Article read
A reader stayed with an article long enough to have read it. Commune records this when someone has had the article open for ten seconds, or has scrolled through it, whichever comes first.
Fires once per reader per article, ever. It reports the one moment the pair crosses from unread to read, so a reader who comes back a year later produces nothing. It is a first-time signal, not a visit counter, and cannot be summed into one.
data.reader names the person. The same actions are readable as engagement records at GET /newsletters/{newsletter}/events?event_type=view.
Three things this topic does not report, each of which will make a count built from it wrong.
- A reader who is not signed in produces nothing, so counting these messages counts signed-in readers and nothing else. Real readership is higher by however much logged-out traffic the newsletter gets, and Commune records nothing durable for an anonymous read, so there is no figure to correct the total by afterwards. On a newsletter with a public archive that can be most of the readership. Treat any number derived from this topic as a floor, label it as signed-in readers, and do not call it an open rate or a view count.
- Marking articles read in bulk produces nothing. A reader clearing a backlog with "mark all as read" is declaring they are not going to read those articles. Only reading reaches this topic.
- A reader who opened an article and left produces nothing. That is a different fact, it has no topic, and its absence is why this is not an open rate.
Delivered as a single HTTPS POST to the consumer's registered endpoint, with the message as the JSON request body. Respond 2xx to acknowledge. Anything else, or a timeout, is retried with backoff, so acknowledge fast and do the work afterwards.
Headers
API version the payload conforms to. Same value as api_version in the envelope.
Same value as id in the envelope. The dedupe key.
Same value as type in the envelope. The topic name.
HMAC-SHA256 over Commune-Timestamp as Unix seconds, a literal ., then the raw request body, keyed by the endpoint's signing secret. Lowercase hex, carried as v0= followed by one or more comma separated digests (more than one while a secret is being rotated). The header's timestamp is an RFC 3339 instant, so convert it to Unix seconds before signing. Verify against the raw bytes before parsing the JSON. The webhooks guide has the full procedure.
When this delivery attempt was made, as an RFC 3339 instant in UTC. It differs on every retry, and it IS covered by the signature, in Unix seconds, so a stale or altered one can be rejected. Enforce a tolerance window of a few minutes and dedupe on the event id as well, since a retry inside that window is legitimate.
1 on the first attempt, incremented on each retry. Not sent today: do not rely on it, and be idempotent regardless. The same number is readable after the fact as attempt on a DeliveryAttempt, so a consumer that needs to tell a retry from a first delivery can ask instead of being told.
Payload
Example payload
{
"id": "018f2a8b-6c4b-7d2e-9f11-6a1c3d5e7b90",
"type": "article.published",
"api_version": "2026-08-26"
}Response
The consumer accepted the delivery. Any 2xx acknowledges the message and it is not sent again.
Changes
Webhooks have no change history: a webhook is keyed by its event name rather than a path and method, so no recorded change can name it.