// the store we break on purpose
Drop is a working storefront we built, run, and break deliberately — so everything this site says about monitoring can be checked against an application whose source you can read.
demonstration store, not a customer. Form/01 is not for sale, no card is ever charged, and nothing about Drop is a usage figure or a customer reference.
Every monitoring product demonstrates itself with an incident it invented. The screenshots are drawn, the graph is plausible, the alert text was written by whoever wrote the marketing copy — and none of it can be checked. We have done it too: the two panels at the top ofthe demo are exactly that, and they say so.
The problem is that the interesting failures are the ones that are hard to fake convincingly. A screenshot of a red dashboard proves nothing about whether a tool would have noticed anything. So instead of drawing one, we built a real store — Form/01 by Soft Theory, a limited run of 100 at $29 — connected it to Beaam the way a customer would, and broke it.
That constrains us usefully. Drop has to actually work, which means the failures have to be real failures, which means what Beaam does or does not catch is a fact rather than a claim. When Beaam misses something, we find out.
Drop illustrates. It never proves.Our proof page exists to refuse invented logos, testimonials and vanity counts, and that refusal is worth more than any demonstration. So Drop appears in the guides, the demo and the comparisons as an example — and never in a customer count, a revenue figure, or a claim about adoption. It is our own application. It says nothing about whether anyone else uses Beaam.
Roughly the stack a solo founder actually ships on, chosen so the failures are the ones you would really meet rather than the ones easiest to trigger.
| Layer | Choice | What it realistically fails as |
|---|---|---|
| App | Next.js on Cloudflare Workers | The storefront is the one part that fails visibly. It is also the least interesting: a page that does not load gets noticed without help. |
| Database | MongoDB Atlas | Everything commercial lives here — release, inventory, orders, entitlements, events. Degrades rather than stops: connection exhaustion and a slow primary, neither of which changes a status code. |
| Payments | Stripe, test mode always | The charge succeeds and the webhook confirming it does not. Stripe is right, the store is wrong, and only the gap between them shows it. |
| Auth | Supabase | Identity only — it holds the creator account and no orders. Its outage locks the creator out of the dashboard without losing a sale, which is why knowing what a dependency carries matters before setting thresholds. |
| Files | Cloudflare R2 | A private bucket reached only through a hashed entitlement token. A missing object returns an error rather than an empty file, because a zero-byte download that the browser saves is worse than no download at all. |
| Resend | Delivery retries four times with backoff and then stops, visibly. "Failed" and "failed and nothing is still trying" need different actions, so they are never the same word. | |
| Telemetry | OpenTelemetry → Beaam | Drop pushes its own metrics rather than only being polled, so the demonstration covers the path a real application uses to report on itself. |
The webhook signing secret was rotated and never updated in Drop. For 8 minutes checkout took payments and delivered nothing, while every page returned 200 and Stripe reported success.
The whole thing, frame by frame →
not recorded yet
There is no video here. The walkthrough is scripted and the application it describes is built, but nobody has sat down and recorded it — and a play button over a still frame would be the kind of small lie this whole exercise is an argument against.
Until then, the written walkthrough covers the same eleven beats, and the source is the version that cannot be edited for the camera.
Drop is MIT-licensed and public. That is not generosity — it is the point. A demonstration you cannot inspect is a screenshot with extra steps, and the parts worth reading are the unglamorous ones: how inventory stays correct under contention, why a failed delivery gives up visibly rather than silently, and what happens when the product file is missing.