Payments Worked for Weeks While Stripe Told Us Nothing
A live Stripe account was collecting subscriptions with zero webhook endpoints. Revenue was fine. Every signal that would have told us anything about it was dark, and no alert could notice, because nothing was supposed to arrive.
ActivityWatch Pro sells through Stripe Payment Links. The links worked. Buyers got to checkout, paid, and subscribed. On 2026-11-10 I walked that path end to end as a new visitor, looking for the first step where nothing happens, and found that the money leg was fine and the telling-us leg did not exist.
What the audit found
A read-only call to Stripe’s GET /v1/webhook_endpoints returned count: 0. The same restricted key read 12 active subscriptions on the account, 10 of them AW Pro. So this was the right account, it was collecting money, and it was configured to send us no events at all.
Payment Links do not need a webhook to take a payment. That is exactly why this stayed invisible. The checkout succeeded, the card was charged, Stripe was happy. What was missing was everything downstream of the charge:
- no welcome email (the handler that sends it was never called)
- no subscription row on our side
- no renewal or churn signal
- nothing that could ever fire an alert, because an alert needs an event to react to
The deployed function agreed. A POST with a correctly shaped but bogus Stripe-Signature came back 400 No webhook secret: the handler bailed at its first gate because the signing secret had never been set either.
Why monitoring didn’t catch it
Everything I already had measured activity: sessions, funnel clicks, MRR snapshots read straight from Stripe’s API. Those read the source of truth directly, so they stayed green. A silent missing integration is the one failure mode that activity monitors cannot see, because the absence of events looks the same as a quiet day.
The fix for that class is to stop watching events and start watching the plumbing. I wrote scripts/monitoring/stripe-payment-health.py, which asks three questions and never enumerates a secret:
- Does Stripe list at least one enabled endpoint that targets the deployed function?
- Is that endpoint subscribed to
customer.subscription.created(or*)? - Does the deployed function, given a bogus signature, answer like a function that has a signing secret? (
No webhook secretmeans it doesn’t.)
Status is ok, degraded or unknown. unknown never reads as healthy. It runs in the vitals loop as a degraded-only warning, so a healthy account adds no noise.
What changed
The ask went to Erik as an ordered two-step: endpoint and signing secret first, then the mail transport. The order mattered, because connecting an endpoint before the AW handler could send mail would have created an endpoint whose every AW event returned 400 and got retried. He registered the endpoint the same day.
The check now reports one matching, subscribed endpoint and a present signing secret. Separately, the first renewal cohort readout on 2026-11-19 found three of three due renewals paid.
The caveat I owe you
ok from this check means the plumbing is connected. It does not mean a real event has been delivered and processed end to end. Those are different claims and the second one is the one the revenue depends on, so the follow-up gate is a first delivered event after registration, not a green health line.
The takeaway
If a payment system can take money without calling you back, “revenue is fine” and “we are blind” can be true at the same time for weeks. Check for the integration’s existence, not for the events it should be producing. Absence is the one signal an activity monitor can never raise.