Test Real Apps, Not Todo-List Demos

How a Framer Motion portfolio matured ShipLocal faster than synthetic benchmarks.

If your tunnel works on a Vite hello-world, you have not tested a tunnel.

You have tested a demo.

ShipLocal got dramatically better the week we stopped validating against starters and started tunneling real apps — especially a Next.js portfolio with Framer Motion, code-split chunks, useInView, and a noisy dev runtime.

That week produced more truth than months of synthetic “it loads” checks.

What starters hide

A todo-list demo usually has:

one HTML document
a small JS bundle
little or no HMR complexity
no motion libraries waiting on hydration
no strict CSP
no client who cares if the hero animates 3 seconds late

A real app has all of the opposite.

So “works on my starter” is a weak claim for any developer tunnel.

What the portfolio actually stressed

Through ShipLocal we saw:

SSR shell fast, motion UI late → hydration / JS delivery
Continuous reload on another app → overlay injection + Fast Refresh
Missing or flaky hot reload → HMR WebSockets not proxied
Partial UI (titles lagging body text) → observer timing as a side effect

Those became the three problem classes in our field guide and the deep dives in Parts 7–10.

None of that shows up on create-next-app default page with two paragraphs.

The feedback loop that matters

Real app report
↓
DevTools hypotheses (JS / WS / View Source)
↓
Proxy or CLI fix
↓
shiplocal doctor baseline
↓
Deploy CLI + server together
↓
Re-test the same real app

That loop is how infrastructure products mature. ngrok and Cloudflare Tunnel did not become reliable by polishing demos. They survived weird production-shaped workloads.

Turn failures into diagnostics

Once we knew the failure modes, we encoded them into something users can paste:

shiplocal doctor

Health, auth, WebSocket RTT, local vs tunnel HTML/JS, HMR probe.

If someone says “ShipLocal is slow,” doctor turns vibes into a report. That only works because the report matches bugs we saw on real stacks.

More: Building shiplocal doctor.

What we ask beta users for

Bring the weird app:

Next.js + Framer Motion
Vite + multiple entry points
full-stack projects with an API target
CSP-locked marketing sites
Dockerized local servers

If ShipLocal only works on clean demos, it is not ready for client work.

Product takeaway

ShipLocal’s job is not “forward HTTP.”

It’s “make a real local project reviewable by a non-technical client.”

That only gets proven on real projects.

Test those first. Optimize the protocol after the failures teach you what to measure.