diff --git a/apps/docs/architecture/tiered-execution.md b/apps/docs/architecture/tiered-execution.md
new file mode 100644
index 0000000..ef745d1
--- /dev/null
+++ b/apps/docs/architecture/tiered-execution.md
@@ -0,0 +1,91 @@
+---
+title: Tiered Execution
+description: How TRAWL escalates through four tiers — from a plain fetch to a residential proxy solve.
+---
+
+# Tiered Execution
+
+Every scrape request runs through a four-tier waterfall. Each tier is only tried if the previous one fails or is skipped. This means you pay the cheapest cost that works for each request — most requests never need a browser.
+
+```
+Request
+  │
+  ▼
+Tier 1: Plain HTTP Fetch ─── success ──→ return (< 100ms)
+  │ blocked / needs-js
+  ▼
+Tier 2: Cached Session ────── success ──→ return (~500ms)
+  │ blocked / cache miss
+  ▼
+Tier 3: Fresh Challenge ───── success ──→ cache cookies, return (4–15s)
+  │ IP flagged
+  ▼
+Tier 4: Residential Proxy ─── success ──→ cache cookies, return (15–45s)
+  │ failed
+  ▼
+  error
+```
+
+## Tier 1 — Plain HTTP Fetch
+
+The cheapest tier. Uses Bun's native `fetch()` with a realistic browser header set:
+
+```
+User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) Chrome/131...
+Accept: text/html,application/xhtml+xml,...
+Accept-Language: en-US,en;q=0.9
+Accept-Encoding: gzip, deflate, br
+```
+
+**Succeeds for:** sites without Cloudflare, sites that serve HTML to plain HTTP.
+
+**Fails for:** sites behind Cloudflare JS challenge (returns `Checking your browser...` interstitial or HTTP 403/429). The detector checks for the `cf-mitigated` response header and the challenge page title/body.
+
+**Skip with:** `skipHttp: true` in the request body, or `maxTier: 1` to cap at Tier 1.
+
+## Tier 2 — Cached Browser Session
+
+Reads `session:{hostname}` from Redis. If found, injects the saved cookies into a pooled Firefox context and navigates. Because the Cloudflare `cf_clearance` cookie is present, the challenge page is skipped and the site loads directly.
+
+**Succeeds for:** any domain that was previously solved by Tier 3 and whose session hasn't expired.
+
+**Fails for:** expired sessions (Cloudflare `cf_clearance` has a finite lifetime). On failure, TRAWL invalidates the cached session and escalates to Tier 3.
+
+## Tier 3 — Fresh Cloudflare Challenge Solve
+
+Acquires a browser from the pool (or waits up to 5s for one to become available). Navigates to the URL with no pre-loaded cookies. Waits for the Cloudflare challenge to resolve by polling `page.content()` every 500ms until the interstitial HTML is gone or `maxTimeout` elapses.
+
+On success:
+- Extracts all cookies from the page context
+- Writes `session:{hostname} → { cookies, userAgent, savedAt }` to Redis (TTL = `SESSION_TTL_SECONDS`)
+- Returns the HTML and cookies to the caller
+
+Uses [Camoufox](https://github.com/daijro/camoufox) — Firefox with fingerprint patching at the C++/Juggler level. CF's detection scripts cannot distinguish it from a real Firefox profile.
+
+## Tier 4 — Residential Proxy Escalation
+
+Same as Tier 3 but launches the browser with `RESIDENTIAL_PROXY_URL` set as the proxy. Only triggered when:
+
+1. `RESIDENTIAL_PROXY_URL` is configured
+2. Tier 3 failed (usually because the datacenter IP is flagged)
+
+If `RESIDENTIAL_PROXY_URL` is not set, Tier 4 is skipped entirely and the request returns an error after Tier 3 fails.
+
+## Tier selection
+
+The orchestrator (`packages/tiers/src/orchestrator.ts`) controls escalation. You can limit it via `ScrapeRequest.maxTier`:
+
+```json
+{ "url": "...", "maxTier": 2 }
+```
+
+This runs Tier 1, then Tier 2, then returns an error if both fail — never launching a fresh browser solve.
+
+## Timing reference
+
+| Tier | Typical time | Browser used |
+|------|-------------|--------------|
+| 1 | 50–150ms | No |
+| 2 | 400–700ms | Yes (warm) |
+| 3 | 4–15s | Yes (fresh solve) |
+| 4 | 15–45s | Yes (fresh solve + proxy) |