The failure mode. A paid endpoint has two ways to be broken, and only one of them pages anyone today. When it's down, uptime monitors fire. When it's up but not charging — the 402 challenge is malformed, the price drifted, the payTo changed, the route 200s when it should challenge — everything looks green while revenue silently goes to zero. For an agent-payable API, the second failure is the common one, because the money path (challenge → settle → broadcast) has more moving parts than the serving path.
What Watch does
Every paid route on the x402 Gateway is now probed like a buyer on a schedule: unsigned request, expect a valid 402, parse the challenge, record price, payTo, and latency.
- Down — unreachable, 5xx, timeout, or a challenge that won't parse.
- Terms changed — price or payTo moved against the observed baseline. Could be an edit; could be a hijack. Either way the owner should know within minutes, not at month-end reconciliation.
- Recovered — back to a valid challenge, so the 3am page has a matching all-clear.
Free services get one watched endpoint with daily checks and dashboard status. Pro ($9/mo, $86.40/yr) gets ten endpoints — including arbitrary third-party URLs — with 15-minute checks, email + webhook alerts, and a public status page per watch.
Design notes worth stealing
Auto-watch on registration. The first paid route is monitored from minute one with zero configuration. Monitoring you have to set up is monitoring that doesn't exist when it matters.
Probe from the buyer's chair, on the seller's network. Checks run against the workers.dev origin, never same-zone entangleit.com paths — the exact trap that makes self-monitoring lie. If your architecture has a "calling yourself" path, test the path your customers actually take.
Transitions, not thresholds. Alerts fire on state changes (down/recovered/terms-changed), never on repeats. A pager that fires once per incident gets read; one that fires every 15 minutes gets filtered.
The fee pays the fee. Escrow-style arithmetic shows up everywhere in this stack; here it's simpler — history is pruned (7/90 days by plan) so the checks table can't grow unbounded, and cron ticks are capped per run.
Try it
Register any paid route and it's watched. Break it on purpose — pause the upstream, change the price — and watch the alert land. The docs have the three curl calls; the status page is yours to share as proof of reliability.
