | Default | Description |
+|----------|---------|-------------|
+| `TZ` | `UTC` | Timezone for the nightly 3 AM run (e.g. `America/New_York`) |
+| `PARK_HOURS_STALENESS_HOURS` | `72` | Hours before park schedule data is re-fetched |
+| `COASTER_STALENESS_HOURS` | `720` | Hours before RCDB coaster lists are re-fetched (720 = 30 days) |
+
+### Manual scrape
+
+To trigger a scrape outside the nightly schedule:
 
 ```bash
-docker run --rm -v sixflagssupercalendar_park_data:/app/data \
-  gitea.thewrightserver.net/josh/sixflagssupercalendar:latest \
-  npm run scrape
+docker compose exec scraper npm run scrape
+```
+
+Force re-scrape of all data (ignores staleness):
+
+```bash
+docker compose exec scraper npm run scrape:force
 ```
 
 ---
 
 ## Data Refresh
 
-The scraper skips any park + month already scraped within the last 72 hours. The nightly Docker scraper service handles this automatically. Parks or months not yet in the database show a `—` placeholder; parks with no open days in the displayed week are hidden from the calendar automatically.
+The scraper skips any park + month already scraped within the staleness window (`PARK_HOURS_STALENESS_HOURS`, default 72h). Past dates are never overwritten — once a day occurs, the API stops returning data for it, so the record written when it was a future date is preserved forever. The nightly scraper handles refresh automatically.
 
-Roller coaster lists (from RCDB) are refreshed every 30 days on each `npm run scrape` run, for parks with a configured `rcdb_id`.
+Roller coaster lists (from RCDB) are refreshed per `COASTER_STALENESS_HOURS` (default 720h = 30 days) for parks with a configured `rcdb_id`.